War on drugs – Wikipedia

War on Drugs is an American term[6][7] usually applied to the U.S. federal government’s campaign of prohibition of drugs, military aid, and military intervention, with the stated aim being to reduce the illegal drug trade.[8][9] The initiative includes a set of drug policies that are intended to discourage the production, distribution, and consumption of psychoactive drugs that the participating governments and the UN have made illegal. The term was popularized by the media shortly after a press conference given on June 18, 1971, by President Richard Nixonthe day after publication of a special message from President Nixon to the Congress on Drug Abuse Prevention and Controlduring which he declared drug abuse “public enemy number one”. That message to the Congress included text about devoting more federal resources to the “prevention of new addicts, and the rehabilitation of those who are addicted”, but that part did not receive the same public attention as the term “war on drugs”.[10][11][12] However, two years prior to this, Nixon had formally declared a “war on drugs” that would be directed toward eradication, interdiction, and incarceration.[13] Today, the Drug Policy Alliance, which advocates for an end to the War on Drugs, estimates that the United States spends $51 billion annually on these initiatives.[14]

On May 13, 2009, Gil Kerlikowskethe Director of the Office of National Drug Control Policy (ONDCP)signaled that the Obama administration did not plan to significantly alter drug enforcement policy, but also that the administration would not use the term “War on Drugs”, because Kerlikowske considers the term to be “counter-productive”.[15] ONDCP’s view is that “drug addiction is a disease that can be successfully prevented and treated… making drugs more available will make it harder to keep our communities healthy and safe”.[16] One of the alternatives that Kerlikowske has showcased is the drug policy of Sweden, which seeks to balance public health concerns with opposition to drug legalization. The prevalence rates for cocaine use in Sweden are barely one-fifth of those in Spain, the biggest consumer of the drug.[17]

In June 2011, the Global Commission on Drug Policy released a critical report on the War on Drugs, declaring: “The global war on drugs has failed, with devastating consequences for individuals and societies around the world. Fifty years after the initiation of the UN Single Convention on Narcotic Drugs, and years after President Nixon launched the US government’s war on drugs, fundamental reforms in national and global drug control policies are urgently needed.”[18] The report was criticized by organizations that oppose a general legalization of drugs.[16]

The first U.S. law that restricted the distribution and use of certain drugs was the Harrison Narcotics Tax Act of 1914. The first local laws came as early as 1860.[19] In 1919, the United States passed the 18th Amendment, prohibiting the sale, manufacture, and transportation of alcohol, with exceptions for religious and medical use. In 1920, the United States passed the National Prohibition Act (Volstead Act), enacted to carry out the provisions in law of the 18th Amendment.

The Federal Bureau of Narcotics was established in the United States Department of the Treasury by an act of June 14, 1930 (46 Stat. 585).[20] In 1933, the federal prohibition for alcohol was repealed by passage of the 21st Amendment. In 1935, President Franklin D. Roosevelt publicly supported the adoption of the Uniform State Narcotic Drug Act. The New York Times used the headline “Roosevelt Asks Narcotic War Aid”.[21][22]

In 1937, the Marihuana Tax Act of 1937 was passed. Several scholars have claimed that the goal was to destroy the hemp industry,[23][24][25] largely as an effort of businessmen Andrew Mellon, Randolph Hearst, and the Du Pont family.[23][25] These scholars argue that with the invention of the decorticator, hemp became a very cheap substitute for the paper pulp that was used in the newspaper industry.[23][26] These scholars believe that Hearst felt[dubious discuss] that this was a threat to his extensive timber holdings. Mellon, United States Secretary of the Treasury and the wealthiest man in America, had invested heavily in the DuPont’s new synthetic fiber, nylon, and considered[dubious discuss] its success to depend on its replacement of the traditional resource, hemp.[23][27][28][29][30][31][32][33] However, there were circumstances that contradict these claims. One reason for doubts about those claims is that the new decorticators did not perform fully satisfactorily in commercial production.[34] To produce fiber from hemp was a labor-intensive process if you include harvest, transport and processing. Technological developments decreased the labor with hemp but not sufficient to eliminate this disadvantage.[35][36]

On October 27, 1970, Congress passes the Comprehensive Drug Abuse Prevention and Control Act of 1970, which, among other things, categorizes controlled substances based on their medicinal use and potential for addiction.[37] In 1971, two congressmen released an explosive report on the growing heroin epidemic among U.S. servicemen in Vietnam; ten to fifteen percent of the servicemen were addicted to heroin, and President Nixon declared drug abuse to be “public enemy number one”.[37][38]

Although Nixon declared “drug abuse” to be public enemy number one in 1971,[39] the policies that his administration implemented as part of the Comprehensive Drug Abuse Prevention and Control Act of 1970 were a continuation of drug prohibition policies in the U.S., which started in 1914.[37][40]

“The Nixon campaign in 1968, and the Nixon White House after that, had two enemies: the antiwar left and black people. You understand what I’m saying? We knew we couldn’t make it illegal to be either against the war or black, but by getting the public to associate the hippies with marijuana and blacks with heroin, and then criminalizing both heavily, we could disrupt those communities. We could arrest their leaders, raid their homes, break up their meetings, and vilify them night after night on the evening news. Did we know we were lying about the drugs? Of course we did.” John Ehrlichman, to Dan Baum[41][42][43] for Harper’s Magazine[44] in 1994, about President Richard Nixon’s war on drugs, declared in 1971.[45][46]

In 1973, the Drug Enforcement Administration was created to replace the Bureau of Narcotics and Dangerous Drugs.[37]

The Nixon Administration also repealed the federal 210-year mandatory minimum sentences for possession of marijuana and started federal demand reduction programs and drug-treatment programs. Robert DuPont, the “Drug czar” in the Nixon Administration, stated it would be more accurate to say that Nixon ended, rather than launched, the “war on drugs”. DuPont also argued that it was the proponents of drug legalization that popularized the term “war on drugs”.[16][unreliable source?]

In 1982, Vice President George H. W. Bush and his aides began pushing for the involvement of the CIA and U.S. military in drug interdiction efforts.[47]

The Office of National Drug Control Policy (ONDCP) was originally established by the National Narcotics Leadership Act of 1988,[48][49] which mandated a national anti-drug media campaign for youth, which would later become the National Youth Anti-Drug Media Campaign.[50] The director of ONDCP is commonly known as the Drug czar,[37] and it was first implemented in 1989 under President George H. W. Bush,[51] and raised to cabinet-level status by Bill Clinton in 1993.[52] These activities were subsequently funded by the Treasury and General Government Appropriations Act of 1998.[53][54] The Drug-Free Media Campaign Act of 1998 codified the campaign at 21 U.S.C.1708.[55]

The Global Commission on Drug Policy released a report on June 2, 2011, alleging that “The War On Drugs Has Failed.” The commissioned was made up of 22 self-appointed members including a number of prominent international politicians and writers. U.S. Surgeon General Regina Benjamin also released the first ever National Prevention Strategy.[56]

On May 21, 2012, the U.S. Government published an updated version of its Drug Policy.[57] The director of ONDCP stated simultaneously that this policy is something different from the “War on Drugs”:

At the same meeting was a declaration signed by the representatives of Italy, the Russian Federation, Sweden, the United Kingdom and the United States in line with this: “Our approach must be a balanced one, combining effective enforcement to restrict the supply of drugs, with efforts to reduce demand and build recovery; supporting people to live a life free of addiction.”[59]

In March 2016 the International Narcotics Control Board stated that the International Drug Control treaties do not mandate a “war on drugs.”[60]

According to Human Rights Watch, the War on Drugs caused soaring arrest rates that disproportionately targeted African Americans due to various factors.[62] John Ehrlichman, an aide to Nixon, said that Nixon used the war on drugs to criminalize and disrupt black and hippie communities and their leaders.[63]

The present state of incarceration in the U.S. as a result of the war on drugs arrived in several stages. By 1971, different stops on drugs had been implemented for more than 50 years (for e.g. since 1914, 1937 etc.) with only a very small increase of inmates per 100,000 citizens. During the first 9 years after Nixon coined the expression “War on Drugs”, statistics showed only a minor increase in the total number of imprisoned.

After 1980, the situation began to change. In the 1980s, while the number of arrests for all crimes had risen by 28%, the number of arrests for drug offenses rose 126%.[64] The result of increased demand was the development of privatization and the for-profit prison industry.[65] The US Department of Justice, reporting on the effects of state initiatives, has stated that, from 1990 through 2000, “the increasing number of drug offenses accounted for 27% of the total growth among black inmates, 7% of the total growth among Hispanic inmates, and 15% of the growth among white inmates.” In addition to prison or jail, the United States provides for the deportation of many non-citizens convicted of drug offenses.[66]

In 1994, the New England Journal of Medicine reported that the “War on Drugs” resulted in the incarceration of one million Americans each year.[67] In 2008, the Washington Post reported that of 1.5 million Americans arrested each year for drug offenses, half a million would be incarcerated.[68] In addition, one in five black Americans would spend time behind bars due to drug laws.[68]

Federal and state policies also impose collateral consequences on those convicted of drug offenses, such as denial of public benefits or licenses, that are not applicable to those convicted of other types of crime.[69] In particular, the passage of the 1990 SolomonLautenberg amendment led many states to impose mandatory driver’s license suspensions (of at least 6 months) for persons committing a drug offense, regardless of whether any motor vehicle was involved.[70][71] Approximately 191,000 licenses were suspended in this manner in 2016, according to a Prison Policy Initiative report.[72]

In 1986, the U.S. Congress passed laws that created a 100 to 1 sentencing disparity for the trafficking or possession of crack when compared to penalties for trafficking of powder cocaine,[73][74][75][76] which had been widely criticized as discriminatory against minorities, mostly blacks, who were more likely to use crack than powder cocaine.[77] This 100:1 ratio had been required under federal law since 1986.[78] Persons convicted in federal court of possession of 5grams of crack cocaine received a minimum mandatory sentence of 5 years in federal prison. On the other hand, possession of 500grams of powder cocaine carries the same sentence.[74][75] In 2010, the Fair Sentencing Act cut the sentencing disparity to 18:1.[77]

According to Human Rights Watch, crime statistics show thatin the United States in 1999compared to non-minorities, African Americans were far more likely to be arrested for drug crimes, and received much stiffer penalties and sentences.[79]

Statistics from 1998 show that there were wide racial disparities in arrests, prosecutions, sentencing and deaths. African-American drug users made up for 35% of drug arrests, 55% of convictions, and 74% of people sent to prison for drug possession crimes.[74] Nationwide African-Americans were sent to state prisons for drug offenses 13 times more often than other races,[80] even though they only supposedly comprised 13% of regular drug users.[74]

Anti-drug legislation over time has also displayed an apparent racial bias. University of Minnesota Professor and social justice author Michael Tonry writes, “The War on Drugs foreseeably and unnecessarily blighted the lives of hundreds and thousands of young disadvantaged black Americans and undermined decades of effort to improve the life chances of members of the urban black underclass.”[81]

In 1968, President Lyndon B. Johnson decided that the government needed to make an effort to curtail the social unrest that blanketed the country at the time. He decided to focus his efforts on illegal drug use, an approach which was in line with expert opinion on the subject at the time. In the 1960s, it was believed that at least half of the crime in the U.S. was drug related, and this number grew as high as 90 percent in the next decade.[82] He created the Reorganization Plan of 1968 which merged the Bureau of Narcotics and the Bureau of Drug Abuse to form the Bureau of Narcotics and Dangerous Drugs within the Department of Justice.[83] The belief during this time about drug use was summarized by journalist Max Lerner in his celebrated[citation needed] work America as a Civilization (1957):

As a case in point we may take the known fact of the prevalence of reefer and dope addiction in Negro areas. This is essentially explained in terms of poverty, slum living, and broken families, yet it would be easy to show the lack of drug addiction among other ethnic groups where the same conditions apply.[84]

Richard Nixon became president in 1969, and did not back away from the anti-drug precedent set by Johnson. Nixon began orchestrating drug raids nationwide to improve his “watchdog” reputation. Lois B. Defleur, a social historian who studied drug arrests during this period in Chicago, stated that, “police administrators indicated they were making the kind of arrests the public wanted”. Additionally, some of Nixon’s newly created drug enforcement agencies would resort to illegal practices to make arrests as they tried to meet public demand for arrest numbers. From 1972 to 1973, the Office of Drug Abuse and Law Enforcement performed 6,000 drug arrests in 18 months, the majority of the arrested black.[85]

The next two Presidents, Gerald Ford and Jimmy Carter, responded with programs that were essentially a continuation of their predecessors. Shortly after Ronald Reagan became President in 1981 he delivered a speech on the topic. Reagan announced, “We’re taking down the surrender flag that has flown over so many drug efforts; we’re running up a battle flag.”[86] For his first five years in office, Reagan slowly strengthened drug enforcement by creating mandatory minimum sentencing and forfeiture of cash and real estate for drug offenses, policies far more detrimental to poor blacks than any other sector affected by the new laws.[citation needed]

Then, driven by the 1986 cocaine overdose of black basketball star Len Bias,[dubious discuss] Reagan was able to pass the Anti-Drug Abuse Act through Congress. This legislation appropriated an additional $1.7 billion to fund the War on Drugs. More importantly, it established 29 new, mandatory minimum sentences for drug offenses. In the entire history of the country up until that point, the legal system had only seen 55 minimum sentences in total.[87] A major stipulation of the new sentencing rules included different mandatory minimums for powder and crack cocaine. At the time of the bill, there was public debate as to the difference in potency and effect of powder cocaine, generally used by whites, and crack cocaine, generally used by blacks, with many believing that “crack” was substantially more powerful and addictive. Crack and powder cocaine are closely related chemicals, crack being a smokeable, freebase form of powdered cocaine hydrochloride which produces a shorter, more intense high while using less of the drug. This method is more cost effective, and therefore more prevalent on the inner-city streets, while powder cocaine remains more popular in white suburbia. The Reagan administration began shoring public opinion against “crack”, encouraging DEA official Robert Putnam to play up the harmful effects of the drug. Stories of “crack whores” and “crack babies” became commonplace; by 1986, Time had declared “crack” the issue of the year.[88] Riding the wave of public fervor, Reagan established much harsher sentencing for crack cocaine, handing down stiffer felony penalties for much smaller amounts of the drug.[89]

Reagan protg and former Vice-President George H. W. Bush was next to occupy the oval office, and the drug policy under his watch held true to his political background. Bush maintained the hard line drawn by his predecessor and former boss, increasing narcotics regulation when the First National Drug Control Strategy was issued by the Office of National Drug Control in 1989.[90]

The next three presidents Clinton, Bush and Obama continued this trend, maintaining the War on Drugs as they inherited it upon taking office.[91] During this time of passivity by the federal government, it was the states that initiated controversial legislation in the War on Drugs. Racial bias manifested itself in the states through such controversial policies as the “stop and frisk” police practices in New York city and the “three strikes” felony laws began in California in 1994.[92]

In August 2010, President Obama signed the Fair Sentencing Act into law that dramatically reduced the 100-to-1 sentencing disparity between powder and crack cocaine, which disproportionately affected minorities.[93]

Commonly used illegal drugs include heroin, cocaine, methamphetamine, and, marijuana.

Heroin is an opiate that is highly addictive. If caught selling or possessing heroin, a perpetrator can be charged with a felony and face twofour years in prison and could be fined to a maximum of $20,000.[94]

Crystal meth is composed of methamphetamine hydrochloride. It is marketed as either a white powder or in a solid (rock) form. The possession of crystal meth can result in a punishment varying from a fine to a jail sentence. As with other drug crimes, sentencing length may increase depending on the amount of the drug found in the possession of the defendant.[95]

Cocaine possession is illegal across the U.S., with the cheaper crack cocaine incurring even greater penalties. Having possession is when the accused knowingly has it on their person, or in a backpack or purse. The possession of cocaine with no prior conviction, for the first offense, the person will be sentenced to a maximum of one year in prison or fined $1,000, or both. If the person has a prior conviction, whether it is a narcotic or cocaine, they will be sentenced to two years in prison, a $2,500 fine, or both. With two or more convictions of possession prior to this present offense, they can be sentenced to 90 days in prison along with a $5,000 fine.[96]

Marijuana is the most popular illegal drug worldwide. The punishment for possession of it is less than for the possession of cocaine or heroin. In some U.S. states, the drug is legal. Over 80 million Americans have tried marijuana. The Criminal Defense Lawyer article claims that, depending on the age of person and how much the person has been caught for possession, they will be fined and could plea bargain into going to a treatment program versus going to prison. In each state the convictions differ along with how much marijuana they have on their person.[97]

Some scholars have claimed that the phrase “War on Drugs” is propaganda cloaking an extension of earlier military or paramilitary operations.[9] Others have argued that large amounts of “drug war” foreign aid money, training, and equipment actually goes to fighting leftist insurgencies and is often provided to groups who themselves are involved in large-scale narco-trafficking, such as corrupt members of the Colombian military.[8]

From 1963 to the end of the Vietnam War in 1975, marijuana usage became common among U.S. soldiers in non-combat situations. Some servicemen also used heroin. Many of the servicemen ended the heroin use after returning to the United States but came home addicted. In 1971, the U.S. military conducted a study of drug use among American servicemen and women. It found that daily usage rates for drugs on a worldwide basis were as low as two percent.[98] However, in the spring of 1971, two congressmen released an alarming report alleging that 15% of the servicemen in Vietnam were addicted to heroin. Marijuana use was also common in Vietnam. Soldiers who used drugs had more disciplinary problems. The frequent drug use had become an issue for the commanders in Vietnam; in 1971 it was estimated that 30,000 servicemen were addicted to drugs, most of them to heroin.[11]

From 1971 on, therefore, returning servicemen were required to take a mandatory heroin test. Servicemen who tested positive upon returning from Vietnam were not allowed to return home until they had passed the test with a negative result. The program also offered a treatment for heroin addicts.[99]

Elliot Borin’s article “The U.S. Military Needs its Speed”published in Wired on February 10, 2003reports:

But the Defense Department, which distributed millions of amphetamine tablets to troops during World War II, Vietnam and the Gulf War, soldiers on, insisting that they are not only harmless but beneficial.

In a news conference held in connection with Schmidt and Umbach’s Article 32 hearing, Dr. Pete Demitry, an Air Force physician and a pilot, claimed that the “Air Force has used (Dexedrine) safely for 60 years” with “no known speed-related mishaps.”

The need for speed, Demitry added “is a life-and-death issue for our military.”[100]

One of the first anti-drug efforts in the realm of foreign policy was President Nixon’s Operation Intercept, announced in September 1969, targeted at reducing the amount of cannabis entering the United States from Mexico. The effort began with an intense inspection crackdown that resulted in an almost shutdown of cross-border traffic.[101] Because the burden on border crossings was controversial in border states, the effort only lasted twenty days.[102]

On December 20, 1989, the United States invaded Panama as part of Operation Just Cause, which involved 25,000 American troops. Gen. Manuel Noriega, head of the government of Panama, had been giving military assistance to Contra groups in Nicaragua at the request of the U.S. which, in exchange, tolerated his drug trafficking activities, which they had known about since the 1960s.[103][104] When the Drug Enforcement Administration (DEA) tried to indict Noriega in 1971, the CIA prevented them from doing so.[103] The CIA, which was then directed by future president George H. W. Bush, provided Noriega with hundreds of thousands of dollars per year as payment for his work in Latin America.[103] When CIA pilot Eugene Hasenfus was shot down over Nicaragua by the Sandinistas, documents aboard the plane revealed many of the CIA’s activities in Latin America, and the CIA’s connections with Noriega became a public relations “liability” for the U.S. government, which finally allowed the DEA to indict him for drug trafficking, after decades of tolerating his drug operations.[103] Operation Just Cause, whose purpose was to capture Noriega and overthrow his government; Noriega found temporary asylum in the Papal Nuncio, and surrendered to U.S. soldiers on January 3, 1990.[105] He was sentenced by a court in Miami to 45 years in prison.[103]

As part of its Plan Colombia program, the United States government currently provides hundreds of millions of dollars per year of military aid, training, and equipment to Colombia,[106] to fight left-wing guerrillas such as the Revolutionary Armed Forces of Colombia (FARC-EP), which has been accused of being involved in drug trafficking.[107]

Private U.S. corporations have signed contracts to carry out anti-drug activities as part of Plan Colombia. DynCorp, the largest private company involved, was among those contracted by the State Department, while others signed contracts with the Defense Department.[108]

Colombian military personnel have received extensive counterinsurgency training from U.S. military and law enforcement agencies, including the School of Americas (SOA). Author Grace Livingstone has stated that more Colombian SOA graduates have been implicated in human rights abuses than currently known SOA graduates from any other country. All of the commanders of the brigades highlighted in a 2001 Human Rights Watch report on Colombia were graduates of the SOA, including the III brigade in Valle del Cauca, where the 2001 Alto Naya Massacre occurred. US-trained officers have been accused of being directly or indirectly involved in many atrocities during the 1990s, including the Massacre of Trujillo and the 1997 Mapiripn Massacre.

In 2000, the Clinton administration initially waived all but one of the human rights conditions attached to Plan Colombia, considering such aid as crucial to national security at the time.[109]

The efforts of U.S. and Colombian governments have been criticized for focusing on fighting leftist guerrillas in southern regions without applying enough pressure on right-wing paramilitaries and continuing drug smuggling operations in the north of the country.[110][111] Human Rights Watch, congressional committees and other entities have documented the existence of connections between members of the Colombian military and the AUC, which the U.S. government has listed as a terrorist group, and that Colombian military personnel have committed human rights abuses which would make them ineligible for U.S. aid under current laws.[citation needed]

In 2010, the Washington Office on Latin America concluded that both Plan Colombia and the Colombian government’s security strategy “came at a high cost in lives and resources, only did part of the job, are yielding diminishing returns and have left important institutions weaker.”[112]

A 2014 report by the RAND Corporation, which was issued to analyze viable strategies for the Mexican drug war considering successes experienced in Columbia, noted:

Between 1999 and 2002, the United States gave Colombia $2.04 billion in aid, 81 percent of which was for military purposes, placing Colombia just below Israel and Egypt among the largest recipients of U.S. military assistance. Colombia increased its defense spending from 3.2 percent of gross domestic product (GDP) in 2000 to 4.19 percent in 2005. Overall, the results were extremely positive. Greater spending on infrastructure and social programs helped the Colombian government increase its political legitimacy, while improved security forces were better able to consolidate control over large swaths of the country previously overrun by insurgents and drug cartels.

It also notes that, “Plan Colombia has been widely hailed as a success, and some analysts believe that, by 2010, Colombian security forces had finally gained the upper hand once and for all.”[113]

The Mrida Initiative is a security cooperation between the United States and the government of Mexico and the countries of Central America. It was approved on June 30, 2008, and its stated aim is combating the threats of drug trafficking and transnational crime. The Mrida Initiative appropriated $1.4 billion in a three-year commitment (20082010) to the Mexican government for military and law enforcement training and equipment, as well as technical advice and training to strengthen the national justice systems. The Mrida Initiative targeted many very important government officials, but it failed to address the thousands of Central Americans who had to flee their countries due to the danger they faced everyday because of the war on drugs. There is still not any type of plan that addresses these people. No weapons are included in the plan.[114][115]

The United States regularly sponsors the spraying of large amounts of herbicides such as glyphosate over the jungles of Central and South America as part of its drug eradication programs. Environmental consequences resulting from aerial fumigation have been criticized as detrimental to some of the world’s most fragile ecosystems;[116] the same aerial fumigation practices are further credited with causing health problems in local populations.[117]

In 2012, the U.S. sent DEA agents to Honduras to assist security forces in counternarcotics operations. Honduras has been a major stop for drug traffickers, who use small planes and landing strips hidden throughout the country to transport drugs. The U.S. government made agreements with several Latin American countries to share intelligence and resources to counter the drug trade. DEA agents, working with other U.S. agencies such as the State Department, the CBP, and Joint Task Force-Bravo, assisted Honduras troops in conducting raids on traffickers’ sites of operation.[118]

The War on Drugs has been a highly contentious issue since its inception. A poll on October 2, 2008, found that three in four Americans believed that the War On Drugs was failing.[119]

At a meeting in Guatemala in 2012, three former presidents from Guatemala, Mexico and Colombia said that the war on drugs had failed and that they would propose a discussion on alternatives, including decriminalization, at the Summit of the Americas in April of that year.[120] Guatemalan President Otto Prez Molina said that the war on drugs was exacting too high a price on the lives of Central Americans and that it was time to “end the taboo on discussing decriminalization”.[121] At the summit, the government of Colombia pushed for the most far-reaching change to drugs policy since the war on narcotics was declared by Nixon four decades prior, citing the catastrophic effects it had had in Colombia.[122]

Several critics have compared the wholesale incarceration of the dissenting minority of drug users to the wholesale incarceration of other minorities in history. Psychiatrist Thomas Szasz, for example, writes in 1997 “Over the past thirty years, we have replaced the medical-political persecution of illegal sex users (‘perverts’ and ‘psychopaths’) with the even more ferocious medical-political persecution of illegal drug users.”[123]

Penalties for drug crimes among American youth almost always involve permanent or semi-permanent removal from opportunities for education, strip them of voting rights, and later involve creation of criminal records which make employment more difficult.[124] Thus, some authors maintain that the War on Drugs has resulted in the creation of a permanent underclass of people who have few educational or job opportunities, often as a result of being punished for drug offenses which in turn have resulted from attempts to earn a living in spite of having no education or job opportunities.[124]

According to a 2008 study published by Harvard economist Jeffrey A. Miron, the annual savings on enforcement and incarceration costs from the legalization of drugs would amount to roughly $41.3 billion, with $25.7 billion being saved among the states and over $15.6 billion accrued for the federal government. Miron further estimated at least $46.7 billion in tax revenue based on rates comparable to those on tobacco and alcohol ($8.7 billion from marijuana, $32.6 billion from cocaine and heroin, remainder from other drugs).[125]

Low taxation in Central American countries has been credited with weakening the region’s response in dealing with drug traffickers. Many cartels, especially Los Zetas have taken advantage of the limited resources of these nations. 2010 tax revenue in El Salvador, Guatemala, and Honduras, composed just 13.53% of GDP. As a comparison, in Chile and the U.S., taxes were 18.6% and 26.9% of GDP respectively. However, direct taxes on income are very hard to enforce and in some cases tax evasion is seen as a national pastime.[126]

The status of coca and coca growers has become an intense political issue in several countries, including Colombia and particularly Bolivia, where the president, Evo Morales, a former coca growers’ union leader, has promised to legalise the traditional cultivation and use of coca.[127] Indeed, legalization efforts have yielded some successes under the Morales administration when combined with aggressive and targeted eradication efforts. The country saw a 1213% decline in coca cultivation[127] in 2011 under Morales, who has used coca growers’ federations to ensure compliance with the law rather than providing a primary role for security forces.[127]

The coca eradication policy has been criticised for its negative impact on the livelihood of coca growers in South America. In many areas of South America the coca leaf has traditionally been chewed and used in tea and for religious, medicinal and nutritional purposes by locals.[128] For this reason many insist that the illegality of traditional coca cultivation is unjust. In many areas the U.S. government and military has forced the eradication of coca without providing for any meaningful alternative crop for farmers, and has additionally destroyed many of their food or market crops, leaving them starving and destitute.[128]

The CIA, DEA, State Department, and several other U.S. government agencies have been alleged to have relations with various groups which are involved in drug trafficking.

Senator John Kerry’s 1988 U.S. Senate Committee on Foreign Relations report on Contra drug links concludes that members of the U.S. State Department “who provided support for the Contras are involved in drug trafficking… and elements of the Contras themselves knowingly receive financial and material assistance from drug traffickers.”[129] The report further states that “the Contra drug links include… payments to drug traffickers by the U.S. State Department of funds authorized by the Congress for humanitarian assistance to the Contras, in some cases after the traffickers had been indicted by federal law enforcement agencies on drug charges, in others while traffickers were under active investigation by these same agencies.”

In 1996, journalist Gary Webb published reports in the San Jose Mercury News, and later in his book Dark Alliance, detailing how Contras, had been involved in distributing crack cocaine into Los Angeles whilst receiving money from the CIA.[citation needed] Contras used money from drug trafficking to buy weapons.[citation needed]

Webb’s premise regarding the U.S. Government connection was initially attacked at the time by the media. It is now widely accepted that Webb’s main assertion of government “knowledge of drug operations, and collaboration with and protection of known drug traffickers” was correct.[130][not in citation given] In 1998, CIA Inspector General Frederick Hitz published a two-volume report[131] that while seemingly refuting Webb’s claims of knowledge and collaboration in its conclusions did not deny them in its body.[citation needed] Hitz went on to admit CIA improprieties in the affair in testimony to a House congressional committee. There has been a reversal amongst mainstream media of its position on Webb’s work, with acknowledgement made of his contribution to exposing a scandal it had ignored.

According to Rodney Campbell, an editorial assistant to Nelson Rockefeller, during World War II, the United States Navy, concerned that strikes and labor disputes in U.S. eastern shipping ports would disrupt wartime logistics, released the mobster Lucky Luciano from prison, and collaborated with him to help the mafia take control of those ports. Labor union members were terrorized and murdered by mafia members as a means of preventing labor unrest and ensuring smooth shipping of supplies to Europe.[132]

According to Alexander Cockburn and Jeffrey St. Clair, in order to prevent Communist party members from being elected in Italy following World War II, the CIA worked closely with the Sicilian Mafia, protecting them and assisting in their worldwide heroin smuggling operations. The mafia was in conflict with leftist groups and was involved in assassinating, torturing, and beating leftist political organizers.[133]

In 1986, the US Defense Department funded a two-year study by the RAND Corporation, which found that the use of the armed forces to interdict drugs coming into the United States would have little or no effect on cocaine traffic and might, in fact, raise the profits of cocaine cartels and manufacturers. The 175-page study, “Sealing the Borders: The Effects of Increased Military Participation in Drug Interdiction”, was prepared by seven researchers, mathematicians and economists at the National Defense Research Institute, a branch of the RAND, and was released in 1988. The study noted that seven prior studies in the past nine years, including one by the Center for Naval Research and the Office of Technology Assessment, had come to similar conclusions. Interdiction efforts, using current armed forces resources, would have almost no effect on cocaine importation into the United States, the report concluded.[135]

During the early-to-mid-1990s, the Clinton administration ordered and funded a major cocaine policy study, again by RAND. The Rand Drug Policy Research Center study concluded that $3 billion should be switched from federal and local law enforcement to treatment. The report said that treatment is the cheapest way to cut drug use, stating that drug treatment is twenty-three times more effective than the supply-side “war on drugs”.[136]

The National Research Council Committee on Data and Research for Policy on Illegal Drugs published its findings in 2001 on the efficacy of the drug war. The NRC Committee found that existing studies on efforts to address drug usage and smuggling, from U.S. military operations to eradicate coca fields in Colombia, to domestic drug treatment centers, have all been inconclusive, if the programs have been evaluated at all: “The existing drug-use monitoring systems are strikingly inadequate to support the full range of policy decisions that the nation must make…. It is unconscionable for this country to continue to carry out a public policy of this magnitude and cost without any way of knowing whether and to what extent it is having the desired effect.”[137] The study, though not ignored by the press, was ignored by top-level policymakers, leading Committee Chair Charles Manski to conclude, as one observer notes, that “the drug war has no interest in its own results”.[138]

In mid-1995, the US government tried to reduce the supply of methamphetamine precursors to disrupt the market of this drug. According to a 2009 study, this effort was successful, but its effects were largely temporary.[139]

During alcohol prohibition, the period from 1920 to 1933, alcohol use initially fell but began to increase as early as 1922. It has been extrapolated that even if prohibition had not been repealed in 1933, alcohol consumption would have quickly surpassed pre-prohibition levels.[140] One argument against the War on Drugs is that it uses similar measures as Prohibition and is no more effective.

In the six years from 2000 to 2006, the U.S. spent $4.7 billion on Plan Colombia, an effort to eradicate coca production in Colombia. The main result of this effort was to shift coca production into more remote areas and force other forms of adaptation. The overall acreage cultivated for coca in Colombia at the end of the six years was found to be the same, after the U.S. Drug Czar’s office announced a change in measuring methodology in 2005 and included new areas in its surveys.[141] Cultivation in the neighboring countries of Peru and Bolivia increased, some would describe this effect like squeezing a balloon.[142]

Richard Davenport-Hines, in his book The Pursuit of Oblivion,[143] criticized the efficacy of the War on Drugs by pointing out that

1015% of illicit heroin and 30% of illicit cocaine is intercepted. Drug traffickers have gross profit margins of up to 300%. At least 75% of illicit drug shipments would have to be intercepted before the traffickers’ profits were hurt.

Alberto Fujimori, president of Peru from 1990 to 2000, described U.S. foreign drug policy as “failed” on grounds that “for 10 years, there has been a considerable sum invested by the Peruvian government and another sum on the part of the American government, and this has not led to a reduction in the supply of coca leaf offered for sale. Rather, in the 10 years from 1980 to 1990, it grew 10-fold.”[144]

At least 500 economists, including Nobel Laureates Milton Friedman,[145] George Akerlof and Vernon L. Smith, have noted that reducing the supply of marijuana without reducing the demand causes the price, and hence the profits of marijuana sellers, to go up, according to the laws of supply and demand.[146] The increased profits encourage the producers to produce more drugs despite the risks, providing a theoretical explanation for why attacks on drug supply have failed to have any lasting effect. The aforementioned economists published an open letter to President George W. Bush stating “We urge…the country to commence an open and honest debate about marijuana prohibition… At a minimum, this debate will force advocates of current policy to show that prohibition has benefits sufficient to justify the cost to taxpayers, foregone tax revenues and numerous ancillary consequences that result from marijuana prohibition.”

The declaration from the World Forum Against Drugs, 2008 state that a balanced policy of drug abuse prevention, education, treatment, law enforcement, research, and supply reduction provides the most effective platform to reduce drug abuse and its associated harms and call on governments to consider demand reduction as one of their first priorities in the fight against drug abuse.[147]

Despite over $7 billion spent annually towards arresting[148] and prosecuting nearly 800,000 people across the country for marijuana offenses in 2005[citation needed] (FBI Uniform Crime Reports), the federally funded Monitoring the Future Survey reports about 85% of high school seniors find marijuana “easy to obtain”. That figure has remained virtually unchanged since 1975, never dropping below 82.7% in three decades of national surveys.[149] The Drug Enforcement Administration states that the number of users of marijuana in the U.S. declined between 2000 and 2005 even with many states passing new medical marijuana laws making access easier,[150] though usage rates remain higher than they were in the 1990s according to the National Survey on Drug Use and Health.[151]

ONDCP stated in April 2011 that there has been a 46 percent drop in cocaine use among young adults over the past five years, and a 65 percent drop in the rate of people testing positive for cocaine in the workplace since 2006.[152] At the same time, a 2007 study found that up to 35% of college undergraduates used stimulants not prescribed to them.[153]

A 2013 study found that prices of heroin, cocaine and cannabis had decreased from 1990 to 2007, but the purity of these drugs had increased during the same time.[154]

The War on Drugs is often called a policy failure.[155][156][157][158][159]

The legality of the War on Drugs has been challenged on four main grounds in the U.S.

Several authors believe that the United States’ federal and state governments have chosen wrong methods for combatting the distribution of illicit substances. Aggressive, heavy-handed enforcement funnels individuals through courts and prisons; instead of treating the cause of the addiction, the focus of government efforts has been on punishment. By making drugs illegal rather than regulating them, the War on Drugs creates a highly profitable black market. Jefferson Fish has edited scholarly collections of articles offering a wide variety of public health based and rights based alternative drug policies.[160][161][162]

In the year 2000, the United States drug-control budget reached 18.4 billion dollars,[163] nearly half of which was spent financing law enforcement while only one sixth was spent on treatment. In the year 2003, 53 percent of the requested drug control budget was for enforcement, 29 percent for treatment, and 18 percent for prevention.[164] The state of New York, in particular, designated 17 percent of its budget towards substance-abuse-related spending. Of that, a mere one percent was put towards prevention, treatment, and research.

In a survey taken by Substance Abuse and Mental Health Services Administration (SAMHSA), it was found that substance abusers that remain in treatment longer are less likely to resume their former drug habits. Of the people that were studied, 66 percent were cocaine users. After experiencing long-term in-patient treatment, only 22 percent returned to the use of cocaine. Treatment had reduced the number of cocaine abusers by two-thirds.[163] By spending the majority of its money on law enforcement, the federal government had underestimated the true value of drug-treatment facilities and their benefit towards reducing the number of addicts in the U.S.

In 2004 the federal government issued the National Drug Control Strategy. It supported programs designed to expand treatment options, enhance treatment delivery, and improve treatment outcomes. For example, the Strategy provided SAMHSA with a $100.6 million grant to put towards their Access to Recovery (ATR) initiative. ATR is a program that provides vouchers to addicts to provide them with the means to acquire clinical treatment or recovery support. The project’s goals are to expand capacity, support client choice, and increase the array of faith-based and community based providers for clinical treatment and recovery support services.[165] The ATR program will also provide a more flexible array of services based on the individual’s treatment needs.

The 2004 Strategy additionally declared a significant 32 million dollar raise in the Drug Courts Program, which provides drug offenders with alternatives to incarceration. As a substitute for imprisonment, drug courts identify substance-abusing offenders and place them under strict court monitoring and community supervision, as well as provide them with long-term treatment services.[166] According to a report issued by the National Drug Court Institute, drug courts have a wide array of benefits, with only 16.4 percent of the nation’s drug court graduates rearrested and charged with a felony within one year of completing the program (versus the 44.1% of released prisoners who end up back in prison within 1-year). Additionally, enrolling an addict in a drug court program costs much less than incarcerating one in prison.[167] According to the Bureau of Prisons, the fee to cover the average cost of incarceration for Federal inmates in 2006 was $24,440.[168] The annual cost of receiving treatment in a drug court program ranges from $900 to $3,500. Drug courts in New York State alone saved $2.54 million in incarceration costs.[167]

Describing the failure of the War on Drugs, New York Times columnist Eduardo Porter noted:

Jeffrey Miron, an economist at Harvard who studies drug policy closely, has suggested that legalizing all illicit drugs would produce net benefits to the United States of some $65 billion a year, mostly by cutting public spending on enforcement as well as through reduced crime and corruption. A study by analysts at the RAND Corporation, a California research organization, suggested that if marijuana were legalized in California and the drug spilled from there to other states, Mexican drug cartels would lose about a fifth of their annual income of some $6.5 billion from illegal exports to the United States.[169]

Many believe that the War on Drugs has been costly and ineffective largely because inadequate emphasis is placed on treatment of addiction. The United States leads the world in both recreational drug usage and incarceration rates. 70% of men arrested in metropolitan areas test positive for an illicit substance,[170] and 54% of all men incarcerated will be repeat offenders.[171]

There are also programs in the United States to combat public health risks of injecting drug users such as the Needle exchange programme. The “needle exchange programme” is intended to provide injecting drug users with new needles in exchange for used needles to prevent needle sharing.

Covert activities and foreign policy

Visit link:

War on drugs – Wikipedia

Homepage – The War On Drugs

Postal Code

Country Select CountryUnited StatesAfghanistanAlbaniaAlgeriaAmerican SamoaAndorraAngolaAnguillaAntarcticaAntigua and BarbudaArgentinaArmeniaArubaAustraliaAustriaAzerbaijanBahamasBahrainBangladeshBarbadosBelarusBelgiumBelizeBeninBermudaBhutanBoliviaBosnia & HerzegovinaBotswanaBouvet IslandBrazilBritish Ind Ocean Ter Brunei DarussalamBulgariaBurkina FasoBurundiCambodiaCameroonCanadaCape VerdeCayman IslandsCentral African RepChadChileChinaChristmas IslandCocos (Keeling Is)ColombiaComorosCongoCook IslandsCosta RicaCote D’Ivoire Croatia (Hrvatska)CubaCyprusCzech RepublicDenmarkDjiboutiDominicaDominican RepublicEast TimorEcuadorEgyptEl SalvadorEquatorial GuineaEritreaEstoniaEthiopiaFalkland Islands Faroe IslandsFijiFinlandFranceFrance, MetroFrench GuianaFrench PolynesiaFrench Southern TerGabonGambiaGeorgiaGeorgia and S. Sand IsGermanyGhanaGibraltarGreeceGreenlandGrenadaGuadeloupeGuamGuatemalaGuineaGuinea-BissauGuyanaHaitiHeard & McDonald IsHondurasHong KongHungaryIcelandIndiaIndonesiaIranIraqIrelandIsraelItalyJamaicaJapanJordanKazakhstanKenyaKiribatiKorea (North) Korea (South)KuwaitKyrgyzstan LaosLatviaLebanonLesothoLiberiaLibyaLiechtensteinLithuaniaLuxembourgMacauMacedoniaMadagascarMalawiMalaysiaMaldivesMaliMaltaMarshall IslandsMartiniqueMauritaniaMauritiusMayotteMexicoMicronesiaMoldovaMonacoMongoliaMontserratMoroccoMozambiqueMyanmarNamibiaNauruNepalNetherlandsNetherlands AntillesNeutral Zone Saudi/IraqNew CaledoniaNew ZealandNicaraguaNigerNigeriaNiueNorfolk IslandNorthern Mariana IsNorwayOmanPakistanPalauPanamaPapua New GuineaParaguayPeruPhilippinesPitcairnPolandPortugalPuerto RicoQatarReunionRomaniaRussian FederationRwandaSaint Kitts and NevisSaint LuciaSt. Vincent/Grenadines SamoaSan MarinoSao Tome and PrincipeSaudi ArabiaSenegalSeychellesSierra LeoneSingaporeSlovakia (Slovak Rep)SloveniaSolomon IslandsSomaliaSouth AfricaSoviet Union (former)SpainSri LankaSt. HelenaSt. Pierre and Miquelo SudanSurinameSvalbard & Jan MayenSwazilandSwedenSwitzerlandSyriaTaiwanTajikistanTanzaniaThailandTogoTokelauTongaTrinidad and TobagoTunisiaTurkeyTurkmenistanTurks and Caicos TuvaluUS Minor Outlying IsUgandaUkraineUnited Arab EmiratesUnited Kingdom UruguayUzbekistanVanuatuVatican City State VenezuelaViet NamVirgin Islands (Brit)Virgin Islands (US)Wallis and Futuna IsWestern SaharaYemenYugoslaviaZaireZambiaZimbabwe

First Name

Birth Date

MonthJanuaryFebruaryMarchAprilMayJuneJulyAugustSeptemberOctoberNovemberDecember

Date12345678910111213141516171819202122232425262728293031

Read the original:

Homepage – The War On Drugs

War on Drugs | United States history | Britannica.com

War on Drugs, the effort in the United States since the 1970s to combat illegal drug use by greatly increasing penalties, enforcement, and incarceration for drug offenders.

The War on Drugs began in June 1971 when U.S. Pres. Richard Nixon declared drug abuse to be public enemy number one and increased federal funding for drug-control agencies and drug-treatment efforts. In 1973 the Drug Enforcement Agency was created out of the merger of the Office for Drug Abuse Law Enforcement, the Bureau of Narcotics and Dangerous Drugs, and the Office of Narcotics Intelligence to consolidate federal efforts to control drug abuse.

The War on Drugs was a relatively small component of federal law-enforcement efforts until the presidency of Ronald Reagan, which began in 1981. Reagan greatly expanded the reach of the drug war and his focus on criminal punishment over treatment led to a massive increase in incarcerations for nonviolent drug offenses, from 50,000 in 1980 to 400,000 in 1997. In 1984 his wife, Nancy, spearheaded another facet of the War on Drugs with her Just Say No campaign, which was a privately funded effort to educate schoolchildren on the dangers of drug use. The expansion of the War on Drugs was in many ways driven by increased media coverage ofand resulting public nervousness overthe crack epidemic that arose in the early 1980s. This heightened concern over illicit drug use helped drive political support for Reagans hard-line stance on drugs. The U.S. Congress passed the Anti-Drug Abuse Act of 1986, which allocated $1.7 billion to the War on Drugs and established a series of mandatory minimum prison sentences for various drug offenses. A notable feature of mandatory minimums was the massive gap between the amounts of crack and of powder cocaine that resulted in the same minimum sentence: possession of five grams of crack led to an automatic five-year sentence while it took the possession of 500 grams of powder cocaine to trigger that sentence. Since approximately 80% of crack users were African American, mandatory minimums led to an unequal increase of incarceration rates for nonviolent black drug offenders, as well as claims that the War on Drugs was a racist institution.

Concerns over the effectiveness of the War on Drugs and increased awareness of the racial disparity of the punishments meted out by it led to decreased public support of the most draconian aspects of the drug war during the early 21st century. Consequently, reforms were enacted during that time, such as the legalization of recreational marijuana in a number of states and the passage of the Fair Sentencing Act of 2010 that reduced the discrepancy of crack-to-powder possession thresholds for minimum sentences from 100-to-1 to 18-to-1. While the War on Drugs is still technically being waged, it is done at much less intense level than it was during its peak in the 1980s.

See original here:

War on Drugs | United States history | Britannica.com

Philippines War on Drugs | Human Rights Watch

Tilted election playing field in Turkey; European Court of Justice confirms rights of same-sex couples; Philippine policepromoting abusers; Vietnam’s cyber security law; Nigerian military trying to smear Amnesty International; Paris names imprisoned Bahrainrights activist Nabeel Rajaban honorary citizen; Intimidation ofjournalists in the US; Brutal US treatment of refugees; and Russia’s World Cup amid Syria slaughter.

Read this article:

Philippines War on Drugs | Human Rights Watch

The War on Drugs (band) – Wikipedia

The War on Drugs is an American indie rock band from Philadelphia, Pennsylvania, formed in 2005. The band consists of Adam Granduciel (lyrics, vocals, guitar), David Hartley (bass), Robbie Bennett (keyboards), Charlie Hall (drums), Jon Natchez (saxophone, keyboards) and Anthony LaMarca (guitar).

Founded by close collaborators Granduciel and Kurt Vile, The War on Drugs released their debut studio album, Wagonwheel Blues, in 2008. Vile departed shortly after its release to focus on his solo career. The band’s second studio album Slave Ambient was released in 2011 to favorable reviews and extensive touring.

The band’s third album, Lost in the Dream, was released in 2014 following extensive touring and a period of loneliness and depression for primary songwriter Granduciel. The album was released to widespread critical acclaim and increased exposure. Previous collaborator Hall joined the band as its full-time drummer during the recording process, with saxophonist Natchez and additional guitarist LaMarca accompanying the band for its world tour. Signing to Atlantic Records, the six-piece band released their fourth album, A Deeper Understanding, in 2017, which won the Grammy Award for Best Rock Album at the 60th Annual Grammy Awards.

In 2003, frontman Adam Granduciel moved from Oakland, California to Philadelphia, where he met Kurt Vile, who had also recently moved back to Philadelphia after living in Boston for two years.[2] The duo subsequently began writing, recording and performing music together.[3] Vile stated, “Adam was the first dude I met when I moved back to Philadelphia in 2003. We saw eye-to-eye on a lot of things. I was obsessed with Bob Dylan at the time, and we totally geeked-out on that. We started playing together in the early days and he would be in my band, The Violators. Then, eventually I played in The War On Drugs.”[4]

Granduciel and Vile began playing together as The War on Drugs in 2005. Regarding the band’s name, Granduciel noted, “My friend Julian and I came up with it a few years ago over a couple bottles of red wine and a few typewriters when we were living in Oakland. We were writing a lot back then, working on a dictionary, and it just came out and we were like “hey, good band name” so eventually when I moved to Philadelphia and got a band together I used it. It was either that or The Rigatoni Danzas. I think we made the right choice. I always felt though that it was the kind of name I could record all sorts of different music under without any sort of predictability inherent in the name”[5]

While Vile and Granduciel formed the backbone of the band, they had a number of accompanists early in the group’s career, before finally settling on a lineup that added Charlie Hall as drummer/organist, Kyle Lloyd as drummer and Dave Hartley on bass.[6] Granduciel had previously toured and recorded with The Capitol Years, and Vile has several solo albums.[7] The group gave away its Barrel of Batteries EP for free early in 2008.[8] Their debut LP for Secretly Canadian, Wagonwheel Blues, was released in 2008.[9]

Following the album’s release, and subsequent European tour, Vile departed from the band to focus on his solo career, stating, “I only went on the first European tour when their album came out, and then I basically left the band. I knew if I stuck with that, it would be all my time and my goal was to have my own musical career.”[4] Fellow Kurt Vile & the Violators bandmate Mike Zanghi joined the band at this time, with Vile noting, “Mike was my drummer first and then when The War On Drugs’ first record came out I thought I was lending Mike to Adam for the European tour but then he just played with them all the time so I kind of had to like, while they were touring a lot, figure out my own thing.”[10]

The lineup underwent several changes, and by the end of 2008, Kurt Vile, Charlie Hall, and Kyle Lloyd had all exited the group. At that time Granduciel and Hartley were joined by drummer Mike Zanghi, whom Granduciel also played with in Kurt Vile’s backing band, the Violators.

After recording much of the band’s forthcoming studio album, Slave Ambient, Zanghi departed from the band in 2010. Drummer Steven Urgo subsequently joined the band, with keyboardist Robbie Bennett also joining at around this time. Regarding Zanghi’s exit, Granduciel noted: “I loved Mike, and I loved the sound of The Violators, but then he wasn’t really the sound of my band. But you have things like friendship, and he’s down to tour and he’s a great guy, but it wasn’t the sound of what this band was.”[11]

Slave Ambient was released to favorable reviews in 2011.[citation needed]

In 2012, Patrick Berkery replaced Urgo as the band’s drummer.[12]

On December 4, 2013 the band announced the upcoming release of its third studio album, Lost in the Dream (March 18, 2014). The band streamed the album in its entirety on NPR’s First Listen site for a week before its release.[13]

Lost in the Dream was featured as the Vinyl Me, Please record of the month in August 2014. The pressing was a limited edition pressing on mint green colored vinyl.

In June 2015, The War on Drugs signed with Atlantic Records for a two-album deal.[14]

On Record Store Day, April 22, 2017, The War on Drugs released their new single “Thinking of a Place.”[15] The single was produced by frontman Granduciel and Shawn Everett.[16] April 28, 2017, The War on Drugs announced a fall 2017 tour in North America and Europe and that a new album was imminent.[17] On June 1, 2017, a new song, “Holding On”, was released, and it was announced that the album would be titled A Deeper Understanding and was released on August 25, 2017.[18]

The 2017 tour begins in September, opening in the band’s hometown, Philadelphia, and it concludes in November in Sweden.[19]

A Deeper Understanding was nominated for the International Album of the Year award at the 2018 UK Americana Awards[20].

At the 60th Annual Grammy Awards, on January 28th, 2018, A Deeper Understanding won the Grammy for Best Rock Album [21]

Granduciel and Zanghi are both former members of founding guitarist Vile’s backing band The Violators, with Granduciel noting, “There was never, despite what lazy journalists have assumed, any sort of falling out, or resentment”[22] following Vile’s departure from The War on Drugs. In 2011, Vile stated, “When my record came out, I assumed Adam would want to focus on The War On Drugs but he came with us in The Violators when we toured the States. The Violators became a unit, and although the cast does rotate, we’ve developed an even tighter unity and sound. Adam is an incredible guitar player these days and there is a certain feeling [between us] that nobody else can tap into. We don’t really have to tell each other what to play, it just happens.”

Both Hartley and Granduciel contributed to singer-songwriter Sharon Van Etten’s fourth studio album, Are We There (2014). Hartley performs bass guitar on the entire album, with Granduciel contributing guitar on two tracks.

Granduciel is currently[when?] producing the new Sore Eros album. They have been recording it in Philadelphia and Los Angeles on and off for the past several years.[4]

In 2016, The War on Drugs contributed a cover of “Touch of Grey” for a Grateful Dead tribute album called Day of the Dead. The album was curated by The National’s Aaron and Bryce Dessner.[19]

Current members

Former members

Follow this link:

The War on Drugs (band) – Wikipedia

A Brief History of the Drug War | Drug Policy Alliance

This video from hip hop legend Jay Z and acclaimed artist Molly Crabapple depicts the drug wars devastating impact on the Black community from decades of biased law enforcement.

The video traces the drug war from President Nixon to the draconian Rockefeller Drug Laws to the emerging aboveground marijuana market that is poised to make legal millions for wealthy investors doing the same thing that generations of people of color have been arrested and locked up for. After you watch the video, read on to learn more about the discriminatory history of the war on drugs.

Many currently illegal drugs, such as marijuana, opium, coca, and psychedelics have been used for thousands of years for both medical and spiritual purposes. So why are some drugs legal and other drugs illegal today? It’s not based on any scientific assessment of the relative risks of these drugs but it has everything to do with who is associated with these drugs.

The first anti-opium laws in the 1870s were directed at Chinese immigrants. The first anti-cocaine laws in the early 1900s were directed at black men in the South. The first anti-marijuana laws, in the Midwest and the Southwest in the 1910s and 20s, were directed at Mexican migrants and Mexican Americans. Today, Latino and especially black communities are still subject to wildly disproportionate drug enforcement and sentencing practices.

In the 1960s, as drugs became symbols of youthful rebellion, social upheaval, and political dissent, the government halted scientific research to evaluate their medical safety and efficacy.

In June 1971, President Nixon declared a war on drugs. He dramatically increased the size and presence of federal drug control agencies, and pushed through measures such as mandatory sentencing and no-knock warrants.

A top Nixon aide, John Ehrlichman, later admitted: You want to know what this was really all about. The Nixon campaign in 1968, and the Nixon White House after that, had two enemies: the antiwar left and black people. You understand what Im saying. We knew we couldnt make it illegal to be either against the war or black, but by getting the public to associate the hippies with marijuana and blacks with heroin, and then criminalizing both heavily, we could disrupt those communities. We could arrest their leaders, raid their homes, break up their meetings, and vilify them night after night on the evening news. Did we know we were lying about the drugs? Of course we did.Nixon temporarily placed marijuana in Schedule One, the most restrictive category of drugs, pending review by a commission he appointed led by Republican Pennsylvania Governor Raymond Shafer.

In 1972, the commission unanimously recommended decriminalizing the possession and distribution of marijuana for personal use. Nixon ignored the report and rejected its recommendations.

Between 1973 and 1977, however, eleven states decriminalized marijuana possession. In January 1977, President Jimmy Carter was inaugurated on a campaign platform that included marijuana decriminalization. In October 1977, the Senate Judiciary Committee voted to decriminalize possession of up to an ounce of marijuana for personal use.

Within just a few years, though, the tide had shifted. Proposals to decriminalize marijuana were abandoned as parents became increasingly concerned about high rates of teen marijuana use. Marijuana was ultimately caught up in a broader cultural backlash against the perceived permissiveness of the 1970s.

The presidency of Ronald Reagan marked the start of a long period of skyrocketing rates of incarceration, largely thanks to his unprecedented expansion of the drug war. The number of people behind bars for nonviolent drug law offenses increased from 50,000 in 1980 to over 400,000 by 1997.

Public concern about illicit drug use built throughout the 1980s, largely due to media portrayals of people addicted to the smokeable form of cocaine dubbed crack. Soon after Ronald Reagan took office in 1981, his wife, Nancy Reagan, began a highly-publicized anti-drug campaign, coining the slogan “Just Say No.”

This set the stage for the zero tolerance policies implemented in the mid-to-late 1980s. Los Angeles Police Chief Daryl Gates, who believed that casual drug users should be taken out and shot, founded the DARE drug education program, which was quickly adopted nationwide despite the lack of evidence of its effectiveness. The increasingly harsh drug policies also blocked the expansion of syringe access programs and other harm reduction policies to reduce the rapid spread of HIV/AIDS.

In the late 1980s, a political hysteria about drugs led to the passage of draconian penalties in Congress and state legislatures that rapidly increased the prison population. In 1985, the proportion of Americans polled who saw drug abuse as the nation’s “number one problem” was just 2-6 percent. The figure grew through the remainder of the 1980s until, in September 1989, it reached a remarkable 64 percent one of the most intense fixations by the American public on any issue in polling history. Within less than a year, however, the figure plummeted to less than 10 percent, as the media lost interest. The draconian policies enacted during the hysteria remained, however, and continued to result in escalating levels of arrests and incarceration.

Although Bill Clinton advocated for treatment instead of incarceration during his 1992 presidential campaign, after his first few months in the White House he reverted to the drug war strategies of his Republican predecessors by continuing to escalate the drug war. Notoriously, Clinton rejected a U.S. Sentencing Commission recommendation to eliminate the disparity between crack and powder cocaine sentences.

He also rejected, with the encouragement of drug czar General Barry McCaffrey, Health Secretary Donna Shalalas advice to end the federal ban on funding for syringe access programs. Yet, a month before leaving office, Clinton asserted in a Rolling Stone interview that “we really need a re-examination of our entire policy on imprisonment” of people who use drugs, and said that marijuana use “should be decriminalized.”

At the height of the drug war hysteria in the late 1980s and early 1990s, a movement emerged seeking a new approach to drug policy. In 1987, Arnold Trebach and Kevin Zeese founded the Drug Policy Foundation describing it as the loyal opposition to the war on drugs. Prominent conservatives such as William Buckley and Milton Friedman had long advocated for ending drug prohibition, as had civil libertarians such as longtime ACLU Executive Director Ira Glasser. In the late 1980s they were joined by Baltimore Mayor Kurt Schmoke, Federal Judge Robert Sweet, Princeton professor Ethan Nadelmann, and other activists, scholars and policymakers.

In 1994, Nadelmann founded The Lindesmith Center as the first U.S. project of George Soros Open Society Institute. In 2000, the growing Center merged with the Drug Policy Foundation to create the Drug Policy Alliance.

George W. Bush arrived in the White House as the drug war was running out of steam yet he allocated more money than ever to it. His drug czar, John Walters, zealously focused on marijuana and launched a major campaign to promote student drug testing. While rates of illicit drug use remained constant, overdose fatalities rose rapidly.

The era of George W. Bush also witnessed the rapid escalation of the militarization of domestic drug law enforcement. By the end of Bush’s term, there were about 40,000 paramilitary-style SWAT raids on Americans every year mostly for nonviolent drug law offenses, often misdemeanors. While federal reform mostly stalled under Bush, state-level reforms finally began to slow the growth of the drug war.

Politicians now routinely admit to having used marijuana, and even cocaine, when they were younger. When Michael Bloomberg was questioned during his 2001 mayoral campaign about whether he had ever used marijuana, he said, “You bet I did and I enjoyed it.” Barack Obama also candidly discussed his prior cocaine and marijuana use: “When I was a kid, I inhaled frequently that was the point.”

Public opinion has shifted dramatically in favor of sensible reforms that expand health-based approaches while reducing the role of criminalization in drug policy.

Marijuana reform has gained unprecedented momentum throughout the Americas. Alaska, California, Colorado, Nevada, Oregon, Maine, Massachusetts, Washington State, and Washington D.C. have legalized marijuana for adults. In December 2013, Uruguay became the first country in the world to legally regulate marijuana. In Canada, Prime Minister Justin Trudeau plans legalize marijuana for adults by 2018.

In response to a worsening overdose epidemic, dozens of U.S. states passed laws to increase access to the overdose antidote, naloxone, as well as 911 Good Samaritan laws to encourage people to seek medical help in the event of an overdose.

Yet the assault on American citizens and others continues, with 700,000 people still arrested for marijuana offenses each year and almost 500,000 people still behind bars for nothing more than a drug law violation.

President Obama, despite supporting several successful policy changes such as reducing the crack/powder sentencing disparity, ending the ban on federal funding for syringe access programs, and ending federal interference with state medical marijuana laws did not shift the majority of drug policy funding to a health-based approach.

Now, the new administration is threatening to take us backward toward a 1980s style drug war. President Trump is calling for a wall to keep drugs out of the country, and Attorney General Jeff Sessions has made it clear that he does not support the sovereignty of states to legalize marijuana, and believes good people dont smoke marijuana.

Progress is inevitably slow, and even with an administration hostile to reform there is still unprecedented momentum behind drug policy reform in states and localities across the country. The Drug Policy Alliance and its allies will continue to advocate for health-based reforms such as marijuana legalization, drug decriminalization, safe consumption sites, naloxone access, bail reform, and more.

We look forward to a future where drug policies are shaped by science and compassion rather than political hysteria.

Here is the original post:

A Brief History of the Drug War | Drug Policy Alliance

Artificial intelligence – Wikipedia

Artificial intelligence (AI, also machine intelligence, MI) is intelligence demonstrated by machines, in contrast to the natural intelligence (NI) displayed by humans and other animals. In computer science AI research is defined as the study of “intelligent agents”: any device that perceives its environment and takes actions that maximize its chance of successfully achieving its goals.[1] Colloquially, the term “artificial intelligence” is applied when a machine mimics “cognitive” functions that humans associate with other human minds, such as “learning” and “problem solving”.

The scope of AI is disputed: as machines become increasingly capable, tasks considered as requiring “intelligence” are often removed from the definition, a phenomenon known as the AI effect, leading to the quip, “AI is whatever hasn’t been done yet.”[3] For instance, optical character recognition is frequently excluded from “artificial intelligence”, having become a routine technology.[4] Capabilities generally classified as AI as of 2017[update] include successfully understanding human speech, competing at the highest level in strategic game systems (such as chess and Go[6]), autonomous cars, intelligent routing in content delivery network and military simulations.

Artificial intelligence was founded as an academic discipline in 1956, and in the years since has experienced several waves of optimism,[7][8] followed by disappointment and the loss of funding (known as an “AI winter”),[9][10] followed by new approaches, success and renewed funding.[8][11] For most of its history, AI research has been divided into subfields that often fail to communicate with each other.[12] These sub-fields are based on technical considerations, such as particular goals (e.g. “robotics” or “machine learning”),[13] the use of particular tools (“logic” or artificial neural networks), or deep philosophical differences.[14][15][16] Subfields have also been based on social factors (particular institutions or the work of particular researchers).[12]

The traditional problems (or goals) of AI research include reasoning, knowledge representation, planning, learning, natural language processing, perception and the ability to move and manipulate objects.[13] General intelligence is among the field’s long-term goals.[17] Approaches include statistical methods, computational intelligence, and traditional symbolic AI. Many tools are used in AI, including versions of search and mathematical optimization, artificial neural networks, and methods based on statistics, probability and economics. The AI field draws upon computer science, mathematics, psychology, linguistics, philosophy and many others.

The field was founded on the claim that human intelligence “can be so precisely described that a machine can be made to simulate it”.[18] This raises philosophical arguments about the nature of the mind and the ethics of creating artificial beings endowed with human-like intelligence which are issues that have been explored by myth, fiction and philosophy since antiquity.[19] Some people also consider AI to be a danger to humanity if it progresses unabatedly.[20] Others believe that AI, unlike previous technological revolutions, will create a risk of mass unemployment.[21]

In the twenty-first century, AI techniques have experienced a resurgence following concurrent advances in computer power, large amounts of data, and theoretical understanding; and AI techniques have become an essential part of the technology industry, helping to solve many challenging problems in computer science.[22][11]

Thought-capable artificial beings appeared as storytelling devices in antiquity,[23] and have been common in fiction, as in Mary Shelley’s Frankenstein or Karel apek’s R.U.R. (Rossum’s Universal Robots).[24] These characters and their fates raised many of the same issues now discussed in the ethics of artificial intelligence.[19]

The study of mechanical or “formal” reasoning began with philosophers and mathematicians in antiquity. The study of mathematical logic led directly to Alan Turing’s theory of computation, which suggested that a machine, by shuffling symbols as simple as “0” and “1”, could simulate any conceivable act of mathematical deduction. This insight, that digital computers can simulate any process of formal reasoning, is known as the ChurchTuring thesis.[25] Along with concurrent discoveries in neurobiology, information theory and cybernetics, this led researchers to consider the possibility of building an electronic brain. Turing proposed that “if a human could not distinguish between responses from a machine and a human, the machine could be considered intelligent”.[26] The first work that is now generally recognized as AI was McCullouch and Pitts’ 1943 formal design for Turing-complete “artificial neurons”.

The field of AI research was born at a workshop at Dartmouth College in 1956.[28] Attendees Allen Newell (CMU), Herbert Simon (CMU), John McCarthy (MIT), Marvin Minsky (MIT) and Arthur Samuel (IBM) became the founders and leaders of AI research.[29] They and their students produced programs that the press described as “astonishing”: computers were learning checkers strategies (c. 1954)[31] (and by 1959 were reportedly playing better than the average human),[32] solving word problems in algebra, proving logical theorems (Logic Theorist, first run c. 1956) and speaking English.[33] By the middle of the 1960s, research in the U.S. was heavily funded by the Department of Defense[34] and laboratories had been established around the world.[35] AI’s founders were optimistic about the future: Herbert Simon predicted, “machines will be capable, within twenty years, of doing any work a man can do”. Marvin Minsky agreed, writing, “within a generation… the problem of creating ‘artificial intelligence’ will substantially be solved”.[7]

They failed to recognize the difficulty of some of the remaining tasks. Progress slowed and in 1974, in response to the criticism of Sir James Lighthill and ongoing pressure from the US Congress to fund more productive projects, both the U.S. and British governments cut off exploratory research in AI. The next few years would later be called an “AI winter”,[9] a period when obtaining funding for AI projects was difficult.

In the early 1980s, AI research was revived by the commercial success of expert systems,[37] a form of AI program that simulated the knowledge and analytical skills of human experts. By 1985 the market for AI had reached over a billion dollars. At the same time, Japan’s fifth generation computer project inspired the U.S and British governments to restore funding for academic research.[8] However, beginning with the collapse of the Lisp Machine market in 1987, AI once again fell into disrepute, and a second, longer-lasting hiatus began.[10]

In the late 1990s and early 21st century, AI began to be used for logistics, data mining, medical diagnosis and other areas.[22] The success was due to increasing computational power (see Moore’s law), greater emphasis on solving specific problems, new ties between AI and other fields (such as statistics, economics and mathematics), and a commitment by researchers to mathematical methods and scientific standards.[38] Deep Blue became the first computer chess-playing system to beat a reigning world chess champion, Garry Kasparov on 11 May 1997.

In 2011, a Jeopardy! quiz show exhibition match, IBM’s question answering system, Watson, defeated the two greatest Jeopardy champions, Brad Rutter and Ken Jennings, by a significant margin. Faster computers, algorithmic improvements, and access to large amounts of data enabled advances in machine learning and perception; data-hungry deep learning methods started to dominate accuracy benchmarks around 2012.[41] The Kinect, which provides a 3D bodymotion interface for the Xbox 360 and the Xbox One use algorithms that emerged from lengthy AI research[42] as do intelligent personal assistants in smartphones.[43] In March 2016, AlphaGo won 4 out of 5 games of Go in a match with Go champion Lee Sedol, becoming the first computer Go-playing system to beat a professional Go player without handicaps.[6][44] In the 2017 Future of Go Summit, AlphaGo won a three-game match with Ke Jie,[45] who at the time continuously held the world No. 1 ranking for two years.[46][47] This marked the completion of a significant milestone in the development of Artificial Intelligence as Go is an extremely complex game, more so than Chess.

According to Bloomberg’s Jack Clark, 2015 was a landmark year for artificial intelligence, with the number of software projects that use AI within Google increased from a “sporadic usage” in 2012 to more than 2,700 projects. Clark also presents factual data indicating that error rates in image processing tasks have fallen significantly since 2011.[48] He attributes this to an increase in affordable neural networks, due to a rise in cloud computing infrastructure and to an increase in research tools and datasets.[11] Other cited examples include Microsoft’s development of a Skype system that can automatically translate from one language to another and Facebook’s system that can describe images to blind people.[48] In a 2017 survey, one in five companies reported they had “incorporated AI in some offerings or processes”.[49][50]

A typical AI perceives its environment and takes actions that maximize its chance of successfully achieving its goals.[1] An AI’s intended goal function can be simple (“1 if the AI wins a game of Go, 0 otherwise”) or complex (“Do actions mathematically similar to the actions that got you rewards in the past”). Goals can be explicitly defined, or can be induced. If the AI is programmed for “reinforcement learning”, goals can be implicitly induced by rewarding some types of behavior and punishing others.[a] Alternatively, an evolutionary system can induce goals by using a “fitness function” to mutate and preferentially replicate high-scoring AI systems; this is similar to how animals evolved to innately desire certain goals such as finding food, or how dogs can be bred via artificial selection to possess desired traits. Some AI systems, such as nearest-neighbor, instead reason by analogy; these systems are not generally given goals, except to the degree that goals are somehow implicit in their training data. Such systems can still be benchmarked if the non-goal system is framed as a system whose “goal” is to successfully accomplish its narrow classification task.[53]

AI often revolves around the use of algorithms. An algorithm is a set of unambiguous instructions that a mechanical computer can execute.[b] A complex algorithm is often built on top of other, simpler, algorithms. A simple example of an algorithm is the following recipe for optimal play at tic-tac-toe:

Many AI algorithms are capable of learning from data; they can enhance themselves by learning new heuristics (strategies, or “rules of thumb”, that have worked well in the past), or can themselves write other algorithms. Some of the “learners” described below, including Bayesian networks, decision trees, and nearest-neighbor, could theoretically, if given infinite data, time, and memory, learn to approximate any function, including whatever combination of mathematical functions would best describe the entire world. These learners could therefore, in theory, derive all possible knowledge, by considering every possible hypothesis and matching it against the data. In practice, it is almost never possible to consider every possibility, because of the phenomenon of “combinatorial explosion”, where the amount of time needed to solve a problem grows exponentially. Much of AI research involves figuring out how to identify and avoid considering broad swaths of possibililities that are unlikely to be fruitful.[55] For example, when viewing a map and looking for the shortest driving route from Denver to New York in the East, one can in most cases skip looking at any path through San Francisco or other areas far to the West; thus, an AI wielding an pathfinding algorithm like A* can avoid the combinatorial explosion that would ensue if every possible route had to be ponderously considered in turn.[57]

The earliest (and easiest to understand) approach to AI was symbolism (such as formal logic): “If an otherwise healthy adult has a fever, then they may have influenza”. A second, more general, approach is Bayesian inference: “If the current patient has a fever, adjust the probability they have influenza in such-and-such way”. The third major approach, extremely popular in routine business AI applications, are analogizers such as SVM and nearest-neighbor: “After examining the records of known past patients whose temperature, symptoms, age, and other factors mostly match the current patient, X% of those patients turned out to have influenza”. A fourth approach is harder to intuitively understand, but is inspired by how the brain’s machinery works: the artificial neural network approach uses artificial “neurons” that can learn by comparing itself to the desired output and altering the strengths of the connections between its internal neurons to “reinforce” connections that seemed to be useful. These four main approaches can overlap with each other and with evolutionary systems; for example, neural nets can learn to make inferences, to generalize, and to make analogies. Some systems implicitly or explicitly use multiple of these approaches, alongside many other AI and non-AI algorithms; the best approach is often different depending on the problem.[59]

Learning algorithms work on the basis that strategies, algorithms, and inferences that worked well in the past are likely to continue working well in the future. These inferences can be obvious, such as “since the sun rose every morning for the last 10,000 days, it will probably rise tomorrow morning as well”. They can be nuanced, such as “X% of families have geographically separate species with color variants, so there is an Y% chance that undiscovered black swans exist”. Learners also work on the basis of “Occam’s razor”: The simplest theory that explains the data is the likeliest. Therefore, to be successful, a learner must be designed such that it prefers simpler theories to complex theories, except in cases where the complex theory is proven substantially better. Settling on a bad, overly complex theory gerrymandered to fit all the past training data is known as overfitting. Many systems attempt to reduce overfitting by rewarding a theory in accordance with how well it fits the data, but penalizing the theory in accordance with how complex the theory is. Besides classic overfitting, learners can also disappoint by “learning the wrong lesson”. A toy example is that an image classifier trained only on pictures of brown horses and black cats might conclude that all brown patches are likely to be horses. A real-world example is that, unlike humans, current image classifiers don’t determine the spatial relationship between components of the picture; instead, they learn abstract patterns of pixels that humans are oblivious to, but that linearly correlate with images of certain types of real objects. Faintly superimposing such a pattern on a legitimate image results in an “adversarial” image that the system misclassifies.[c][62][63][64]

Compared with humans, existing AI lacks several features of human “commonsense reasoning”; most notably, humans have powerful mechanisms for reasoning about “nave physics” such as space, time, and physical interactions. This enables even young children to easily make inferences like “If I roll this pen off a table, it will fall on the floor”. Humans also have a powerful mechanism of “folk psychology” that helps them to interpret natural-language sentences such as “The city councilmen refused the demonstrators a permit because they advocated violence”. (A generic AI has difficulty inferring whether the councilmen or the demonstrators are the ones alleged to be advocating violence.)[67][68][69] This lack of “common knowledge” means that AI often makes different mistakes than humans make, in ways that can seem incomprehensible. For example, existing self-driving cars cannot reason about the location nor the intentions of pedestrians in the exact way that humans do, and instead must use non-human modes of reasoning to avoid accidents.[70][71][72]

The overall research goal of artificial intelligence is to create technology that allows computers and machines to function in an intelligent manner. The general problem of simulating (or creating) intelligence has been broken down into sub-problems. These consist of particular traits or capabilities that researchers expect an intelligent system to display. The traits described below have received the most attention.[13]

Early researchers developed algorithms that imitated step-by-step reasoning that humans use when they solve puzzles or make logical deductions.[73] By the late 1980s and 1990s, AI research had developed methods for dealing with uncertain or incomplete information, employing concepts from probability and economics.[74]

These algorithms proved to be insufficient for solving large reasoning problems, because they experienced a “combinatorial explosion”: they became exponentially slower as the problems grew larger.[55] In fact, even humans rarely use the step-by-step deduction that early AI research was able to model. They solve most of their problems using fast, intuitive judgements.[75]

Knowledge representation[76] and knowledge engineering[77] are central to classical AI research. Some “expert systems” attempt to gather together explicit knowledge possessed by experts in some narrow domain. In addition, some projects attempt to gather the “commonsense knowledge” known to the average person into a database containing extensive knowledge about the world. Among the things a comprehensive commonsense knowledge base would contain are: objects, properties, categories and relations between objects;[78] situations, events, states and time;[79] causes and effects;[80] knowledge about knowledge (what we know about what other people know);[81] and many other, less well researched domains. A representation of “what exists” is an ontology: the set of objects, relations, concepts, and properties formally described so that software agents can interpret them. The semantics of these are captured as description logic concepts, roles, and individuals, and typically implemented as classes, properties, and individuals in the Web Ontology Language.[82] The most general ontologies are called upper ontologies, which attempt to provide a foundation for all other knowledge[83] by acting as mediators between domain ontologies that cover specific knowledge about a particular knowledge domain (field of interest or area of concern). Such formal knowledge representations can be used in content-based indexing and retrieval,[84] scene interpretation,[85] clinical decision support,[86] knowledge discovery (mining “interesting” and actionable inferences from large databases),[87] and other areas.[88]

Among the most difficult problems in knowledge representation are:

Intelligent agents must be able to set goals and achieve them.[95] They need a way to visualize the futurea representation of the state of the world and be able to make predictions about how their actions will change itand be able to make choices that maximize the utility (or “value”) of available choices.[96]

In classical planning problems, the agent can assume that it is the only system acting in the world, allowing the agent to be certain of the consequences of its actions.[97] However, if the agent is not the only actor, then it requires that the agent can reason under uncertainty. This calls for an agent that can not only assess its environment and make predictions, but also evaluate its predictions and adapt based on its assessment.[98]

Multi-agent planning uses the cooperation and competition of many agents to achieve a given goal. Emergent behavior such as this is used by evolutionary algorithms and swarm intelligence.[99]

Machine learning, a fundamental concept of AI research since the field’s inception,[100] is the study of computer algorithms that improve automatically through experience.[101][102]

Unsupervised learning is the ability to find patterns in a stream of input. Supervised learning includes both classification and numerical regression. Classification is used to determine what category something belongs in, after seeing a number of examples of things from several categories. Regression is the attempt to produce a function that describes the relationship between inputs and outputs and predicts how the outputs should change as the inputs change.[102] Both classifiers and regression learners can be viewed as “function approximators” trying to learn an unknown (possibly implicit) function; for example, a spam classifier can be viewed as learning a function that maps from the text of an email to one of two categories, “spam” or “not spam”. Computational learning theory can assess learners by computational complexity, by sample complexity (how much data is required), or by other notions of optimization.[103] In reinforcement learning[104] the agent is rewarded for good responses and punished for bad ones. The agent uses this sequence of rewards and punishments to form a strategy for operating in its problem space.

Natural language processing[105] (NLP) gives machines the ability to read and understand human language. A sufficiently powerful natural language processing system would enable natural-language user interfaces and the acquisition of knowledge directly from human-written sources, such as newswire texts. Some straightforward applications of natural language processing include information retrieval, text mining, question answering[106] and machine translation.[107] Many current approaches use word co-occurrence frequencies to construct syntactic representations of text. “Keyword spotting” strategies for search are popular and scalable but dumb; a search query for “dog” might only match documents with the literal word “dog” and miss a document with the word “poodle”. “Lexical affinity” strategies use the occurrence of words such as “accident” to assess the sentiment of a document. Modern statistical NLP approaches can combine all these strategies as well as others, and often achieve acceptable accuracy at the page or paragraph level, but continue to lack the semantic understanding required to classify isolated sentences well. Besides the usual difficulties with encoding semantic commonsense knowledge, existing semantic NLP sometimes scales too poorly to be viable in business applications. Beyond semantic NLP, the ultimate goal of “narrative” NLP is to embody a full understanding of commonsense reasoning.[108]

Machine perception[109] is the ability to use input from sensors (such as cameras (visible spectrum or infrared), microphones, wireless signals, and active lidar, sonar, radar, and tactile sensors) to deduce aspects of the world. Applications include speech recognition,[110] facial recognition, and object recognition.[111] Computer vision is the ability to analyze visual input. Such input is usually ambiguous; a giant, fifty-meter-tall pedestrian far away may produce exactly the same pixels as a nearby normal-sized pedestrian, requiring the AI to judge the relative likelihood and reasonableness of different interpretations, for example by using its “object model” to assess that fifty-meter pedestrians do not exist.[112]

AI is heavily used in robotics.[113] Advanced robotic arms and other industrial robots, widely used in modern factories, can learn from experience how to move efficiently despite the presence of friction and gear slippage.[114] A modern mobile robot, when given a small, static, and visible environment, can easily determine its location and map its environment; however, dynamic environments, such as (in endoscopy) the interior of a patient’s breathing body, pose a greater challenge. Motion planning is the process of breaking down a movement task into “primitives” such as individual joint movements. Such movement often involves compliant motion, a process where movement requires maintaining physical contact with an object.[116][117] Moravec’s paradox generalizes that low-level sensorimotor skills that humans take for granted are, counterintuitively, difficult to program into a robot; the paradox is named after Hans Moravec, who stated in 1988 that “it is comparatively easy to make computers exhibit adult level performance on intelligence tests or playing checkers, and difficult or impossible to give them the skills of a one-year-old when it comes to perception and mobility”.[118][119] This is attributed to the fact that, unlike checkers, physical dexterity has been a direct target of natural selection for millions of years.[120]

Moravec’s paradox can be extended to many forms of social intelligence.[122][123] Distributed multi-agent coordination of autonomous vehicles remains a difficult problem.[124] Affective computing is an interdisciplinary umbrella that comprises systems which recognize, interpret, process, or simulate human affects. Moderate successes related to affective computing include textual sentiment analysis and, more recently, multimodal affect analysis, wherein AI classifies the affects displayed by a videotaped subject.[128]

In the long run, social skills and an understanding of human emotion and game theory would be valuable to a social agent. Being able to predict the actions of others by understanding their motives and emotional states would allow an agent to make better decisions. Some computer systems mimic human emotion and expressions to appear more sensitive to the emotional dynamics of human interaction, or to otherwise facilitate humancomputer interaction.[129] Similarly, some virtual assistants are programmed to speak conversationally or even to banter humorously; this tends to give naive users an unrealistic conception of how intelligent existing computer agents actually are.[130]

Historically, projects such as the Cyc knowledge base (1984) and the massive Japanese Fifth Generation Computer Systems initiative (19821992) attempted to cover the breadth of human cognition. These early projects failed to escape the limitations of non-quantitative symbolic logic models and, in retrospect, greatly underestimated the difficulty of cross-domain AI. Nowadays, the vast majority of current AI researchers work instead on tractable “narrow AI” applications (such as medical diagnosis or automobile navigation).[131] Many researchers predict that such “narrow AI” work in different individual domains will eventually be incorporated into a machine with artificial general intelligence (AGI), combining most of the narrow skills mentioned in this article and at some point even exceeding human ability in most or all these areas.[17][132] Many advances have general, cross-domain significance. One high-profile example is that DeepMind in the 2010s developed a “generalized artificial intelligence” that could learn many diverse Atari games on its own, and later developed a variant of the system which succeeds at sequential learning.[133][134][135] Besides transfer learning,[136] hypothetical AGI breakthroughs could include the development of reflective architectures that can engage in decision-theoretic metareasoning, and figuring out how to “slurp up” a comprehensive knowledge base from the entire unstructured Web. Some argue that some kind of (currently-undiscovered) conceptually straightforward, but mathematically difficult, “Master Algorithm” could lead to AGI. Finally, a few “emergent” approaches look to simulating human intelligence extremely closely, and believe that anthropomorphic features like an artificial brain or simulated child development may someday reach a critical point where general intelligence emerges.[138][139]

Many of the problems in this article may also require general intelligence, if machines are to solve the problems as well as people do. For example, even specific straightforward tasks, like machine translation, require that a machine read and write in both languages (NLP), follow the author’s argument (reason), know what is being talked about (knowledge), and faithfully reproduce the author’s original intent (social intelligence). A problem like machine translation is considered “AI-complete”, because all of these problems need to be solved simultaneously in order to reach human-level machine performance.

There is no established unifying theory or paradigm that guides AI research. Researchers disagree about many issues.[140] A few of the most long standing questions that have remained unanswered are these: should artificial intelligence simulate natural intelligence by studying psychology or neurobiology? Or is human biology as irrelevant to AI research as bird biology is to aeronautical engineering?[14] Can intelligent behavior be described using simple, elegant principles (such as logic or optimization)? Or does it necessarily require solving a large number of completely unrelated problems?[15]

In the 1940s and 1950s, a number of researchers explored the connection between neurobiology, information theory, and cybernetics. Some of them built machines that used electronic networks to exhibit rudimentary intelligence, such as W. Grey Walter’s turtles and the Johns Hopkins Beast. Many of these researchers gathered for meetings of the Teleological Society at Princeton University and the Ratio Club in England.[141] By 1960, this approach was largely abandoned, although elements of it would be revived in the 1980s.

When access to digital computers became possible in the middle 1950s, AI research began to explore the possibility that human intelligence could be reduced to symbol manipulation. The research was centered in three institutions: Carnegie Mellon University, Stanford and MIT, and as described below, each one developed its own style of research. John Haugeland named these symbolic approaches to AI “good old fashioned AI” or “GOFAI”.[142] During the 1960s, symbolic approaches had achieved great success at simulating high-level thinking in small demonstration programs. Approaches based on cybernetics or artificial neural networks were abandoned or pushed into the background.[143] Researchers in the 1960s and the 1970s were convinced that symbolic approaches would eventually succeed in creating a machine with artificial general intelligence and considered this the goal of their field.

Economist Herbert Simon and Allen Newell studied human problem-solving skills and attempted to formalize them, and their work laid the foundations of the field of artificial intelligence, as well as cognitive science, operations research and management science. Their research team used the results of psychological experiments to develop programs that simulated the techniques that people used to solve problems. This tradition, centered at Carnegie Mellon University would eventually culminate in the development of the Soar architecture in the middle 1980s.[144][145]

Unlike Simon and Newell, John McCarthy felt that machines did not need to simulate human thought, but should instead try to find the essence of abstract reasoning and problem solving, regardless of whether people used the same algorithms.[14] His laboratory at Stanford (SAIL) focused on using formal logic to solve a wide variety of problems, including knowledge representation, planning and learning.[146] Logic was also the focus of the work at the University of Edinburgh and elsewhere in Europe which led to the development of the programming language Prolog and the science of logic programming.[147]

Researchers at MIT (such as Marvin Minsky and Seymour Papert)[148] found that solving difficult problems in vision and natural language processing required ad-hoc solutions they argued that there was no simple and general principle (like logic) that would capture all the aspects of intelligent behavior. Roger Schank described their “anti-logic” approaches as “scruffy” (as opposed to the “neat” paradigms at CMU and Stanford).[15] Commonsense knowledge bases (such as Doug Lenat’s Cyc) are an example of “scruffy” AI, since they must be built by hand, one complicated concept at a time.[149]

When computers with large memories became available around 1970, researchers from all three traditions began to build knowledge into AI applications.[150] This “knowledge revolution” led to the development and deployment of expert systems (introduced by Edward Feigenbaum), the first truly successful form of AI software.[37] The knowledge revolution was also driven by the realization that enormous amounts of knowledge would be required by many simple AI applications.

By the 1980s progress in symbolic AI seemed to stall and many believed that symbolic systems would never be able to imitate all the processes of human cognition, especially perception, robotics, learning and pattern recognition. A number of researchers began to look into “sub-symbolic” approaches to specific AI problems.[16] Sub-symbolic methods manage to approach intelligence without specific representations of knowledge.

This includes embodied, situated, behavior-based, and nouvelle AI. Researchers from the related field of robotics, such as Rodney Brooks, rejected symbolic AI and focused on the basic engineering problems that would allow robots to move and survive.[151] Their work revived the non-symbolic viewpoint of the early cybernetics researchers of the 1950s and reintroduced the use of control theory in AI. This coincided with the development of the embodied mind thesis in the related field of cognitive science: the idea that aspects of the body (such as movement, perception and visualization) are required for higher intelligence.

Within developmental robotics, developmental learning approaches are elaborated upon to allow robots to accumulate repertoires of novel skills through autonomous self-exploration, social interaction with human teachers, and the use of guidance mechanisms (active learning, maturation, motor synergies, etc.).[152][153]

Interest in neural networks and “connectionism” was revived by David Rumelhart and others in the middle of the 1980s.[156] Artificial neural networks are an example of soft computing — they are solutions to problems which cannot be solved with complete logical certainty, and where an approximate solution is often sufficient. Other soft computing approaches to AI include fuzzy systems, evolutionary computation and many statistical tools. The application of soft computing to AI is studied collectively by the emerging discipline of computational intelligence.[157]

Much of GOFAI got bogged down on ad hoc patches to symbolic computation that worked on their own toy models but failed to generalize to real-world results. However, around the 1990s, AI researchers adopted sophisticated mathematical tools, such as hidden Markov models (HMM), information theory, and normative Bayesian decision theory to compare or to unify competing architectures. The shared mathematical language permitted a high level of collaboration with more established fields (like mathematics, economics or operations research).[d] Compared with GOFAI, new “statistical learning” techniques such as HMM and neural networks were gaining higher levels of accuracy in many practical domains such as data mining, without necessarily acquiring semantic understanding of the datasets. The increased successes with real-world data led to increasing emphasis on comparing different approaches against shared test data to see which approach performed best in a broader context than that provided by idiosyncratic toy models; AI research was becoming more scientific. Nowadays results of experiments are often rigorously measurable, and are sometimes (with difficulty) reproducible.[38][158] Different statistical learning techniques have different limitations; for example, basic HMM cannot model the infinite possible combinations of natural language. Critics note that the shift from GOFAI to statistical learning is often also a shift away from Explainable AI. In AGI research, some scholars caution against over-reliance on statistical learning, and argue that continuing research into GOFAI will still be necessary to attain general intelligence.

AI has developed a large number of tools to solve the most difficult problems in computer science. A few of the most general of these methods are discussed below.

Many problems in AI can be solved in theory by intelligently searching through many possible solutions:[167] Reasoning can be reduced to performing a search. For example, logical proof can be viewed as searching for a path that leads from premises to conclusions, where each step is the application of an inference rule.[168] Planning algorithms search through trees of goals and subgoals, attempting to find a path to a target goal, a process called means-ends analysis.[169] Robotics algorithms for moving limbs and grasping objects use local searches in configuration space.[114] Many learning algorithms use search algorithms based on optimization.

Simple exhaustive searches[170] are rarely sufficient for most real world problems: the search space (the number of places to search) quickly grows to astronomical numbers. The result is a search that is too slow or never completes. The solution, for many problems, is to use “heuristics” or “rules of thumb” that prioritize choices in favor of those that are more likely to reach a goal, and to do so in a shorter number of steps. In some search methodologies heuristics can also serve to entirely eliminate some choices that are unlikely to lead to a goal (called “pruning the search tree”). Heuristics supply the program with a “best guess” for the path on which the solution lies.[171] Heuristics limit the search for solutions into a smaller sample size.

A very different kind of search came to prominence in the 1990s, based on the mathematical theory of optimization. For many problems, it is possible to begin the search with some form of a guess and then refine the guess incrementally until no more refinements can be made. These algorithms can be visualized as blind hill climbing: we begin the search at a random point on the landscape, and then, by jumps or steps, we keep moving our guess uphill, until we reach the top. Other optimization algorithms are simulated annealing, beam search and random optimization.[172]

Evolutionary computation uses a form of optimization search. For example, they may begin with a population of organisms (the guesses) and then allow them to mutate and recombine, selecting only the fittest to survive each generation (refining the guesses). Classic evolutionary algorithms include genetic algorithms, gene expression programming, and genetic programming.[173] Alternatively, distributed search processes can coordinate via swarm intelligence algorithms. Two popular swarm algorithms used in search are particle swarm optimization (inspired by bird flocking) and ant colony optimization (inspired by ant trails).[174][175]

Logic[176] is used for knowledge representation and problem solving, but it can be applied to other problems as well. For example, the satplan algorithm uses logic for planning[177] and inductive logic programming is a method for learning.[178]

Several different forms of logic are used in AI research. Propositional logic[179] involves truth functions such as “or” and “not”. First-order logic[180] adds quantifiers and predicates, and can express facts about objects, their properties, and their relations with each other. Fuzzy set theory assigns a “degree of truth” (between 0 and 1) to vague statements such as “Alice is old” (or rich, or tall, or hungry) that are too linguistically imprecise to be completely true or false. Fuzzy logic is successfully used in control systems to allow experts to contribute vague rules such as “if you are close to the destination station and moving fast, increase the train’s brake pressure”; these vague rules can then be numerically refined within the system. Fuzzy logic fails to scale well in knowledge bases; many AI researchers question the validity of chaining fuzzy-logic inferences.[e][182][183]

Default logics, non-monotonic logics and circumscription[90] are forms of logic designed to help with default reasoning and the qualification problem. Several extensions of logic have been designed to handle specific domains of knowledge, such as: description logics;[78] situation calculus, event calculus and fluent calculus (for representing events and time);[79] causal calculus;[80] belief calculus;[184] and modal logics.[81]

Overall, qualitiative symbolic logic is brittle and scales poorly in the presence of noise or other uncertainty. Exceptions to rules are numerous, and it is difficult for logical systems to function in the presence of contradictory rules.[186]

Many problems in AI (in reasoning, planning, learning, perception and robotics) require the agent to operate with incomplete or uncertain information. AI researchers have devised a number of powerful tools to solve these problems using methods from probability theory and economics.[187]

Bayesian networks[188] are a very general tool that can be used for a large number of problems: reasoning (using the Bayesian inference algorithm),[189] learning (using the expectation-maximization algorithm),[f][191] planning (using decision networks)[192] and perception (using dynamic Bayesian networks).[193] Probabilistic algorithms can also be used for filtering, prediction, smoothing and finding explanations for streams of data, helping perception systems to analyze processes that occur over time (e.g., hidden Markov models or Kalman filters).[193] Compared with symbolic logic, formal Bayesian inference is computationally expensive. For inference to be tractable, most observations must be conditionally independent of one another. Complicated graphs with diamonds or other “loops” (undirected cycles) can require a sophisticated method such as Markov Chain Monte Carlo, which spreads an ensemble of random walkers throughout the Bayesian network and attempts to converge to an assessment of the conditional probabilities. Bayesian networks are used on XBox Live to rate and match players; wins and losses are “evidence” of how good a player is. AdSense uses a Bayesian network with over 300 million edges to learn which ads to serve.

A key concept from the science of economics is “utility”: a measure of how valuable something is to an intelligent agent. Precise mathematical tools have been developed that analyze how an agent can make choices and plan, using decision theory, decision analysis,[195] and information value theory.[96] These tools include models such as Markov decision processes,[196] dynamic decision networks,[193] game theory and mechanism design.[197]

The simplest AI applications can be divided into two types: classifiers (“if shiny then diamond”) and controllers (“if shiny then pick up”). Controllers do, however, also classify conditions before inferring actions, and therefore classification forms a central part of many AI systems. Classifiers are functions that use pattern matching to determine a closest match. They can be tuned according to examples, making them very attractive for use in AI. These examples are known as observations or patterns. In supervised learning, each pattern belongs to a certain predefined class. A class can be seen as a decision that has to be made. All the observations combined with their class labels are known as a data set. When a new observation is received, that observation is classified based on previous experience.[198]

A classifier can be trained in various ways; there are many statistical and machine learning approaches. The decision tree[199] is perhaps the most widely used machine learning algorithm. Other widely used classifiers are the neural network,[201] k-nearest neighbor algorithm,[g][203] kernel methods such as the support vector machine (SVM),[h][205] Gaussian mixture model[206] and the extremely popular naive Bayes classifier.[i][208] Classifier performance depends greatly on the characteristics of the data to be classified, such as the dataset size, the dimensionality, and the level of noise. Model-based classifiers perform well if the assumed model is an extremely good fit for the actual data. Otherwise, if no matching model is available, and if accuracy (rather than speed or scalability) is the sole concern, conventional wisdom is that discriminative classifiers (especially SVM) tend to be more accurate than model-based classifiers such as “naive Bayes” on most practical data sets.[209]

Neural networks, or neural nets, were inspired by the architecture of neurons in the human brain. A simple “neuron” N accepts input from multiple other neurons, each of which, when activated (or “fired”), cast a weighted “vote” for or against whether neuron N should itself activate. Learning requires an algorithm to adjust these weights based on the training data; one simple algorithm (dubbed “fire together, wire together”) is to increase the weight between two connected neurons when the activation of one triggers the successful activation of another. The net forms “concepts” that are distributed among a subnetwork of shared[j] neurons that tend to fire together; a concept meaning “leg” might be coupled with a subnetwork meaning “foot” that includes the sound for “foot”. Neurons have a continuous spectrum of activation; in addition, neurons can process inputs in a nonlinear way rather than weighing straightforward votes. Modern neural nets can learn both continuous functions and, surprisingly, digital logical operations. Neural networks’ early successes included predicting the stock market and (in 1995) a mostly self-driving car.[k] In the 2010s, advances in neural networks using deep learning thrust AI into widespread public consciousness and contributed to an enormous upshift in corporate AI spending; for example, AI-related M&A in 2017 was over 25 times as large as in 2015.[212][213]

The study of non-learning artificial neural networks[201] began in the decade before the field of AI research was founded, in the work of Walter Pitts and Warren McCullouch. Frank Rosenblatt invented the perceptron, a learning network with a single layer, similar to the old concept of linear regression. Early pioneers also include Alexey Grigorevich Ivakhnenko, Teuvo Kohonen, Stephen Grossberg, Kunihiko Fukushima, Christoph von der Malsburg, David Willshaw, Shun-Ichi Amari, Bernard Widrow, John Hopfield, Eduardo R. Caianiello, and others.

The main categories of networks are acyclic or feedforward neural networks (where the signal passes in only one direction) and recurrent neural networks (which allow feedback and short-term memories of previous input events). Among the most popular feedforward networks are perceptrons, multi-layer perceptrons and radial basis networks.[214] Neural networks can be applied to the problem of intelligent control (for robotics) or learning, using such techniques as Hebbian learning (“fire together, wire together”), GMDH or competitive learning.[215]

Today, neural networks are often trained by the backpropagation algorithm, which had been around since 1970 as the reverse mode of automatic differentiation published by Seppo Linnainmaa,[216][217] and was introduced to neural networks by Paul Werbos.[218][219][220]

Hierarchical temporal memory is an approach that models some of the structural and algorithmic properties of the neocortex.[221]

In short, most neural networks use some form of gradient descent on a hand-created neural topology. However, some research groups, such as Uber, argue that simple neuroevolution to mutate new neural network topologies and weights may be competitive with sophisticated gradient descent approaches. One advantage of neuroevolution is that it may be less prone to get caught in “dead ends”.[222]

Deep learning is any artificial neural network that can learn a long chain of causal links. For example, a feedforward network with six hidden layers can learn a seven-link causal chain (six hidden layers + output layer) and has a “credit assignment path” (CAP) depth of seven. Many deep learning systems need to be able to learn chains ten or more causal links in length.[223] Deep learning has transformed many important subfields of artificial intelligence, including computer vision, speech recognition, natural language processing and others.[224][225][223]

According to one overview,[226] the expression “Deep Learning” was introduced to the Machine Learning community by Rina Dechter in 1986[227] and gained traction after Igor Aizenberg and colleagues introduced it to Artificial Neural Networks in 2000.[228] The first functional Deep Learning networks were published by Alexey Grigorevich Ivakhnenko and V. G. Lapa in 1965.[229][pageneeded] These networks are trained one layer at a time. Ivakhnenko’s 1971 paper[230] describes the learning of a deep feedforward multilayer perceptron with eight layers, already much deeper than many later networks. In 2006, a publication by Geoffrey Hinton and Ruslan Salakhutdinov introduced another way of pre-training many-layered feedforward neural networks (FNNs) one layer at a time, treating each layer in turn as an unsupervised restricted Boltzmann machine, then using supervised backpropagation for fine-tuning. Similar to shallow artificial neural networks, deep neural networks can model complex non-linear relationships. Over the last few years, advances in both machine learning algorithms and computer hardware have led to more efficient methods for training deep neural networks that contain many layers of non-linear hidden units and a very large output layer.[232]

Deep learning often uses convolutional neural networks (CNNs), whose origins can be traced back to the Neocognitron introduced by Kunihiko Fukushima in 1980.[233] In 1989, Yann LeCun and colleagues applied backpropagation to such an architecture. In the early 2000s, in an industrial application CNNs already processed an estimated 10% to 20% of all the checks written in the US.[234] Since 2011, fast implementations of CNNs on GPUs have won many visual pattern recognition competitions.[223]

CNNs with 12 convolutional layers were used in conjunction with reinforcement learning by Deepmind’s “AlphaGo Lee”, the program that beat a top Go champion in 2016.[235]

Early on, deep learning was also applied to sequence learning with recurrent neural networks (RNNs)[236] which are in theory Turing complete[237] and can run arbitrary programs to process arbitrary sequences of inputs. The depth of an RNN is unlimited and depends on the length of its input sequence; thus, an RNN is an example of deep learning.[223] RNNs can be trained by gradient descent[238][239][240] but suffer from the vanishing gradient problem.[224][241] In 1992, it was shown that unsupervised pre-training of a stack of recurrent neural networks can speed up subsequent supervised learning of deep sequential problems.[242]

Numerous researchers now use variants of a deep learning recurrent NN called the long short-term memory (LSTM) network published by Hochreiter & Schmidhuber in 1997.[243] LSTM is often trained by Connectionist Temporal Classification (CTC).[244] At Google, Microsoft and Baidu this approach has revolutionised speech recognition.[245][246][247] For example, in 2015, Google’s speech recognition experienced a dramatic performance jump of 49% through CTC-trained LSTM, which is now available through Google Voice to billions of smartphone users.[248] Google also used LSTM to improve machine translation,[249] Language Modeling[250] and Multilingual Language Processing.[251] LSTM combined with CNNs also improved automatic image captioning[252] and a plethora of other applications.

AI, like electricity or the steam engine, is a general purpose technology. There is no consensus on how to characterize which tasks AI tends to excel at.[253] While projects such as AlphaZero have succeeded in generating their own knowledge from scratch, many other machine learning projects require large training datasets.[254][255] Researcher Andrew Ng has suggested, as a “highly imperfect rule of thumb”, that “almost anything a typical human can do with less than one second of mental thought, we can probably now or in the near future automate using AI.”[256] Moravec’s paradox suggests that AI lags humans at many tasks that the human brain has specifically evolved to perform well.[120]

Games provide a well-publicized benchmark for assessing rates of progress. AlphaGo around 2016 brought the era of classical board-game benchmarks to an close. Games of imperfect knowledge provide new challenges to AI in the area of game theory.[257][258] E-sports such as StarCraft continue to provide additional public benchmarks.[259][260] There are many competitions and prizes, such as the Imagenet Challenge, to promote research in artificial intelligence. The main areas of competition include general machine intelligence, conversational behavior, data-mining, robotic cars, and robot soccer as well as conventional games.[citation needed]

The “imitation game” (an interpretation of the 1950 Turing test that assesses whether a computer can imitate a human) is nowadays considered too exploitable to be a meaningful benchmark.[261] A derivative of the Turing test is the Completely Automated Public Turing test to tell Computers and Humans Apart (CAPTCHA). As the name implies, this helps to determine that a user is an actual person and not a computer posing as a human. In contrast to the standard Turing test, CAPTCHA is administered by a machine and targeted to a human as opposed to being administered by a human and targeted to a machine. A computer asks a user to complete a simple test then generates a grade for that test. Computers are unable to solve the problem, so correct solutions are deemed to be the result of a person taking the test. A common type of CAPTCHA is the test that requires the typing of distorted letters, numbers or symbols that appear in an image undecipherable by a computer.

Proposed “universal intelligence” tests aim to compare how well machines, humans, and even non-human animals perform on problem sets that are generic as possible. At an extreme, the test suite can contain every possible problem, weighted by Kolmogorov complexity; unfortunately, these problem sets tend to be dominated by impoverished pattern-matching exercises where a tuned AI can easily exceed human performance levels.[263][264]

AI is relevant to any intellectual task. Modern artificial intelligence techniques are pervasive and are too numerous to list here. Frequently, when a technique reaches mainstream use, it is no longer considered artificial intelligence; this phenomenon is described as the AI effect.

High-profile examples of AI include autonomous vehicles (such as drones and self-driving cars), medical diagnosis, creating art (such as poetry), proving mathematical theorems, playing games (such as Chess or Go), search engines (such as Google search), online assistants (such as Siri), image recognition in photographs, spam filtering, prediction of judicial decisions[267] and targeting online advertisements.[268][269]

With social media sites overtaking TV as a source for news for young people and news organisations increasingly reliant on social media platforms for generating distribution,[270] major publishers now use artificial intelligence (AI) technology to post stories more effectively and generate higher volumes of traffic.[271]

Artificial intelligence is breaking into the healthcare industry by assisting doctors. According to Bloomberg Technology, Microsoft has developed AI to help doctors find the right treatments for cancer.[272] There is a great amount of research and drugs developed relating to cancer. In detail, there are more than 800 medicines and vaccines to treat cancer. This negatively affects the doctors, because there are too many options to choose from, making it more difficult to choose the right drugs for the patients. Microsoft is working on a project to develop a machine called “Hanover”. Its goal is to memorize all the papers necessary to cancer and help predict which combinations of drugs will be most effective for each patient. One project that is being worked on at the moment is fighting myeloid leukemia, a fatal cancer where the treatment has not improved in decades. Another study was reported to have found that artificial intelligence was as good as trained doctors in identifying skin cancers.[273] Another study is using artificial intelligence to try and monitor multiple high-risk patients, and this is done by asking each patient numerous questions based on data acquired from live doctor to patient interactions.[274]

According to CNN, a recent study by surgeons at the Children’s National Medical Center in Washington successfully demonstrated surgery with an autonomous robot. The team supervised the robot while it performed soft-tissue surgery, stitching together a pig’s bowel during open surgery, and doing so better than a human surgeon, the team claimed.[275] IBM has created its own artificial intelligence computer, the IBM Watson, which has beaten human intelligence (at some levels). Watson not only won at the game show Jeopardy! against former champions,[276] but was declared a hero after successfully diagnosing a woman who was suffering from leukemia.[277]

Advancements in AI have contributed to the growth of the automotive industry through the creation and evolution of self-driving vehicles. As of 2016, there are over 30 companies utilizing AI into the creation of driverless cars. A few companies involved with AI include Tesla, Google, and Apple.[278]

Many components contribute to the functioning of self-driving cars. These vehicles incorporate systems such as braking, lane changing, collision prevention, navigation and mapping. Together, these systems, as well as high performance computers, are integrated into one complex vehicle.[279]

Recent developments in autonomous automobiles have made the innovation of self-driving trucks possible, though they are still in the testing phase. The UK government has passed legislation to begin testing of self-driving truck platoons in 2018.[280] Self-driving truck platoons are a fleet of self-driving trucks following the lead of one non-self-driving truck, so the truck platoons aren’t entirely autonomous yet. Meanwhile, the Daimler, a German automobile corporation, is testing the Freightliner Inspiration which is a semi-autonomous truck that will only be used on the highway.[281]

One main factor that influences the ability for a driver-less automobile to function is mapping. In general, the vehicle would be pre-programmed with a map of the area being driven. This map would include data on the approximations of street light and curb heights in order for the vehicle to be aware of its surroundings. However, Google has been working on an algorithm with the purpose of eliminating the need for pre-programmed maps and instead, creating a device that would be able to adjust to a variety of new surroundings.[282] Some self-driving cars are not equipped with steering wheels or brake pedals, so there has also been research focused on creating an algorithm that is capable of maintaining a safe environment for the passengers in the vehicle through awareness of speed and driving conditions.[283]

Another factor that is influencing the ability for a driver-less automobile is the safety of the passenger. To make a driver-less automobile, engineers must program it to handle high risk situations. These situations could include a head on collision with pedestrians. The car’s main goal should be to make a decision that would avoid hitting the pedestrians and saving the passengers in the car. But there is a possibility the car would need to make a decision that would put someone in danger. In other words, the car would need to decide to save the pedestrians or the passengers.[284] The programing of the car in these situations is crucial to a successful driver-less automobile.

Financial institutions have long used artificial neural network systems to detect charges or claims outside of the norm, flagging these for human investigation. The use of AI in banking can be traced back to 1987 when Security Pacific National Bank in US set-up a Fraud Prevention Task force to counter the unauthorised use of debit cards. Programs like Kasisto and Moneystream are using AI in financial services.

Banks use artificial intelligence systems today to organize operations, maintain book-keeping, invest in stocks, and manage properties. AI can react to changes overnight or when business is not taking place.[285] In August 2001, robots beat humans in a simulated financial trading competition.[286] AI has also reduced fraud and financial crimes by monitoring behavioral patterns of users for any abnormal changes or anomalies.[287]

The use of AI machines in the market in applications such as online trading and decision making has changed major economic theories.[288] For example, AI based buying and selling platforms have changed the law of supply and demand in that it is now possible to easily estimate individualized demand and supply curves and thus individualized pricing. Furthermore, AI machines reduce information asymmetry in the market and thus making markets more efficient while reducing the volume of trades. Furthermore, AI in the markets limits the consequences of behavior in the markets again making markets more efficient. Other theories where AI has had impact include in rational choice, rational expectations, game theory, Lewis turning point, portfolio optimization and counterfactual thinking.

In video games, artificial intelligence is routinely used to generate dynamic purposeful behavior in non-player characters (NPCs). In addition, well-understood AI techniques are routinely used for pathfinding. Some researchers consider NPC AI in games to be a “solved problem” for most production tasks. Games with more atypical AI include the AI director of Left 4 Dead (2008) and the neuroevolutionary training of platoons in Supreme Commander 2 (2010).[289][290]

Worldwide annual military spending on robotics rose from 5.1 billion USD in 2010 to 7.5 billion USD in 2015.[291][292] Military drones capable of autonomous action are widely considered a useful asset. In 2017, Vladimir Putin stated that “Whoever becomes the leader in (artificial intelligence) will become the ruler of the world”.[293][294] Many artificial intelligence researchers seek to distance themselves from military applications of AI.[295]

For financial statements audit, AI makes continuous audit possible. AI tools could analyze many sets of different information immediately. The potential benefit would be the overall audit risk will be reduced, the level of assurance will be increased and the time duration of audit will be reduced.[296]

A report by the Guardian newspaper in the UK in 2018 found that online gambling companies were using AI to predict the behavior of customers in order to target them with personalized promotions.[297]

There are three philosophical questions related to AI:

Can a machine be intelligent? Can it “think”?

Widespread use of artificial intelligence could have unintended consequences that are dangerous or undesirable. Scientists from the Future of Life Institute, among others, described some short-term research goals to see how AI influences the economy, the laws and ethics that are involved with AI and how to minimize AI security risks. In the long-term, the scientists have proposed to continue optimizing function while minimizing possible security risks that come along with new technologies.[308]

Machines with intelligence have the potential to use their intelligence to make ethical decisions. Research in this area includes “machine ethics”, “artificial moral agents”, and the study of “malevolent vs. friendly AI”.

See the original post:

Artificial intelligence – Wikipedia

A.I. Artificial Intelligence (2001) – IMDb

Edit Storyline

In the not-so-far future the polar ice caps have melted and the resulting rise of the ocean waters has drowned all the coastal cities of the world. Withdrawn to the interior of the continents, the human race keeps advancing, reaching the point of creating realistic robots (called mechas) to serve them. One of the mecha-producing companies builds David, an artificial kid which is the first to have real feelings, especially a never-ending love for his “mother”, Monica. Monica is the woman who adopted him as a substitute for her real son, who remains in cryo-stasis, stricken by an incurable disease. David is living happily with Monica and her husband, but when their real son returns home after a cure is discovered, his life changes dramatically. Written byChris Makrozahopoulos

Budget:$100,000,000 (estimated)

Opening Weekend USA: $29,352,630,1 July 2001, Wide Release

Gross USA: $78,616,689, 23 September 2001

Cumulative Worldwide Gross: $235,927,000

Runtime: 146 min

Aspect Ratio: 1.85 : 1

View original post here:

A.I. Artificial Intelligence (2001) – IMDb

Online Artificial Intelligence Courses | Microsoft …

The Microsoft Professional Program (MPP) is a collection of courses that teach skills in several core technology tracks that help you excel in the industry’s newest job roles.

These courses are created and taught by experts and feature quizzes, hands-on labs, and engaging communities. For each track you complete, you earn a certificate of completion from Microsoft proving that you mastered those skills.

See the original post:

Online Artificial Intelligence Courses | Microsoft …

What is artificial intelligence? – BBC News

A computer can beat the world chess champion and understand voice commands on your smartphone, but real artificial intelligence has yet to arrive. The pace of change is quickening, though.

Some people say it will save humanity, even make us immortal.

Others say it could destroy us all.

But, the truth is, most of us don’t really know what AI is.

Video production by Valery Eremenko

Intelligent Machines – a BBC News series looking at AI and robotics

See more here:

What is artificial intelligence? – BBC News

Artificial intelligence – Wikipedia

Artificial intelligence (AI, also machine intelligence, MI) is intelligence demonstrated by machines, in contrast to the natural intelligence (NI) displayed by humans and other animals. In computer science AI research is defined as the study of “intelligent agents”: any device that perceives its environment and takes actions that maximize its chance of successfully achieving its goals.[1] Colloquially, the term “artificial intelligence” is applied when a machine mimics “cognitive” functions that humans associate with other human minds, such as “learning” and “problem solving”.

The scope of AI is disputed: as machines become increasingly capable, tasks considered as requiring “intelligence” are often removed from the definition, a phenomenon known as the AI effect, leading to the quip, “AI is whatever hasn’t been done yet.”[3] For instance, optical character recognition is frequently excluded from “artificial intelligence”, having become a routine technology.[4] Capabilities generally classified as AI as of 2017[update] include successfully understanding human speech, competing at the highest level in strategic game systems (such as chess and Go[6]), autonomous cars, intelligent routing in content delivery network and military simulations.

Artificial intelligence was founded as an academic discipline in 1956, and in the years since has experienced several waves of optimism,[7][8] followed by disappointment and the loss of funding (known as an “AI winter”),[9][10] followed by new approaches, success and renewed funding.[8][11] For most of its history, AI research has been divided into subfields that often fail to communicate with each other.[12] These sub-fields are based on technical considerations, such as particular goals (e.g. “robotics” or “machine learning”),[13] the use of particular tools (“logic” or artificial neural networks), or deep philosophical differences.[14][15][16] Subfields have also been based on social factors (particular institutions or the work of particular researchers).[12]

The traditional problems (or goals) of AI research include reasoning, knowledge representation, planning, learning, natural language processing, perception and the ability to move and manipulate objects.[13] General intelligence is among the field’s long-term goals.[17] Approaches include statistical methods, computational intelligence, and traditional symbolic AI. Many tools are used in AI, including versions of search and mathematical optimization, artificial neural networks, and methods based on statistics, probability and economics. The AI field draws upon computer science, mathematics, psychology, linguistics, philosophy and many others.

The field was founded on the claim that human intelligence “can be so precisely described that a machine can be made to simulate it”.[18] This raises philosophical arguments about the nature of the mind and the ethics of creating artificial beings endowed with human-like intelligence which are issues that have been explored by myth, fiction and philosophy since antiquity.[19] Some people also consider AI to be a danger to humanity if it progresses unabatedly.[20] Others believe that AI, unlike previous technological revolutions, will create a risk of mass unemployment.[21]

In the twenty-first century, AI techniques have experienced a resurgence following concurrent advances in computer power, large amounts of data, and theoretical understanding; and AI techniques have become an essential part of the technology industry, helping to solve many challenging problems in computer science.[22][11]

Thought-capable artificial beings appeared as storytelling devices in antiquity,[23] and have been common in fiction, as in Mary Shelley’s Frankenstein or Karel apek’s R.U.R. (Rossum’s Universal Robots).[24] These characters and their fates raised many of the same issues now discussed in the ethics of artificial intelligence.[19]

The study of mechanical or “formal” reasoning began with philosophers and mathematicians in antiquity. The study of mathematical logic led directly to Alan Turing’s theory of computation, which suggested that a machine, by shuffling symbols as simple as “0” and “1”, could simulate any conceivable act of mathematical deduction. This insight, that digital computers can simulate any process of formal reasoning, is known as the ChurchTuring thesis.[25] Along with concurrent discoveries in neurobiology, information theory and cybernetics, this led researchers to consider the possibility of building an electronic brain. Turing proposed that “if a human could not distinguish between responses from a machine and a human, the machine could be considered intelligent”.[26] The first work that is now generally recognized as AI was McCullouch and Pitts’ 1943 formal design for Turing-complete “artificial neurons”.

The field of AI research was born at a workshop at Dartmouth College in 1956.[28] Attendees Allen Newell (CMU), Herbert Simon (CMU), John McCarthy (MIT), Marvin Minsky (MIT) and Arthur Samuel (IBM) became the founders and leaders of AI research.[29] They and their students produced programs that the press described as “astonishing”: computers were learning checkers strategies (c. 1954)[31] (and by 1959 were reportedly playing better than the average human),[32] solving word problems in algebra, proving logical theorems (Logic Theorist, first run c. 1956) and speaking English.[33] By the middle of the 1960s, research in the U.S. was heavily funded by the Department of Defense[34] and laboratories had been established around the world.[35] AI’s founders were optimistic about the future: Herbert Simon predicted, “machines will be capable, within twenty years, of doing any work a man can do”. Marvin Minsky agreed, writing, “within a generation… the problem of creating ‘artificial intelligence’ will substantially be solved”.[7]

They failed to recognize the difficulty of some of the remaining tasks. Progress slowed and in 1974, in response to the criticism of Sir James Lighthill and ongoing pressure from the US Congress to fund more productive projects, both the U.S. and British governments cut off exploratory research in AI. The next few years would later be called an “AI winter”,[9] a period when obtaining funding for AI projects was difficult.

In the early 1980s, AI research was revived by the commercial success of expert systems,[37] a form of AI program that simulated the knowledge and analytical skills of human experts. By 1985 the market for AI had reached over a billion dollars. At the same time, Japan’s fifth generation computer project inspired the U.S and British governments to restore funding for academic research.[8] However, beginning with the collapse of the Lisp Machine market in 1987, AI once again fell into disrepute, and a second, longer-lasting hiatus began.[10]

In the late 1990s and early 21st century, AI began to be used for logistics, data mining, medical diagnosis and other areas.[22] The success was due to increasing computational power (see Moore’s law), greater emphasis on solving specific problems, new ties between AI and other fields (such as statistics, economics and mathematics), and a commitment by researchers to mathematical methods and scientific standards.[38] Deep Blue became the first computer chess-playing system to beat a reigning world chess champion, Garry Kasparov on 11 May 1997.

In 2011, a Jeopardy! quiz show exhibition match, IBM’s question answering system, Watson, defeated the two greatest Jeopardy champions, Brad Rutter and Ken Jennings, by a significant margin. Faster computers, algorithmic improvements, and access to large amounts of data enabled advances in machine learning and perception; data-hungry deep learning methods started to dominate accuracy benchmarks around 2012.[41] The Kinect, which provides a 3D bodymotion interface for the Xbox 360 and the Xbox One use algorithms that emerged from lengthy AI research[42] as do intelligent personal assistants in smartphones.[43] In March 2016, AlphaGo won 4 out of 5 games of Go in a match with Go champion Lee Sedol, becoming the first computer Go-playing system to beat a professional Go player without handicaps.[6][44] In the 2017 Future of Go Summit, AlphaGo won a three-game match with Ke Jie,[45] who at the time continuously held the world No. 1 ranking for two years.[46][47] This marked the completion of a significant milestone in the development of Artificial Intelligence as Go is an extremely complex game, more so than Chess.

According to Bloomberg’s Jack Clark, 2015 was a landmark year for artificial intelligence, with the number of software projects that use AI within Google increased from a “sporadic usage” in 2012 to more than 2,700 projects. Clark also presents factual data indicating that error rates in image processing tasks have fallen significantly since 2011.[48] He attributes this to an increase in affordable neural networks, due to a rise in cloud computing infrastructure and to an increase in research tools and datasets.[11] Other cited examples include Microsoft’s development of a Skype system that can automatically translate from one language to another and Facebook’s system that can describe images to blind people.[48] In a 2017 survey, one in five companies reported they had “incorporated AI in some offerings or processes”.[49][50]

A typical AI perceives its environment and takes actions that maximize its chance of successfully achieving its goals.[1] An AI’s intended goal function can be simple (“1 if the AI wins a game of Go, 0 otherwise”) or complex (“Do actions mathematically similar to the actions that got you rewards in the past”). Goals can be explicitly defined, or can be induced. If the AI is programmed for “reinforcement learning”, goals can be implicitly induced by rewarding some types of behavior and punishing others.[a] Alternatively, an evolutionary system can induce goals by using a “fitness function” to mutate and preferentially replicate high-scoring AI systems; this is similar to how animals evolved to innately desire certain goals such as finding food, or how dogs can be bred via artificial selection to possess desired traits. Some AI systems, such as nearest-neighbor, instead reason by analogy; these systems are not generally given goals, except to the degree that goals are somehow implicit in their training data. Such systems can still be benchmarked if the non-goal system is framed as a system whose “goal” is to successfully accomplish its narrow classification task.[53]

AI often revolves around the use of algorithms. An algorithm is a set of unambiguous instructions that a mechanical computer can execute.[b] A complex algorithm is often built on top of other, simpler, algorithms. A simple example of an algorithm is the following recipe for optimal play at tic-tac-toe:

Many AI algorithms are capable of learning from data; they can enhance themselves by learning new heuristics (strategies, or “rules of thumb”, that have worked well in the past), or can themselves write other algorithms. Some of the “learners” described below, including Bayesian networks, decision trees, and nearest-neighbor, could theoretically, if given infinite data, time, and memory, learn to approximate any function, including whatever combination of mathematical functions would best describe the entire world. These learners could therefore, in theory, derive all possible knowledge, by considering every possible hypothesis and matching it against the data. In practice, it is almost never possible to consider every possibility, because of the phenomenon of “combinatorial explosion”, where the amount of time needed to solve a problem grows exponentially. Much of AI research involves figuring out how to identify and avoid considering broad swaths of possibililities that are unlikely to be fruitful.[55] For example, when viewing a map and looking for the shortest driving route from Denver to New York in the East, one can in most cases skip looking at any path through San Francisco or other areas far to the West; thus, an AI wielding an pathfinding algorithm like A* can avoid the combinatorial explosion that would ensue if every possible route had to be ponderously considered in turn.[57]

The earliest (and easiest to understand) approach to AI was symbolism (such as formal logic): “If an otherwise healthy adult has a fever, then they may have influenza”. A second, more general, approach is Bayesian inference: “If the current patient has a fever, adjust the probability they have influenza in such-and-such way”. The third major approach, extremely popular in routine business AI applications, are analogizers such as SVM and nearest-neighbor: “After examining the records of known past patients whose temperature, symptoms, age, and other factors mostly match the current patient, X% of those patients turned out to have influenza”. A fourth approach is harder to intuitively understand, but is inspired by how the brain’s machinery works: the artificial neural network approach uses artificial “neurons” that can learn by comparing itself to the desired output and altering the strengths of the connections between its internal neurons to “reinforce” connections that seemed to be useful. These four main approaches can overlap with each other and with evolutionary systems; for example, neural nets can learn to make inferences, to generalize, and to make analogies. Some systems implicitly or explicitly use multiple of these approaches, alongside many other AI and non-AI algorithms; the best approach is often different depending on the problem.[59]

Learning algorithms work on the basis that strategies, algorithms, and inferences that worked well in the past are likely to continue working well in the future. These inferences can be obvious, such as “since the sun rose every morning for the last 10,000 days, it will probably rise tomorrow morning as well”. They can be nuanced, such as “X% of families have geographically separate species with color variants, so there is an Y% chance that undiscovered black swans exist”. Learners also work on the basis of “Occam’s razor”: The simplest theory that explains the data is the likeliest. Therefore, to be successful, a learner must be designed such that it prefers simpler theories to complex theories, except in cases where the complex theory is proven substantially better. Settling on a bad, overly complex theory gerrymandered to fit all the past training data is known as overfitting. Many systems attempt to reduce overfitting by rewarding a theory in accordance with how well it fits the data, but penalizing the theory in accordance with how complex the theory is. Besides classic overfitting, learners can also disappoint by “learning the wrong lesson”. A toy example is that an image classifier trained only on pictures of brown horses and black cats might conclude that all brown patches are likely to be horses. A real-world example is that, unlike humans, current image classifiers don’t determine the spatial relationship between components of the picture; instead, they learn abstract patterns of pixels that humans are oblivious to, but that linearly correlate with images of certain types of real objects. Faintly superimposing such a pattern on a legitimate image results in an “adversarial” image that the system misclassifies.[c][62][63][64]

Compared with humans, existing AI lacks several features of human “commonsense reasoning”; most notably, humans have powerful mechanisms for reasoning about “nave physics” such as space, time, and physical interactions. This enables even young children to easily make inferences like “If I roll this pen off a table, it will fall on the floor”. Humans also have a powerful mechanism of “folk psychology” that helps them to interpret natural-language sentences such as “The city councilmen refused the demonstrators a permit because they advocated violence”. (A generic AI has difficulty inferring whether the councilmen or the demonstrators are the ones alleged to be advocating violence.)[67][68][69] This lack of “common knowledge” means that AI often makes different mistakes than humans make, in ways that can seem incomprehensible. For example, existing self-driving cars cannot reason about the location nor the intentions of pedestrians in the exact way that humans do, and instead must use non-human modes of reasoning to avoid accidents.[70][71][72]

The overall research goal of artificial intelligence is to create technology that allows computers and machines to function in an intelligent manner. The general problem of simulating (or creating) intelligence has been broken down into sub-problems. These consist of particular traits or capabilities that researchers expect an intelligent system to display. The traits described below have received the most attention.[13]

Early researchers developed algorithms that imitated step-by-step reasoning that humans use when they solve puzzles or make logical deductions.[73] By the late 1980s and 1990s, AI research had developed methods for dealing with uncertain or incomplete information, employing concepts from probability and economics.[74]

These algorithms proved to be insufficient for solving large reasoning problems, because they experienced a “combinatorial explosion”: they became exponentially slower as the problems grew larger.[55] In fact, even humans rarely use the step-by-step deduction that early AI research was able to model. They solve most of their problems using fast, intuitive judgements.[75]

Knowledge representation[76] and knowledge engineering[77] are central to classical AI research. Some “expert systems” attempt to gather together explicit knowledge possessed by experts in some narrow domain. In addition, some projects attempt to gather the “commonsense knowledge” known to the average person into a database containing extensive knowledge about the world. Among the things a comprehensive commonsense knowledge base would contain are: objects, properties, categories and relations between objects;[78] situations, events, states and time;[79] causes and effects;[80] knowledge about knowledge (what we know about what other people know);[81] and many other, less well researched domains. A representation of “what exists” is an ontology: the set of objects, relations, concepts, and properties formally described so that software agents can interpret them. The semantics of these are captured as description logic concepts, roles, and individuals, and typically implemented as classes, properties, and individuals in the Web Ontology Language.[82] The most general ontologies are called upper ontologies, which attempt to provide a foundation for all other knowledge[83] by acting as mediators between domain ontologies that cover specific knowledge about a particular knowledge domain (field of interest or area of concern). Such formal knowledge representations can be used in content-based indexing and retrieval,[84] scene interpretation,[85] clinical decision support,[86] knowledge discovery (mining “interesting” and actionable inferences from large databases),[87] and other areas.[88]

Among the most difficult problems in knowledge representation are:

Intelligent agents must be able to set goals and achieve them.[95] They need a way to visualize the futurea representation of the state of the world and be able to make predictions about how their actions will change itand be able to make choices that maximize the utility (or “value”) of available choices.[96]

In classical planning problems, the agent can assume that it is the only system acting in the world, allowing the agent to be certain of the consequences of its actions.[97] However, if the agent is not the only actor, then it requires that the agent can reason under uncertainty. This calls for an agent that can not only assess its environment and make predictions, but also evaluate its predictions and adapt based on its assessment.[98]

Multi-agent planning uses the cooperation and competition of many agents to achieve a given goal. Emergent behavior such as this is used by evolutionary algorithms and swarm intelligence.[99]

Machine learning, a fundamental concept of AI research since the field’s inception,[100] is the study of computer algorithms that improve automatically through experience.[101][102]

Unsupervised learning is the ability to find patterns in a stream of input. Supervised learning includes both classification and numerical regression. Classification is used to determine what category something belongs in, after seeing a number of examples of things from several categories. Regression is the attempt to produce a function that describes the relationship between inputs and outputs and predicts how the outputs should change as the inputs change.[102] Both classifiers and regression learners can be viewed as “function approximators” trying to learn an unknown (possibly implicit) function; for example, a spam classifier can be viewed as learning a function that maps from the text of an email to one of two categories, “spam” or “not spam”. Computational learning theory can assess learners by computational complexity, by sample complexity (how much data is required), or by other notions of optimization.[103] In reinforcement learning[104] the agent is rewarded for good responses and punished for bad ones. The agent uses this sequence of rewards and punishments to form a strategy for operating in its problem space.

Natural language processing[105] (NLP) gives machines the ability to read and understand human language. A sufficiently powerful natural language processing system would enable natural-language user interfaces and the acquisition of knowledge directly from human-written sources, such as newswire texts. Some straightforward applications of natural language processing include information retrieval, text mining, question answering[106] and machine translation.[107] Many current approaches use word co-occurrence frequencies to construct syntactic representations of text. “Keyword spotting” strategies for search are popular and scalable but dumb; a search query for “dog” might only match documents with the literal word “dog” and miss a document with the word “poodle”. “Lexical affinity” strategies use the occurrence of words such as “accident” to assess the sentiment of a document. Modern statistical NLP approaches can combine all these strategies as well as others, and often achieve acceptable accuracy at the page or paragraph level, but continue to lack the semantic understanding required to classify isolated sentences well. Besides the usual difficulties with encoding semantic commonsense knowledge, existing semantic NLP sometimes scales too poorly to be viable in business applications. Beyond semantic NLP, the ultimate goal of “narrative” NLP is to embody a full understanding of commonsense reasoning.[108]

Machine perception[109] is the ability to use input from sensors (such as cameras (visible spectrum or infrared), microphones, wireless signals, and active lidar, sonar, radar, and tactile sensors) to deduce aspects of the world. Applications include speech recognition,[110] facial recognition, and object recognition.[111] Computer vision is the ability to analyze visual input. Such input is usually ambiguous; a giant, fifty-meter-tall pedestrian far away may produce exactly the same pixels as a nearby normal-sized pedestrian, requiring the AI to judge the relative likelihood and reasonableness of different interpretations, for example by using its “object model” to assess that fifty-meter pedestrians do not exist.[112]

AI is heavily used in robotics.[113] Advanced robotic arms and other industrial robots, widely used in modern factories, can learn from experience how to move efficiently despite the presence of friction and gear slippage.[114] A modern mobile robot, when given a small, static, and visible environment, can easily determine its location and map its environment; however, dynamic environments, such as (in endoscopy) the interior of a patient’s breathing body, pose a greater challenge. Motion planning is the process of breaking down a movement task into “primitives” such as individual joint movements. Such movement often involves compliant motion, a process where movement requires maintaining physical contact with an object.[116][117] Moravec’s paradox generalizes that low-level sensorimotor skills that humans take for granted are, counterintuitively, difficult to program into a robot; the paradox is named after Hans Moravec, who stated in 1988 that “it is comparatively easy to make computers exhibit adult level performance on intelligence tests or playing checkers, and difficult or impossible to give them the skills of a one-year-old when it comes to perception and mobility”.[118][119] This is attributed to the fact that, unlike checkers, physical dexterity has been a direct target of natural selection for millions of years.[120]

Moravec’s paradox can be extended to many forms of social intelligence.[122][123] Distributed multi-agent coordination of autonomous vehicles remains a difficult problem.[124] Affective computing is an interdisciplinary umbrella that comprises systems which recognize, interpret, process, or simulate human affects. Moderate successes related to affective computing include textual sentiment analysis and, more recently, multimodal affect analysis, wherein AI classifies the affects displayed by a videotaped subject.[128]

In the long run, social skills and an understanding of human emotion and game theory would be valuable to a social agent. Being able to predict the actions of others by understanding their motives and emotional states would allow an agent to make better decisions. Some computer systems mimic human emotion and expressions to appear more sensitive to the emotional dynamics of human interaction, or to otherwise facilitate humancomputer interaction.[129] Similarly, some virtual assistants are programmed to speak conversationally or even to banter humorously; this tends to give naive users an unrealistic conception of how intelligent existing computer agents actually are.[130]

Historically, projects such as the Cyc knowledge base (1984) and the massive Japanese Fifth Generation Computer Systems initiative (19821992) attempted to cover the breadth of human cognition. These early projects failed to escape the limitations of non-quantitative symbolic logic models and, in retrospect, greatly underestimated the difficulty of cross-domain AI. Nowadays, the vast majority of current AI researchers work instead on tractable “narrow AI” applications (such as medical diagnosis or automobile navigation).[131] Many researchers predict that such “narrow AI” work in different individual domains will eventually be incorporated into a machine with artificial general intelligence (AGI), combining most of the narrow skills mentioned in this article and at some point even exceeding human ability in most or all these areas.[17][132] Many advances have general, cross-domain significance. One high-profile example is that DeepMind in the 2010s developed a “generalized artificial intelligence” that could learn many diverse Atari games on its own, and later developed a variant of the system which succeeds at sequential learning.[133][134][135] Besides transfer learning,[136] hypothetical AGI breakthroughs could include the development of reflective architectures that can engage in decision-theoretic metareasoning, and figuring out how to “slurp up” a comprehensive knowledge base from the entire unstructured Web. Some argue that some kind of (currently-undiscovered) conceptually straightforward, but mathematically difficult, “Master Algorithm” could lead to AGI. Finally, a few “emergent” approaches look to simulating human intelligence extremely closely, and believe that anthropomorphic features like an artificial brain or simulated child development may someday reach a critical point where general intelligence emerges.[138][139]

Many of the problems in this article may also require general intelligence, if machines are to solve the problems as well as people do. For example, even specific straightforward tasks, like machine translation, require that a machine read and write in both languages (NLP), follow the author’s argument (reason), know what is being talked about (knowledge), and faithfully reproduce the author’s original intent (social intelligence). A problem like machine translation is considered “AI-complete”, because all of these problems need to be solved simultaneously in order to reach human-level machine performance.

There is no established unifying theory or paradigm that guides AI research. Researchers disagree about many issues.[140] A few of the most long standing questions that have remained unanswered are these: should artificial intelligence simulate natural intelligence by studying psychology or neurobiology? Or is human biology as irrelevant to AI research as bird biology is to aeronautical engineering?[14] Can intelligent behavior be described using simple, elegant principles (such as logic or optimization)? Or does it necessarily require solving a large number of completely unrelated problems?[15]

In the 1940s and 1950s, a number of researchers explored the connection between neurobiology, information theory, and cybernetics. Some of them built machines that used electronic networks to exhibit rudimentary intelligence, such as W. Grey Walter’s turtles and the Johns Hopkins Beast. Many of these researchers gathered for meetings of the Teleological Society at Princeton University and the Ratio Club in England.[141] By 1960, this approach was largely abandoned, although elements of it would be revived in the 1980s.

When access to digital computers became possible in the middle 1950s, AI research began to explore the possibility that human intelligence could be reduced to symbol manipulation. The research was centered in three institutions: Carnegie Mellon University, Stanford and MIT, and as described below, each one developed its own style of research. John Haugeland named these symbolic approaches to AI “good old fashioned AI” or “GOFAI”.[142] During the 1960s, symbolic approaches had achieved great success at simulating high-level thinking in small demonstration programs. Approaches based on cybernetics or artificial neural networks were abandoned or pushed into the background.[143] Researchers in the 1960s and the 1970s were convinced that symbolic approaches would eventually succeed in creating a machine with artificial general intelligence and considered this the goal of their field.

Economist Herbert Simon and Allen Newell studied human problem-solving skills and attempted to formalize them, and their work laid the foundations of the field of artificial intelligence, as well as cognitive science, operations research and management science. Their research team used the results of psychological experiments to develop programs that simulated the techniques that people used to solve problems. This tradition, centered at Carnegie Mellon University would eventually culminate in the development of the Soar architecture in the middle 1980s.[144][145]

Unlike Simon and Newell, John McCarthy felt that machines did not need to simulate human thought, but should instead try to find the essence of abstract reasoning and problem solving, regardless of whether people used the same algorithms.[14] His laboratory at Stanford (SAIL) focused on using formal logic to solve a wide variety of problems, including knowledge representation, planning and learning.[146] Logic was also the focus of the work at the University of Edinburgh and elsewhere in Europe which led to the development of the programming language Prolog and the science of logic programming.[147]

Researchers at MIT (such as Marvin Minsky and Seymour Papert)[148] found that solving difficult problems in vision and natural language processing required ad-hoc solutions they argued that there was no simple and general principle (like logic) that would capture all the aspects of intelligent behavior. Roger Schank described their “anti-logic” approaches as “scruffy” (as opposed to the “neat” paradigms at CMU and Stanford).[15] Commonsense knowledge bases (such as Doug Lenat’s Cyc) are an example of “scruffy” AI, since they must be built by hand, one complicated concept at a time.[149]

When computers with large memories became available around 1970, researchers from all three traditions began to build knowledge into AI applications.[150] This “knowledge revolution” led to the development and deployment of expert systems (introduced by Edward Feigenbaum), the first truly successful form of AI software.[37] The knowledge revolution was also driven by the realization that enormous amounts of knowledge would be required by many simple AI applications.

By the 1980s progress in symbolic AI seemed to stall and many believed that symbolic systems would never be able to imitate all the processes of human cognition, especially perception, robotics, learning and pattern recognition. A number of researchers began to look into “sub-symbolic” approaches to specific AI problems.[16] Sub-symbolic methods manage to approach intelligence without specific representations of knowledge.

This includes embodied, situated, behavior-based, and nouvelle AI. Researchers from the related field of robotics, such as Rodney Brooks, rejected symbolic AI and focused on the basic engineering problems that would allow robots to move and survive.[151] Their work revived the non-symbolic viewpoint of the early cybernetics researchers of the 1950s and reintroduced the use of control theory in AI. This coincided with the development of the embodied mind thesis in the related field of cognitive science: the idea that aspects of the body (such as movement, perception and visualization) are required for higher intelligence.

Within developmental robotics, developmental learning approaches are elaborated upon to allow robots to accumulate repertoires of novel skills through autonomous self-exploration, social interaction with human teachers, and the use of guidance mechanisms (active learning, maturation, motor synergies, etc.).[152][153]

Interest in neural networks and “connectionism” was revived by David Rumelhart and others in the middle of the 1980s.[156] Artificial neural networks are an example of soft computing — they are solutions to problems which cannot be solved with complete logical certainty, and where an approximate solution is often sufficient. Other soft computing approaches to AI include fuzzy systems, evolutionary computation and many statistical tools. The application of soft computing to AI is studied collectively by the emerging discipline of computational intelligence.[157]

Much of GOFAI got bogged down on ad hoc patches to symbolic computation that worked on their own toy models but failed to generalize to real-world results. However, around the 1990s, AI researchers adopted sophisticated mathematical tools, such as hidden Markov models (HMM), information theory, and normative Bayesian decision theory to compare or to unify competing architectures. The shared mathematical language permitted a high level of collaboration with more established fields (like mathematics, economics or operations research).[d] Compared with GOFAI, new “statistical learning” techniques such as HMM and neural networks were gaining higher levels of accuracy in many practical domains such as data mining, without necessarily acquiring semantic understanding of the datasets. The increased successes with real-world data led to increasing emphasis on comparing different approaches against shared test data to see which approach performed best in a broader context than that provided by idiosyncratic toy models; AI research was becoming more scientific. Nowadays results of experiments are often rigorously measurable, and are sometimes (with difficulty) reproducible.[38][158] Different statistical learning techniques have different limitations; for example, basic HMM cannot model the infinite possible combinations of natural language. Critics note that the shift from GOFAI to statistical learning is often also a shift away from Explainable AI. In AGI research, some scholars caution against over-reliance on statistical learning, and argue that continuing research into GOFAI will still be necessary to attain general intelligence.

AI has developed a large number of tools to solve the most difficult problems in computer science. A few of the most general of these methods are discussed below.

Many problems in AI can be solved in theory by intelligently searching through many possible solutions:[167] Reasoning can be reduced to performing a search. For example, logical proof can be viewed as searching for a path that leads from premises to conclusions, where each step is the application of an inference rule.[168] Planning algorithms search through trees of goals and subgoals, attempting to find a path to a target goal, a process called means-ends analysis.[169] Robotics algorithms for moving limbs and grasping objects use local searches in configuration space.[114] Many learning algorithms use search algorithms based on optimization.

Simple exhaustive searches[170] are rarely sufficient for most real world problems: the search space (the number of places to search) quickly grows to astronomical numbers. The result is a search that is too slow or never completes. The solution, for many problems, is to use “heuristics” or “rules of thumb” that prioritize choices in favor of those that are more likely to reach a goal, and to do so in a shorter number of steps. In some search methodologies heuristics can also serve to entirely eliminate some choices that are unlikely to lead to a goal (called “pruning the search tree”). Heuristics supply the program with a “best guess” for the path on which the solution lies.[171] Heuristics limit the search for solutions into a smaller sample size.

A very different kind of search came to prominence in the 1990s, based on the mathematical theory of optimization. For many problems, it is possible to begin the search with some form of a guess and then refine the guess incrementally until no more refinements can be made. These algorithms can be visualized as blind hill climbing: we begin the search at a random point on the landscape, and then, by jumps or steps, we keep moving our guess uphill, until we reach the top. Other optimization algorithms are simulated annealing, beam search and random optimization.[172]

Evolutionary computation uses a form of optimization search. For example, they may begin with a population of organisms (the guesses) and then allow them to mutate and recombine, selecting only the fittest to survive each generation (refining the guesses). Classic evolutionary algorithms include genetic algorithms, gene expression programming, and genetic programming.[173] Alternatively, distributed search processes can coordinate via swarm intelligence algorithms. Two popular swarm algorithms used in search are particle swarm optimization (inspired by bird flocking) and ant colony optimization (inspired by ant trails).[174][175]

Logic[176] is used for knowledge representation and problem solving, but it can be applied to other problems as well. For example, the satplan algorithm uses logic for planning[177] and inductive logic programming is a method for learning.[178]

Several different forms of logic are used in AI research. Propositional logic[179] involves truth functions such as “or” and “not”. First-order logic[180] adds quantifiers and predicates, and can express facts about objects, their properties, and their relations with each other. Fuzzy set theory assigns a “degree of truth” (between 0 and 1) to vague statements such as “Alice is old” (or rich, or tall, or hungry) that are too linguistically imprecise to be completely true or false. Fuzzy logic is successfully used in control systems to allow experts to contribute vague rules such as “if you are close to the destination station and moving fast, increase the train’s brake pressure”; these vague rules can then be numerically refined within the system. Fuzzy logic fails to scale well in knowledge bases; many AI researchers question the validity of chaining fuzzy-logic inferences.[e][182][183]

Default logics, non-monotonic logics and circumscription[90] are forms of logic designed to help with default reasoning and the qualification problem. Several extensions of logic have been designed to handle specific domains of knowledge, such as: description logics;[78] situation calculus, event calculus and fluent calculus (for representing events and time);[79] causal calculus;[80] belief calculus;[184] and modal logics.[81]

Overall, qualitiative symbolic logic is brittle and scales poorly in the presence of noise or other uncertainty. Exceptions to rules are numerous, and it is difficult for logical systems to function in the presence of contradictory rules.[186]

Many problems in AI (in reasoning, planning, learning, perception and robotics) require the agent to operate with incomplete or uncertain information. AI researchers have devised a number of powerful tools to solve these problems using methods from probability theory and economics.[187]

Bayesian networks[188] are a very general tool that can be used for a large number of problems: reasoning (using the Bayesian inference algorithm),[189] learning (using the expectation-maximization algorithm),[f][191] planning (using decision networks)[192] and perception (using dynamic Bayesian networks).[193] Probabilistic algorithms can also be used for filtering, prediction, smoothing and finding explanations for streams of data, helping perception systems to analyze processes that occur over time (e.g., hidden Markov models or Kalman filters).[193] Compared with symbolic logic, formal Bayesian inference is computationally expensive. For inference to be tractable, most observations must be conditionally independent of one another. Complicated graphs with diamonds or other “loops” (undirected cycles) can require a sophisticated method such as Markov Chain Monte Carlo, which spreads an ensemble of random walkers throughout the Bayesian network and attempts to converge to an assessment of the conditional probabilities. Bayesian networks are used on XBox Live to rate and match players; wins and losses are “evidence” of how good a player is. AdSense uses a Bayesian network with over 300 million edges to learn which ads to serve.

A key concept from the science of economics is “utility”: a measure of how valuable something is to an intelligent agent. Precise mathematical tools have been developed that analyze how an agent can make choices and plan, using decision theory, decision analysis,[195] and information value theory.[96] These tools include models such as Markov decision processes,[196] dynamic decision networks,[193] game theory and mechanism design.[197]

The simplest AI applications can be divided into two types: classifiers (“if shiny then diamond”) and controllers (“if shiny then pick up”). Controllers do, however, also classify conditions before inferring actions, and therefore classification forms a central part of many AI systems. Classifiers are functions that use pattern matching to determine a closest match. They can be tuned according to examples, making them very attractive for use in AI. These examples are known as observations or patterns. In supervised learning, each pattern belongs to a certain predefined class. A class can be seen as a decision that has to be made. All the observations combined with their class labels are known as a data set. When a new observation is received, that observation is classified based on previous experience.[198]

A classifier can be trained in various ways; there are many statistical and machine learning approaches. The decision tree[199] is perhaps the most widely used machine learning algorithm. Other widely used classifiers are the neural network,[201] k-nearest neighbor algorithm,[g][203] kernel methods such as the support vector machine (SVM),[h][205] Gaussian mixture model[206] and the extremely popular naive Bayes classifier.[i][208] Classifier performance depends greatly on the characteristics of the data to be classified, such as the dataset size, the dimensionality, and the level of noise. Model-based classifiers perform well if the assumed model is an extremely good fit for the actual data. Otherwise, if no matching model is available, and if accuracy (rather than speed or scalability) is the sole concern, conventional wisdom is that discriminative classifiers (especially SVM) tend to be more accurate than model-based classifiers such as “naive Bayes” on most practical data sets.[209]

Neural networks, or neural nets, were inspired by the architecture of neurons in the human brain. A simple “neuron” N accepts input from multiple other neurons, each of which, when activated (or “fired”), cast a weighted “vote” for or against whether neuron N should itself activate. Learning requires an algorithm to adjust these weights based on the training data; one simple algorithm (dubbed “fire together, wire together”) is to increase the weight between two connected neurons when the activation of one triggers the successful activation of another. The net forms “concepts” that are distributed among a subnetwork of shared[j] neurons that tend to fire together; a concept meaning “leg” might be coupled with a subnetwork meaning “foot” that includes the sound for “foot”. Neurons have a continuous spectrum of activation; in addition, neurons can process inputs in a nonlinear way rather than weighing straightforward votes. Modern neural nets can learn both continuous functions and, surprisingly, digital logical operations. Neural networks’ early successes included predicting the stock market and (in 1995) a mostly self-driving car.[k] In the 2010s, advances in neural networks using deep learning thrust AI into widespread public consciousness and contributed to an enormous upshift in corporate AI spending; for example, AI-related M&A in 2017 was over 25 times as large as in 2015.[212][213]

The study of non-learning artificial neural networks[201] began in the decade before the field of AI research was founded, in the work of Walter Pitts and Warren McCullouch. Frank Rosenblatt invented the perceptron, a learning network with a single layer, similar to the old concept of linear regression. Early pioneers also include Alexey Grigorevich Ivakhnenko, Teuvo Kohonen, Stephen Grossberg, Kunihiko Fukushima, Christoph von der Malsburg, David Willshaw, Shun-Ichi Amari, Bernard Widrow, John Hopfield, Eduardo R. Caianiello, and others.

The main categories of networks are acyclic or feedforward neural networks (where the signal passes in only one direction) and recurrent neural networks (which allow feedback and short-term memories of previous input events). Among the most popular feedforward networks are perceptrons, multi-layer perceptrons and radial basis networks.[214] Neural networks can be applied to the problem of intelligent control (for robotics) or learning, using such techniques as Hebbian learning (“fire together, wire together”), GMDH or competitive learning.[215]

Today, neural networks are often trained by the backpropagation algorithm, which had been around since 1970 as the reverse mode of automatic differentiation published by Seppo Linnainmaa,[216][217] and was introduced to neural networks by Paul Werbos.[218][219][220]

Hierarchical temporal memory is an approach that models some of the structural and algorithmic properties of the neocortex.[221]

In short, most neural networks use some form of gradient descent on a hand-created neural topology. However, some research groups, such as Uber, argue that simple neuroevolution to mutate new neural network topologies and weights may be competitive with sophisticated gradient descent approaches. One advantage of neuroevolution is that it may be less prone to get caught in “dead ends”.[222]

Deep learning is any artificial neural network that can learn a long chain of causal links. For example, a feedforward network with six hidden layers can learn a seven-link causal chain (six hidden layers + output layer) and has a “credit assignment path” (CAP) depth of seven. Many deep learning systems need to be able to learn chains ten or more causal links in length.[223] Deep learning has transformed many important subfields of artificial intelligence, including computer vision, speech recognition, natural language processing and others.[224][225][223]

According to one overview,[226] the expression “Deep Learning” was introduced to the Machine Learning community by Rina Dechter in 1986[227] and gained traction after Igor Aizenberg and colleagues introduced it to Artificial Neural Networks in 2000.[228] The first functional Deep Learning networks were published by Alexey Grigorevich Ivakhnenko and V. G. Lapa in 1965.[229][pageneeded] These networks are trained one layer at a time. Ivakhnenko’s 1971 paper[230] describes the learning of a deep feedforward multilayer perceptron with eight layers, already much deeper than many later networks. In 2006, a publication by Geoffrey Hinton and Ruslan Salakhutdinov introduced another way of pre-training many-layered feedforward neural networks (FNNs) one layer at a time, treating each layer in turn as an unsupervised restricted Boltzmann machine, then using supervised backpropagation for fine-tuning. Similar to shallow artificial neural networks, deep neural networks can model complex non-linear relationships. Over the last few years, advances in both machine learning algorithms and computer hardware have led to more efficient methods for training deep neural networks that contain many layers of non-linear hidden units and a very large output layer.[232]

Deep learning often uses convolutional neural networks (CNNs), whose origins can be traced back to the Neocognitron introduced by Kunihiko Fukushima in 1980.[233] In 1989, Yann LeCun and colleagues applied backpropagation to such an architecture. In the early 2000s, in an industrial application CNNs already processed an estimated 10% to 20% of all the checks written in the US.[234] Since 2011, fast implementations of CNNs on GPUs have won many visual pattern recognition competitions.[223]

CNNs with 12 convolutional layers were used in conjunction with reinforcement learning by Deepmind’s “AlphaGo Lee”, the program that beat a top Go champion in 2016.[235]

Early on, deep learning was also applied to sequence learning with recurrent neural networks (RNNs)[236] which are in theory Turing complete[237] and can run arbitrary programs to process arbitrary sequences of inputs. The depth of an RNN is unlimited and depends on the length of its input sequence; thus, an RNN is an example of deep learning.[223] RNNs can be trained by gradient descent[238][239][240] but suffer from the vanishing gradient problem.[224][241] In 1992, it was shown that unsupervised pre-training of a stack of recurrent neural networks can speed up subsequent supervised learning of deep sequential problems.[242]

Numerous researchers now use variants of a deep learning recurrent NN called the long short-term memory (LSTM) network published by Hochreiter & Schmidhuber in 1997.[243] LSTM is often trained by Connectionist Temporal Classification (CTC).[244] At Google, Microsoft and Baidu this approach has revolutionised speech recognition.[245][246][247] For example, in 2015, Google’s speech recognition experienced a dramatic performance jump of 49% through CTC-trained LSTM, which is now available through Google Voice to billions of smartphone users.[248] Google also used LSTM to improve machine translation,[249] Language Modeling[250] and Multilingual Language Processing.[251] LSTM combined with CNNs also improved automatic image captioning[252] and a plethora of other applications.

AI, like electricity or the steam engine, is a general purpose technology. There is no consensus on how to characterize which tasks AI tends to excel at.[253] While projects such as AlphaZero have succeeded in generating their own knowledge from scratch, many other machine learning projects require large training datasets.[254][255] Researcher Andrew Ng has suggested, as a “highly imperfect rule of thumb”, that “almost anything a typical human can do with less than one second of mental thought, we can probably now or in the near future automate using AI.”[256] Moravec’s paradox suggests that AI lags humans at many tasks that the human brain has specifically evolved to perform well.[120]

Games provide a well-publicized benchmark for assessing rates of progress. AlphaGo around 2016 brought the era of classical board-game benchmarks to an close. Games of imperfect knowledge provide new challenges to AI in the area of game theory.[257][258] E-sports such as StarCraft continue to provide additional public benchmarks.[259][260] There are many competitions and prizes, such as the Imagenet Challenge, to promote research in artificial intelligence. The main areas of competition include general machine intelligence, conversational behavior, data-mining, robotic cars, and robot soccer as well as conventional games.[citation needed]

The “imitation game” (an interpretation of the 1950 Turing test that assesses whether a computer can imitate a human) is nowadays considered too exploitable to be a meaningful benchmark.[261] A derivative of the Turing test is the Completely Automated Public Turing test to tell Computers and Humans Apart (CAPTCHA). As the name implies, this helps to determine that a user is an actual person and not a computer posing as a human. In contrast to the standard Turing test, CAPTCHA is administered by a machine and targeted to a human as opposed to being administered by a human and targeted to a machine. A computer asks a user to complete a simple test then generates a grade for that test. Computers are unable to solve the problem, so correct solutions are deemed to be the result of a person taking the test. A common type of CAPTCHA is the test that requires the typing of distorted letters, numbers or symbols that appear in an image undecipherable by a computer.

Proposed “universal intelligence” tests aim to compare how well machines, humans, and even non-human animals perform on problem sets that are generic as possible. At an extreme, the test suite can contain every possible problem, weighted by Kolmogorov complexity; unfortunately, these problem sets tend to be dominated by impoverished pattern-matching exercises where a tuned AI can easily exceed human performance levels.[263][264]

AI is relevant to any intellectual task. Modern artificial intelligence techniques are pervasive and are too numerous to list here. Frequently, when a technique reaches mainstream use, it is no longer considered artificial intelligence; this phenomenon is described as the AI effect.

High-profile examples of AI include autonomous vehicles (such as drones and self-driving cars), medical diagnosis, creating art (such as poetry), proving mathematical theorems, playing games (such as Chess or Go), search engines (such as Google search), online assistants (such as Siri), image recognition in photographs, spam filtering, prediction of judicial decisions[267] and targeting online advertisements.[268][269]

With social media sites overtaking TV as a source for news for young people and news organisations increasingly reliant on social media platforms for generating distribution,[270] major publishers now use artificial intelligence (AI) technology to post stories more effectively and generate higher volumes of traffic.[271]

Artificial intelligence is breaking into the healthcare industry by assisting doctors. According to Bloomberg Technology, Microsoft has developed AI to help doctors find the right treatments for cancer.[272] There is a great amount of research and drugs developed relating to cancer. In detail, there are more than 800 medicines and vaccines to treat cancer. This negatively affects the doctors, because there are too many options to choose from, making it more difficult to choose the right drugs for the patients. Microsoft is working on a project to develop a machine called “Hanover”. Its goal is to memorize all the papers necessary to cancer and help predict which combinations of drugs will be most effective for each patient. One project that is being worked on at the moment is fighting myeloid leukemia, a fatal cancer where the treatment has not improved in decades. Another study was reported to have found that artificial intelligence was as good as trained doctors in identifying skin cancers.[273] Another study is using artificial intelligence to try and monitor multiple high-risk patients, and this is done by asking each patient numerous questions based on data acquired from live doctor to patient interactions.[274]

According to CNN, a recent study by surgeons at the Children’s National Medical Center in Washington successfully demonstrated surgery with an autonomous robot. The team supervised the robot while it performed soft-tissue surgery, stitching together a pig’s bowel during open surgery, and doing so better than a human surgeon, the team claimed.[275] IBM has created its own artificial intelligence computer, the IBM Watson, which has beaten human intelligence (at some levels). Watson not only won at the game show Jeopardy! against former champions,[276] but was declared a hero after successfully diagnosing a woman who was suffering from leukemia.[277]

Advancements in AI have contributed to the growth of the automotive industry through the creation and evolution of self-driving vehicles. As of 2016, there are over 30 companies utilizing AI into the creation of driverless cars. A few companies involved with AI include Tesla, Google, and Apple.[278]

Many components contribute to the functioning of self-driving cars. These vehicles incorporate systems such as braking, lane changing, collision prevention, navigation and mapping. Together, these systems, as well as high performance computers, are integrated into one complex vehicle.[279]

Recent developments in autonomous automobiles have made the innovation of self-driving trucks possible, though they are still in the testing phase. The UK government has passed legislation to begin testing of self-driving truck platoons in 2018.[280] Self-driving truck platoons are a fleet of self-driving trucks following the lead of one non-self-driving truck, so the truck platoons aren’t entirely autonomous yet. Meanwhile, the Daimler, a German automobile corporation, is testing the Freightliner Inspiration which is a semi-autonomous truck that will only be used on the highway.[281]

One main factor that influences the ability for a driver-less automobile to function is mapping. In general, the vehicle would be pre-programmed with a map of the area being driven. This map would include data on the approximations of street light and curb heights in order for the vehicle to be aware of its surroundings. However, Google has been working on an algorithm with the purpose of eliminating the need for pre-programmed maps and instead, creating a device that would be able to adjust to a variety of new surroundings.[282] Some self-driving cars are not equipped with steering wheels or brake pedals, so there has also been research focused on creating an algorithm that is capable of maintaining a safe environment for the passengers in the vehicle through awareness of speed and driving conditions.[283]

Another factor that is influencing the ability for a driver-less automobile is the safety of the passenger. To make a driver-less automobile, engineers must program it to handle high risk situations. These situations could include a head on collision with pedestrians. The car’s main goal should be to make a decision that would avoid hitting the pedestrians and saving the passengers in the car. But there is a possibility the car would need to make a decision that would put someone in danger. In other words, the car would need to decide to save the pedestrians or the passengers.[284] The programing of the car in these situations is crucial to a successful driver-less automobile.

Financial institutions have long used artificial neural network systems to detect charges or claims outside of the norm, flagging these for human investigation. The use of AI in banking can be traced back to 1987 when Security Pacific National Bank in US set-up a Fraud Prevention Task force to counter the unauthorised use of debit cards. Programs like Kasisto and Moneystream are using AI in financial services.

Banks use artificial intelligence systems today to organize operations, maintain book-keeping, invest in stocks, and manage properties. AI can react to changes overnight or when business is not taking place.[285] In August 2001, robots beat humans in a simulated financial trading competition.[286] AI has also reduced fraud and financial crimes by monitoring behavioral patterns of users for any abnormal changes or anomalies.[287]

The use of AI machines in the market in applications such as online trading and decision making has changed major economic theories.[288] For example, AI based buying and selling platforms have changed the law of supply and demand in that it is now possible to easily estimate individualized demand and supply curves and thus individualized pricing. Furthermore, AI machines reduce information asymmetry in the market and thus making markets more efficient while reducing the volume of trades. Furthermore, AI in the markets limits the consequences of behavior in the markets again making markets more efficient. Other theories where AI has had impact include in rational choice, rational expectations, game theory, Lewis turning point, portfolio optimization and counterfactual thinking.

In video games, artificial intelligence is routinely used to generate dynamic purposeful behavior in non-player characters (NPCs). In addition, well-understood AI techniques are routinely used for pathfinding. Some researchers consider NPC AI in games to be a “solved problem” for most production tasks. Games with more atypical AI include the AI director of Left 4 Dead (2008) and the neuroevolutionary training of platoons in Supreme Commander 2 (2010).[289][290]

Worldwide annual military spending on robotics rose from 5.1 billion USD in 2010 to 7.5 billion USD in 2015.[291][292] Military drones capable of autonomous action are widely considered a useful asset. In 2017, Vladimir Putin stated that “Whoever becomes the leader in (artificial intelligence) will become the ruler of the world”.[293][294] Many artificial intelligence researchers seek to distance themselves from military applications of AI.[295]

For financial statements audit, AI makes continuous audit possible. AI tools could analyze many sets of different information immediately. The potential benefit would be the overall audit risk will be reduced, the level of assurance will be increased and the time duration of audit will be reduced.[296]

A report by the Guardian newspaper in the UK in 2018 found that online gambling companies were using AI to predict the behavior of customers in order to target them with personalized promotions.[297]

There are three philosophical questions related to AI:

Can a machine be intelligent? Can it “think”?

Widespread use of artificial intelligence could have unintended consequences that are dangerous or undesirable. Scientists from the Future of Life Institute, among others, described some short-term research goals to see how AI influences the economy, the laws and ethics that are involved with AI and how to minimize AI security risks. In the long-term, the scientists have proposed to continue optimizing function while minimizing possible security risks that come along with new technologies.[308]

Machines with intelligence have the potential to use their intelligence to make ethical decisions. Research in this area includes “machine ethics”, “artificial moral agents”, and the study of “malevolent vs. friendly AI”.

Link:

Artificial intelligence – Wikipedia

What is artificial intelligence? – BBC News

A computer can beat the world chess champion and understand voice commands on your smartphone, but real artificial intelligence has yet to arrive. The pace of change is quickening, though.

Some people say it will save humanity, even make us immortal.

Others say it could destroy us all.

But, the truth is, most of us don’t really know what AI is.

Video production by Valery Eremenko

Intelligent Machines – a BBC News series looking at AI and robotics

See the original post:

What is artificial intelligence? – BBC News

Online Artificial Intelligence Courses | Microsoft …

The Microsoft Professional Program (MPP) is a collection of courses that teach skills in several core technology tracks that help you excel in the industry’s newest job roles.

These courses are created and taught by experts and feature quizzes, hands-on labs, and engaging communities. For each track you complete, you earn a certificate of completion from Microsoft proving that you mastered those skills.

View post:

Online Artificial Intelligence Courses | Microsoft …

A.I. Artificial Intelligence – Wikipedia

A.I. Artificial Intelligence, also known as A.I., is a 2001 American science fiction drama film directed by Steven Spielberg. The screenplay by Spielberg and screen story by Ian Watson were based on the 1969 short story “Supertoys Last All Summer Long” by Brian Aldiss. The film was produced by Kathleen Kennedy, Spielberg and Bonnie Curtis. It stars Haley Joel Osment, Jude Law, Frances O’Connor, Brendan Gleeson and William Hurt. Set in a futuristic post-climate change society, A.I. tells the story of David (Osment), a childlike android uniquely programmed with the ability to love.

Development of A.I. originally began with producer-director Stanley Kubrick, after he acquired the rights to Aldiss’ story in the early 1970s. Kubrick hired a series of writers until the mid-1990s, including Brian Aldiss, Bob Shaw, Ian Watson, and Sara Maitland. The film languished in protracted development for years, partly because Kubrick felt computer-generated imagery was not advanced enough to create the David character, whom he believed no child actor would convincingly portray. In 1995, Kubrick handed A.I. to Spielberg, but the film did not gain momentum until Kubrick’s death in 1999. Spielberg remained close to Watson’s film treatment for the screenplay.

The film divided critics, with the overall balance being positive, and grossed approximately $235 million. The film was nominated for two Academy Awards at the 74th Academy Awards, for Best Visual Effects and Best Original Score (by John Williams).

In a 2016 BBC poll of 177 critics around the world, Steven Spielberg’s A.I. Artificial Intelligence was voted the eighty-third greatest film since 2000.[3] A.I. is dedicated to Stanley Kubrick.

In the late 22nd century, rising sea levels from global warming have wiped out coastal cities such as Amsterdam, Venice, and New York, and drastically reduced the world’s population. A new type of robots called Mecha, advanced humanoids capable of thoughts and emotions, have been created.

David, a Mecha that resembles a human child and is programmed to display love for its owners, is sent to Henry Swinton, and his wife, Monica, as a replacement for their son, Martin, who has been placed in suspended animation until he can be cured of a rare disease. Monica warms to David and activates his imprinting protocol, causing him to have an enduring childlike love for her. David is befriended by Teddy, a robotic teddy bear, who cares for David’s well-being.

Martin is cured of his disease and brought home; as he recovers, he grows jealous of David. He makes David go to Monica in the night and cut off a lock of her hair. This upsets the parents, particularly Henry, who fears that the scissors are a weapon.

At a pool party, one of Martin’s friends pokes David with a knife, activating his self-protection programming. David grabs Martin and they fall into the pool. Martin is saved from drowning, but Henry persuades Monica to return David to his creator for destruction. Instead, Monica abandons both David and Teddy in the forest to hide as an unregistered Mecha.

David is captured for an anti-Mecha “Flesh Fair”, where obsolete and unlicensed Mecha are destroyed before cheering crowds. David is nearly killed, but tricks the crowd into thinking that he is human, and escapes with Gigolo Joe, a male prostitute Mecha who is on the run after being framed for murder. The two set out to find the Blue Fairy, whom David remembers from The Adventures of Pinocchio, and believes can turn him into a human, allowing Monica to love him and take him home.

Joe and David make their way to the resort town, Rouge City, where “Dr. Know”, a holographic answer engine, leads them to the top of Rockefeller Center in the flooded ruins of Manhattan. There, David meets a copy of himself and destroys it. David then meets his creator, Professor Hobby, who tells David that he was built in the image of the professor’s dead son David, and that more copies, including female versions called Darlene, are being manufactured.

Disheartened, David falls from a ledge, but is rescued by Joe using their amphibicopter. David tells Joe he saw the Blue Fairy underwater and wants to go down to meet her. Joe is captured by the authorities using an electromagnet. David and Teddy use the amphibicopter to go to the Fairy, which turns out to be a statue at the now-sunken Coney Island. The two become trapped when the Wonder Wheel falls on their vehicle. David asks repeatedly to be turned into a real boy until the ocean freezes and is deactivated once his power source is drained.

Two thousand years later, humans have become extinct, and Manhattan is buried under glacial ice. The Mecha have evolved into an advanced, intelligent, silicon-based form. They find David and Teddy, and discover they are original Mecha that knew living humans, making them special.

David is revived and walks to the frozen Fairy statue, which collapses when he touches it. The Mecha use Davids memories to reconstruct the Swinton home and explain to him that they cannot make him human. However, David insists that they recreate Monica from DNA in the lock of hair. The Mecha warn David that the clone can only live for a day, and that the process cannot be repeated. David spends the next day with Monica and Teddy. Before she drifts off to sleep, Monica tells David she has always loved him. Teddy climbs onto the bed and watches the two lie peacefully together.

Kubrick began development on an adaptation of “Super-Toys Last All Summer Long” in the late 1970s, hiring the story’s author, Brian Aldiss, to write a film treatment. In 1985, Kubrick asked Steven Spielberg to direct the film, with Kubrick producing.[6] Warner Bros. agreed to co-finance A.I. and cover distribution duties.[7] The film labored in development hell, and Aldiss was fired by Kubrick over creative differences in 1989.[8] Bob Shaw served as writer very briefly, leaving after six weeks because of Kubrick’s demanding work schedule, and Ian Watson was hired as the new writer in March 1990. Aldiss later remarked, “Not only did the bastard fire me, he hired my enemy [Watson] instead.” Kubrick handed Watson The Adventures of Pinocchio for inspiration, calling A.I. “a picaresque robot version of Pinocchio”.[7][9]

Three weeks later Watson gave Kubrick his first story treatment, and concluded his work on A.I. in May 1991 with another treatment, at 90 pages. Gigolo Joe was originally conceived as a G.I. Mecha, but Watson suggested changing him to a male prostitute. Kubrick joked, “I guess we lost the kiddie market.”[7] In the meantime, Kubrick dropped A.I. to work on a film adaptation of Wartime Lies, feeling computer animation was not advanced enough to create the David character. However, after the release of Spielberg’s Jurassic Park (with its innovative use of computer-generated imagery), it was announced in November 1993 that production would begin in 1994.[10] Dennis Muren and Ned Gorman, who worked on Jurassic Park, became visual effects supervisors,[8] but Kubrick was displeased with their previsualization, and with the expense of hiring Industrial Light & Magic.[11]

Stanley [Kubrick] showed Steven [Spielberg] 650 drawings which he had, and the script and the story, everything. Stanley said, “Look, why don’t you direct it and I’ll produce it.” Steven was almost in shock.

Producer Jan Harlan, on Spielberg’s first meeting with Kubrick about A.I.[12]

In early 1994, the film was in pre-production with Christopher “Fangorn” Baker as concept artist, and Sara Maitland assisting on the story, which gave it “a feminist fairy-tale focus”.[7] Maitland said that Kubrick never referred to the film as A.I., but as Pinocchio.[11] Chris Cunningham became the new visual effects supervisor. Some of his unproduced work for A.I. can be seen on the DVD, The Work of Director Chris Cunningham.[13] Aside from considering computer animation, Kubrick also had Joseph Mazzello do a screen test for the lead role.[11] Cunningham helped assemble a series of “little robot-type humans” for the David character. “We tried to construct a little boy with a movable rubber face to see whether we could make it look appealing,” producer Jan Harlan reflected. “But it was a total failure, it looked awful.” Hans Moravec was brought in as a technical consultant.[11] Meanwhile, Kubrick and Harlan thought A.I. would be closer to Steven Spielberg’s sensibilities as director.[14][15] Kubrick handed the position to Spielberg in 1995, but Spielberg chose to direct other projects, and convinced Kubrick to remain as director.[12][16] The film was put on hold due to Kubrick’s commitment to Eyes Wide Shut (1999).[17] After the filmmaker’s death in March 1999, Harlan and Christiane Kubrick approached Spielberg to take over the director’s position.[18][19] By November 1999, Spielberg was writing the screenplay based on Watson’s 90-page story treatment. It was his first solo screenplay credit since Close Encounters of the Third Kind (1977).[20] Spielberg remained close to Watson’s treatment, but removed various sex scenes with Gigolo Joe. Pre-production was briefly halted during February 2000, because Spielberg pondered directing other projects, which were Harry Potter and the Philosopher’s Stone, Minority Report and Memoirs of a Geisha.[17][21] The following month Spielberg announced that A.I. would be his next project, with Minority Report as a follow-up.[22] When he decided to fast track A.I., Spielberg brought Chris Baker back as concept artist.[16]

The original start date was July 10, 2000,[15] but filming was delayed until August.[23] Aside from a couple of weeks shooting on location in Oxbow Regional Park in Oregon, A.I. was shot entirely using sound stages at Warner Bros. Studios and the Spruce Goose Dome in Long Beach, California.[24] The Swinton house was constructed on Stage 16, while Stage 20 was used for Rouge City and other sets.[25][26] Spielberg copied Kubrick’s obsessively secretive approach to filmmaking by refusing to give the complete script to cast and crew, banning press from the set, and making actors sign confidentiality agreements. Social robotics expert Cynthia Breazeal served as technical consultant during production.[15][27] Haley Joel Osment and Jude Law applied prosthetic makeup daily in an attempt to look shinier and robotic.[4] Costume designer Bob Ringwood (Batman, Troy) studied pedestrians on the Las Vegas Strip for his influence on the Rouge City extras.[28] Spielberg found post-production on A.I. difficult because he was simultaneously preparing to shoot Minority Report.[29]

The film’s soundtrack was released by Warner Sunset Records in 2001. The original score was composed and conducted by John Williams and featured singers Lara Fabian on two songs and Josh Groban on one. The film’s score also had a limited release as an official “For your consideration Academy Promo”, as well as a complete score issue by La-La Land Records in 2015.[30] The band Ministry appears in the film playing the song “What About Us?” (but the song does not appear on the official soundtrack album).

Warner Bros. used an alternate reality game titled The Beast to promote the film. Over forty websites were created by Atomic Pictures in New York City (kept online at Cloudmakers.org) including the website for Cybertronics Corp. There were to be a series of video games for the Xbox video game console that followed the storyline of The Beast, but they went undeveloped. To avoid audiences mistaking A.I. for a family film, no action figures were created, although Hasbro released a talking Teddy following the film’s release in June 2001.[15]

A.I. had its premiere at the Venice Film Festival in 2001.[31]

A.I. Artificial Intelligence was released on VHS and DVD by Warner Home Video on March 5, 2002 in both a standard full-screen release which included no bonus features and as a 2-Disc Special Edition featuring the film in its original 1.85:1 anamorphic widescreen format as well as an eight-part documentary detailing the film’s development, production, music and visual effects. The bonus features also included interviews with Haley Joel Osment, Jude Law, Frances O’Connor, Steven Spielberg and John Williams, two teaser trailers for the film’s original theatrical release and an extensive photo gallery featuring production sills and Stanley Kubrick’s original storyboards.[32]

The film was released on Blu-ray Disc on April 5, 2011 by Paramount Home Media Distribution for the U.S. and by Warner Home Video for international markets. This release featured the film a newly restored high-definition print and incorporated all the bonus features previously included on the 2-Disc Special Edition DVD.[33]

The film opened in 3,242 theaters in the United States on June 29, 2001, earning $29,352,630 during its opening weekend. A.I went on to gross $78.62 million in US totals as well as $157.31 million in foreign countries, coming to a worldwide total of $235.93 million.[34]

Based on 191 reviews collected by Rotten Tomatoes, 73% of critics gave the film positive notices with a score of 6.6 out of 10. The website’s statement of the critical consensus reads, “A curious, not always seamless, amalgamation of Kubrick’s chilly bleakness and Spielberg’s warm-hearted optimism. [The film] is, in a word, fascinating.”[35] By comparison, Metacritic collected an average score of 65, based on 32 reviews, which is considered favorable.[36]

Producer Jan Harlan stated that Kubrick “would have applauded” the final film, while Kubrick’s widow Christiane also enjoyed A.I.[37] Brian Aldiss admired the film as well: “I thought what an inventive, intriguing, ingenious, involving film this was. There are flaws in it and I suppose I might have a personal quibble but it’s so long since I wrote it.” Of the film’s ending, he wondered how it might have been had Kubrick directed the film: “That is one of the ‘ifs’ of film historyat least the ending indicates Spielberg adding some sugar to Kubrick’s wine. The actual ending is overly sympathetic and moreover rather overtly engineered by a plot device that does not really bear credence. But it’s a brilliant piece of film and of course it’s a phenomenon because it contains the energies and talents of two brilliant filmmakers.”[38] Richard Corliss heavily praised Spielberg’s direction, as well as the cast and visual effects.[39] Roger Ebert gave the film four stars, saying that it was “wonderful and maddening.”[40] Leonard Maltin, on the other hand, gives the film two stars out of four in his Movie Guide, writing: “[The] intriguing story draws us in, thanks in part to Osment’s exceptional performance, but takes several wrong turns; ultimately, it just doesn’t work. Spielberg rewrote the adaptation Stanley Kubrick commissioned of the Brian Aldiss short story ‘Super Toys Last All Summer Long’; [the] result is a curious and uncomfortable hybrid of Kubrick and Spielberg sensibilities.” However, he calls John Williams’ music score “striking”. Jonathan Rosenbaum compared A.I. to Solaris (1972), and praised both “Kubrick for proposing that Spielberg direct the project and Spielberg for doing his utmost to respect Kubrick’s intentions while making it a profoundly personal work.”[41] Film critic Armond White, of the New York Press, praised the film noting that “each part of Davids journey through carnal and sexual universes into the final eschatological devastation becomes as profoundly philosophical and contemplative as anything by cinemas most thoughtful, speculative artists Borzage, Ozu, Demy, Tarkovsky.”[42] Filmmaker Billy Wilder hailed A.I. as “the most underrated film of the past few years.”[43] When British filmmaker Ken Russell saw the film, he wept during the ending.[44]

Mick LaSalle gave a largely negative review. “A.I. exhibits all its creators’ bad traits and none of the good. So we end up with the structureless, meandering, slow-motion endlessness of Kubrick combined with the fuzzy, cuddly mindlessness of Spielberg.” Dubbing it Spielberg’s “first boring movie”, LaSalle also believed the robots at the end of the film were aliens, and compared Gigolo Joe to the “useless” Jar Jar Binks, yet praised Robin Williams for his portrayal of a futuristic Albert Einstein.[45][not in citation given] Peter Travers gave a mixed review, concluding “Spielberg cannot live up to Kubrick’s darker side of the future.” But he still put the film on his top ten list that year for best movies.[46] David Denby in The New Yorker criticized A.I. for not adhering closely to his concept of the Pinocchio character. Spielberg responded to some of the criticisms of the film, stating that many of the “so called sentimental” elements of A.I., including the ending, were in fact Kubrick’s and the darker elements were his own.[47] However, Sara Maitland, who worked on the project with Kubrick in the 1990s, claimed that one of the reasons Kubrick never started production on A.I. was because he had a hard time making the ending work.[48] James Berardinelli found the film “consistently involving, with moments of near-brilliance, but far from a masterpiece. In fact, as the long-awaited ‘collaboration’ of Kubrick and Spielberg, it ranks as something of a disappointment.” Of the film’s highly debated finale, he claimed, “There is no doubt that the concluding 30 minutes are all Spielberg; the outstanding question is where Kubrick’s vision left off and Spielberg’s began.”[49]

Screenwriter Ian Watson has speculated, “Worldwide, A.I. was very successful (and the 4th highest earner of the year) but it didn’t do quite so well in America, because the film, so I’m told, was too poetical and intellectual in general for American tastes. Plus, quite a few critics in America misunderstood the film, thinking for instance that the Giacometti-style beings in the final 20 minutes were aliens (whereas they were robots of the future who had evolved themselves from the robots in the earlier part of the film) and also thinking that the final 20 minutes were a sentimental addition by Spielberg, whereas those scenes were exactly what I wrote for Stanley and exactly what he wanted, filmed faithfully by Spielberg.”[50]

In 2002, Spielberg told film critic Joe Leydon that “People pretend to think they know Stanley Kubrick, and think they know me, when most of them don’t know either of us”. “And what’s really funny about that is, all the parts of A.I. that people assume were Stanley’s were mine. And all the parts of A.I. that people accuse me of sweetening and softening and sentimentalizing were all Stanley’s. The teddy bear was Stanley’s. The whole last 20 minutes of the movie was completely Stanley’s. The whole first 35, 40 minutes of the film all the stuff in the house was word for word, from Stanley’s screenplay. This was Stanley’s vision.” “Eighty percent of the critics got it all mixed up. But I could see why. Because, obviously, I’ve done a lot of movies where people have cried and have been sentimental. And I’ve been accused of sentimentalizing hard-core material. But in fact it was Stanley who did the sweetest parts of A.I., not me. I’m the guy who did the dark center of the movie, with the Flesh Fair and everything else. That’s why he wanted me to make the movie in the first place. He said, ‘This is much closer to your sensibilities than my own.'”[51]

Upon rewatching the film many years after its release, BBC film critic Mark Kermode apologized to Spielberg in an interview in January 2013 for “getting it wrong” on the film when he first viewed it in 2001. He now believes the film to be Spielberg’s “enduring masterpiece”.[52]

Visual effects supervisors Dennis Muren, Stan Winston, Michael Lantieri and Scott Farrar were nominated for the Academy Award for Best Visual Effects, while John Williams was nominated for Best Original Music Score.[53] Steven Spielberg, Jude Law and Williams received nominations at the 59th Golden Globe Awards.[54] A.I. was successful at the Saturn Awards, winning five awards, including Best Science Fiction Film along with Best Writing for Spielberg and Best Performance by a Younger Actor for Osment.[55]

View original post here:

A.I. Artificial Intelligence – Wikipedia

Benefits & Risks of Artificial Intelligence – Future of …

Many AI researchers roll their eyes when seeing this headline:Stephen Hawking warns that rise of robots may be disastrous for mankind. And as many havelost count of how many similar articles theyveseen.Typically, these articles are accompanied by an evil-looking robot carrying a weapon, and they suggest we should worry about robots rising up and killing us because theyve become conscious and/or evil.On a lighter note, such articles are actually rather impressive, because they succinctly summarize the scenario that AI researchers dontworry about. That scenario combines as many as three separate misconceptions: concern about consciousness, evil, androbots.

If you drive down the road, you have a subjective experience of colors, sounds, etc. But does a self-driving car have a subjective experience? Does it feel like anything at all to be a self-driving car?Although this mystery of consciousness is interesting in its own right, its irrelevant to AI risk. If you get struck by a driverless car, it makes no difference to you whether it subjectively feels conscious. In the same way, what will affect us humans is what superintelligent AIdoes, not how it subjectively feels.

The fear of machines turning evil is another red herring. The real worry isnt malevolence, but competence. A superintelligent AI is by definition very good at attaining its goals, whatever they may be, so we need to ensure that its goals are aligned with ours. Humans dont generally hate ants, but were more intelligent than they are so if we want to build a hydroelectric dam and theres an anthill there, too bad for the ants. The beneficial-AI movement wants to avoid placing humanity in the position of those ants.

The consciousness misconception is related to the myth that machines cant have goals.Machines can obviously have goals in the narrow sense of exhibiting goal-oriented behavior: the behavior of a heat-seeking missile is most economically explained as a goal to hit a target.If you feel threatened by a machine whose goals are misaligned with yours, then it is precisely its goals in this narrow sense that troubles you, not whether the machine is conscious and experiences a sense of purpose.If that heat-seeking missile were chasing you, you probably wouldnt exclaim: Im not worried, because machines cant have goals!

I sympathize with Rodney Brooks and other robotics pioneers who feel unfairly demonized by scaremongering tabloids,because some journalists seem obsessively fixated on robots and adorn many of their articles with evil-looking metal monsters with red shiny eyes. In fact, the main concern of the beneficial-AI movement isnt with robots but with intelligence itself: specifically, intelligence whose goals are misaligned with ours. To cause us trouble, such misaligned superhuman intelligence needs no robotic body, merely an internet connection this may enable outsmarting financial markets, out-inventing human researchers, out-manipulating human leaders, and developing weapons we cannot even understand. Even if building robots were physically impossible, a super-intelligent and super-wealthy AI could easily pay or manipulate many humans to unwittingly do its bidding.

The robot misconception is related to the myth that machines cant control humans. Intelligence enables control: humans control tigers not because we are stronger, but because we are smarter. This means that if we cede our position as smartest on our planet, its possible that we might also cede control.

Continued here:

Benefits & Risks of Artificial Intelligence – Future of …

What is AI (artificial intelligence)? – Definition from …

AI (artificial intelligence) is the simulation of human intelligence processes by machines, especially computer systems. These processes include learning (the acquisition of information and rules for using the information), reasoning (using the rules to reach approximate or definite conclusions) and self-correction. Particular applications of AI include expert systems, speech recognition and machine vision.

AI was coined by John McCarthy, an American computer scientist, in 1956 at The Dartmouth Conference where the discipline was born. Today, it is an umbrella term that encompasses everything from robotic process automation to actual robotics. It has gained prominence recently due, in part, to big data, or the increase in speed, size and variety of data businesses are now collecting. AI can perform tasks such as identifying patterns in the data more efficiently than humans, enabling businesses to gain more insight out of their data.

AI can be categorized in any number of ways, but here are two examples.

The first classifies AI systems as either weak AI or strong AI. Weak AI, also known as narrow AI, is an AI system that is designed and trained for a particular task. Virtual personal assistants, such as Apple’s Siri, are a form of weak AI.

Strong AI, also known as artificial general intelligence, is an AI system with generalized human cognitive abilities so that when presented with an unfamiliar task, it has enough intelligence to find a solution. The Turing Test, developed by mathematician Alan Turing in 1950, is a method used to determine if a computer can actually think like a human, although the method is controversial.

The second example is from Arend Hintze, an assistant professor of integrative biology and computer science and engineering at Michigan State University. He categorizes AI into four types, from the kind of AI systems that exist today to sentient systems, which do not yet exist. His categories are as follows:

Continued here:

What is AI (artificial intelligence)? – Definition from …

Artificial intelligence – Wikipedia

Artificial intelligence (AI, also machine intelligence, MI) is intelligence demonstrated by machines, in contrast to the natural intelligence (NI) displayed by humans and other animals. In computer science AI research is defined as the study of “intelligent agents”: any device that perceives its environment and takes actions that maximize its chance of successfully achieving its goals.[1] Colloquially, the term “artificial intelligence” is applied when a machine mimics “cognitive” functions that humans associate with other human minds, such as “learning” and “problem solving”.

The scope of AI is disputed: as machines become increasingly capable, tasks considered as requiring “intelligence” are often removed from the definition, a phenomenon known as the AI effect, leading to the quip, “AI is whatever hasn’t been done yet.”[3] For instance, optical character recognition is frequently excluded from “artificial intelligence”, having become a routine technology.[4] Capabilities generally classified as AI as of 2017[update] include successfully understanding human speech, competing at the highest level in strategic game systems (such as chess and Go[6]), autonomous cars, intelligent routing in content delivery network and military simulations.

Artificial intelligence was founded as an academic discipline in 1956, and in the years since has experienced several waves of optimism,[7][8] followed by disappointment and the loss of funding (known as an “AI winter”),[9][10] followed by new approaches, success and renewed funding.[8][11] For most of its history, AI research has been divided into subfields that often fail to communicate with each other.[12] These sub-fields are based on technical considerations, such as particular goals (e.g. “robotics” or “machine learning”),[13] the use of particular tools (“logic” or “neural networks”), or deep philosophical differences.[14][15][16] Subfields have also been based on social factors (particular institutions or the work of particular researchers).[12]

The traditional problems (or goals) of AI research include reasoning, knowledge representation, planning, learning, natural language processing, perception and the ability to move and manipulate objects.[13] General intelligence is among the field’s long-term goals.[17] Approaches include statistical methods, computational intelligence, and traditional symbolic AI. Many tools are used in AI, including versions of search and mathematical optimization, neural networks and methods based on statistics, probability and economics. The AI field draws upon computer science, mathematics, psychology, linguistics, philosophy and many others.

The field was founded on the claim that human intelligence “can be so precisely described that a machine can be made to simulate it”.[18] This raises philosophical arguments about the nature of the mind and the ethics of creating artificial beings endowed with human-like intelligence which are issues that have been explored by myth, fiction and philosophy since antiquity.[19] Some people also consider AI to be a danger to humanity if it progresses unabatedly.[20] Others believe that AI, unlike previous technological revolutions, will create a risk of mass unemployment.[21]

In the twenty-first century, AI techniques have experienced a resurgence following concurrent advances in computer power, large amounts of data, and theoretical understanding; and AI techniques have become an essential part of the technology industry, helping to solve many challenging problems in computer science.[22][11]

Thought-capable artificial beings appeared as storytelling devices in antiquity,[23] and have been common in fiction, as in Mary Shelley’s Frankenstein or Karel apek’s R.U.R. (Rossum’s Universal Robots).[24] These characters and their fates raised many of the same issues now discussed in the ethics of artificial intelligence.[19]

The study of mechanical or “formal” reasoning began with philosophers and mathematicians in antiquity. The study of mathematical logic led directly to Alan Turing’s theory of computation, which suggested that a machine, by shuffling symbols as simple as “0” and “1”, could simulate any conceivable act of mathematical deduction. This insight, that digital computers can simulate any process of formal reasoning, is known as the ChurchTuring thesis.[25] Along with concurrent discoveries in neurobiology, information theory and cybernetics, this led researchers to consider the possibility of building an electronic brain. Turing proposed that “if a human could not distinguish between responses from a machine and a human, the machine could be considered intelligent”.[26] The first work that is now generally recognized as AI was McCullouch and Pitts’ 1943 formal design for Turing-complete “artificial neurons”.

The field of AI research was born at a workshop at Dartmouth College in 1956.[28] Attendees Allen Newell (CMU), Herbert Simon (CMU), John McCarthy (MIT), Marvin Minsky (MIT) and Arthur Samuel (IBM) became the founders and leaders of AI research.[29] They and their students produced programs that the press described as “astonishing”: computers were learning checkers strategies (c. 1954)[31] (and by 1959 were reportedly playing better than the average human),[32] solving word problems in algebra, proving logical theorems (Logic Theorist, first run c. 1956) and speaking English.[33] By the middle of the 1960s, research in the U.S. was heavily funded by the Department of Defense[34] and laboratories had been established around the world.[35] AI’s founders were optimistic about the future: Herbert Simon predicted, “machines will be capable, within twenty years, of doing any work a man can do”. Marvin Minsky agreed, writing, “within a generation… the problem of creating ‘artificial intelligence’ will substantially be solved”.[7]

They failed to recognize the difficulty of some of the remaining tasks. Progress slowed and in 1974, in response to the criticism of Sir James Lighthill and ongoing pressure from the US Congress to fund more productive projects, both the U.S. and British governments cut off exploratory research in AI. The next few years would later be called an “AI winter”,[9] a period when obtaining funding for AI projects was difficult.

In the early 1980s, AI research was revived by the commercial success of expert systems,[37] a form of AI program that simulated the knowledge and analytical skills of human experts. By 1985 the market for AI had reached over a billion dollars. At the same time, Japan’s fifth generation computer project inspired the U.S and British governments to restore funding for academic research.[8] However, beginning with the collapse of the Lisp Machine market in 1987, AI once again fell into disrepute, and a second, longer-lasting hiatus began.[10]

In the late 1990s and early 21st century, AI began to be used for logistics, data mining, medical diagnosis and other areas.[22] The success was due to increasing computational power (see Moore’s law), greater emphasis on solving specific problems, new ties between AI and other fields (such as statistics, economics and mathematics), and a commitment by researchers to mathematical methods and scientific standards.[38] Deep Blue became the first computer chess-playing system to beat a reigning world chess champion, Garry Kasparov on 11 May 1997.

In 2011, a Jeopardy! quiz show exhibition match, IBM’s question answering system, Watson, defeated the two greatest Jeopardy champions, Brad Rutter and Ken Jennings, by a significant margin. Faster computers, algorithmic improvements, and access to large amounts of data enabled advances in machine learning and perception; data-hungry deep learning methods started to dominate accuracy benchmarks around 2012.[41] The Kinect, which provides a 3D bodymotion interface for the Xbox 360 and the Xbox One use algorithms that emerged from lengthy AI research[42] as do intelligent personal assistants in smartphones.[43] In March 2016, AlphaGo won 4 out of 5 games of Go in a match with Go champion Lee Sedol, becoming the first computer Go-playing system to beat a professional Go player without handicaps.[6][44] In the 2017 Future of Go Summit, AlphaGo won a three-game match with Ke Jie,[45] who at the time continuously held the world No. 1 ranking for two years.[46][47] This marked the completion of a significant milestone in the development of Artificial Intelligence as Go is an extremely complex game, more so than Chess.

According to Bloomberg’s Jack Clark, 2015 was a landmark year for artificial intelligence, with the number of software projects that use AI within Google increased from a “sporadic usage” in 2012 to more than 2,700 projects. Clark also presents factual data indicating that error rates in image processing tasks have fallen significantly since 2011.[48] He attributes this to an increase in affordable neural networks, due to a rise in cloud computing infrastructure and to an increase in research tools and datasets.[11] Other cited examples include Microsoft’s development of a Skype system that can automatically translate from one language to another and Facebook’s system that can describe images to blind people.[48] In a 2017 survey, one in five companies reported they had “incorporated AI in some offerings or processes”.[49][50]

A typical AI perceives its environment and takes actions that maximize its chance of successfully achieving its goals.[1] An AI’s intended goal function can be simple (“1 if the AI wins a game of Go, 0 otherwise”) or complex (“Do actions mathematically similar to the actions that got you rewards in the past”). Goals can be explicitly defined, or can be induced. If the AI is programmed for “reinforcement learning”, goals can be implicitly induced by rewarding some types of behavior and punishing others.[a] Alternatively, an evolutionary system can induce goals by using a “fitness function” to mutate and preferentially replicate high-scoring AI systems; this is similar to how animals evolved to innately desire certain goals such as finding food, or how dogs can be bred via artificial selection to possess desired traits. Some AI systems, such as nearest-neighbor, instead reason by analogy; these systems are not generally given goals, except to the degree that goals are somehow implicit in their training data. Such systems can still be benchmarked if the non-goal system is framed as a system whose “goal” is to successfully accomplish its narrow classification task.[53]

AI often revolves around the use of algorithms. An algorithm is a set of unambiguous instructions that a mechanical computer can execute.[b] A complex algorithm is often built on top of other, simpler, algorithms. A simple example of an algorithm is the following recipe for optimal play at tic-tac-toe:

Many AI algorithms are capable of learning from data; they can enhance themselves by learning new heuristics (strategies, or “rules of thumb”, that have worked well in the past), or can themselves write other algorithms. Some of the “learners” described below, including Bayesian networks, decision trees, and nearest-neighbor, could theoretically, if given infinite data, time, and memory, learn to approximate any function, including whatever combination of mathematical functions would best describe the entire world. These learners could therefore, in theory, derive all possible knowledge, by considering every possible hypothesis and matching it against the data. In practice, it is almost never possible to consider every possibility, because of the phenomenon of “combinatorial explosion”, where the amount of time needed to solve a problem grows exponentially. Much of AI research involves figuring out how to identify and avoid considering broad swaths of possibililities that are unlikely to be fruitful.[55] For example, when viewing a map and looking for the shortest driving route from Denver to New York in the East, one can in most cases skip looking at any path through San Francisco or other areas far to the West; thus, an AI wielding an pathfinding algorithm like A* can avoid the combinatorial explosion that would ensue if every possible route had to be ponderously considered in turn.[57]

The earliest (and easiest to understand) approach to AI was symbolism (such as formal logic): “If an otherwise healthy adult has a fever, then they may have influenza”. A second, more general, approach is Bayesian inference: “If the current patient has a fever, adjust the probability they have influenza in such-and-such way”. The third major approach, extremely popular in routine business AI applications, is analogizers such as SVM and nearest-neighbor: “After examining the records of known past patients whose temperature, symptoms, age, and other factors mostly match the current patient, X% of those patients turned out to have influenza”. A fourth approach is harder to intuitively understand, but is inspired by how the brain’s machinery works: the neural network approach uses artificial “neurons” that can learn by comparing itself to the desired output and altering the strengths of the connections between its internal neurons to “reinforce” connections that seemed to be useful. These four main approaches can overlap with each other and with evolutionary systems; for example, neural nets can learn to make inferences, to generalize, and to make analogies. Some systems implicitly or explicitly use multiple of these approaches, alongside many other AI and non-AI algorithms; the best approach is often different depending on the problem.[59]

Learning algorithms work on the basis that strategies, algorithms, and inferences that worked well in the past are likely to continue working well in the future. These inferences can be obvious, such as “since the sun rose every morning for the last 10,000 days, it will probably rise tomorrow morning as well”. They can be nuanced, such as “X% of families have geographically separate species with color variants, so there is an Y% chance that undiscovered black swans exist”. Learners also work on the basis of “Occam’s razor”: The simplest theory that explains the data is the likeliest. Therefore, to be successful, a learner must be designed such that it prefers simpler theories to complex theories, except in cases where the complex theory is proven substantially better. Settling on a bad, overly complex theory gerrymandered to fit all the past training data is known as overfitting. Many systems attempt to reduce overfitting by rewarding a theory in accordance with how well it fits the data, but penalizing the theory in accordance with how complex the theory is. Besides classic overfitting, learners can also disappoint by “learning the wrong lesson”. A toy example is that an image classifier trained only on pictures of brown horses and black cats might conclude that all brown patches are likely to be horses. A real-world example is that, unlike humans, current image classifiers don’t determine the spatial relationship between components of the picture; instead, they learn abstract patterns of pixels that humans are oblivious to, but that linearly correlate with images of certain types of real objects. Faintly superimposing such a pattern on a legitimate image results in an “adversarial” image that the system misclassifies.[c][62][63][64]

Compared with humans, existing AI lacks several features of human “commonsense reasoning”; most notably, humans have powerful mechanisms for reasoning about “nave physics” such as space, time, and physical interactions. This enables even young children to easily make inferences like “If I roll this pen off a table, it will fall on the floor”. Humans also have a powerful mechanism of “folk psychology” that helps them to interpret natural-language sentences such as “The city councilmen refused the demonstrators a permit because they advocated violence”. (A generic AI has difficulty inferring whether the councilmen or the demonstrators are the ones alleged to be advocating violence.)[67][68][69] This lack of “common knowledge” means that AI often makes different mistakes than humans make, in ways that can seem incomprehensible. For example, existing self-driving cars cannot reason about the location nor the intentions of pedestrians in the exact way that humans do, and instead must use non-human modes of reasoning to avoid accidents.[70][71][72]

The overall research goal of artificial intelligence is to create technology that allows computers and machines to function in an intelligent manner. The general problem of simulating (or creating) intelligence has been broken down into sub-problems. These consist of particular traits or capabilities that researchers expect an intelligent system to display. The traits described below have received the most attention.[13]

Early researchers developed algorithms that imitated step-by-step reasoning that humans use when they solve puzzles or make logical deductions.[73] By the late 1980s and 1990s, AI research had developed methods for dealing with uncertain or incomplete information, employing concepts from probability and economics.[74]

These algorithms proved to be insufficient for solving large reasoning problems, because they experienced a “combinatorial explosion”: they became exponentially slower as the problems grew larger.[55] In fact, even humans rarely use the step-by-step deduction that early AI research was able to model. They solve most of their problems using fast, intuitive judgements.[75]

Knowledge representation[76] and knowledge engineering[77] are central to classical AI research. Some “expert systems” attempt to gather together explicit knowledge possessed by experts in some narrow domain. In addition, some projects attempt to gather the “commonsense knowledge” known to the average person into a database containing extensive knowledge about the world. Among the things a comprehensive commonsense knowledge base would contain are: objects, properties, categories and relations between objects;[78] situations, events, states and time;[79] causes and effects;[80] knowledge about knowledge (what we know about what other people know);[81] and many other, less well researched domains. A representation of “what exists” is an ontology: the set of objects, relations, concepts, and properties formally described so that software agents can interpret them. The semantics of these are captured as description logic concepts, roles, and individuals, and typically implemented as classes, properties, and individuals in the Web Ontology Language.[82] The most general ontologies are called upper ontologies, which attempt to provide a foundation for all other knowledge[83] by acting as mediators between domain ontologies that cover specific knowledge about a particular knowledge domain (field of interest or area of concern). Such formal knowledge representations can be used in content-based indexing and retrieval,[84] scene interpretation,[85] clinical decision support,[86] knowledge discovery (mining “interesting” and actionable inferences from large databases),[87] and other areas.[88]

Among the most difficult problems in knowledge representation are:

Intelligent agents must be able to set goals and achieve them.[95] They need a way to visualize the futurea representation of the state of the world and be able to make predictions about how their actions will change itand be able to make choices that maximize the utility (or “value”) of available choices.[96]

In classical planning problems, the agent can assume that it is the only system acting in the world, allowing the agent to be certain of the consequences of its actions.[97] However, if the agent is not the only actor, then it requires that the agent can reason under uncertainty. This calls for an agent that can not only assess its environment and make predictions, but also evaluate its predictions and adapt based on its assessment.[98]

Multi-agent planning uses the cooperation and competition of many agents to achieve a given goal. Emergent behavior such as this is used by evolutionary algorithms and swarm intelligence.[99]

Machine learning, a fundamental concept of AI research since the field’s inception,[100] is the study of computer algorithms that improve automatically through experience.[101][102]

Unsupervised learning is the ability to find patterns in a stream of input. Supervised learning includes both classification and numerical regression. Classification is used to determine what category something belongs in, after seeing a number of examples of things from several categories. Regression is the attempt to produce a function that describes the relationship between inputs and outputs and predicts how the outputs should change as the inputs change.[102] Both classifiers and regression learners can be viewed as “function approximators” trying to learn an unknown (possibly implicit) function; for example, a spam classifier can be viewed as learning a function that maps from the text of an email to one of two categories, “spam” or “not spam”. Computational learning theory can assess learners by computational complexity, by sample complexity (how much data is required), or by other notions of optimization.[103] In reinforcement learning[104] the agent is rewarded for good responses and punished for bad ones. The agent uses this sequence of rewards and punishments to form a strategy for operating in its problem space.

Natural language processing[105] (NLP) gives machines the ability to read and understand human language. A sufficiently powerful natural language processing system would enable natural-language user interfaces and the acquisition of knowledge directly from human-written sources, such as newswire texts. Some straightforward applications of natural language processing include information retrieval, text mining, question answering[106] and machine translation.[107] Many current approaches use word co-occurrence frequencies to construct syntactic representations of text. “Keyword spotting” strategies for search are popular and scalable but dumb; a search query for “dog” might only match documents with the literal word “dog” and miss a document with the word “poodle”. “Lexical affinity” strategies use the occurrence of words such as “accident” to assess the sentiment of a document. Modern statistical NLP approaches can combine all these strategies as well as others, and often achieve acceptable accuracy at the page or paragraph level, but continue to lack the semantic understanding required to classify isolated sentences well. Besides the usual difficulties with encoding semantic commonsense knowledge, existing semantic NLP sometimes scales too poorly to be viable in business applications. Beyond semantic NLP, the ultimate goal of “narrative” NLP is to embody a full understanding of commonsense reasoning.[108]

Machine perception[109] is the ability to use input from sensors (such as cameras (visible spectrum or infrared), microphones, wireless signals, and active lidar, sonar, radar, and tactile sensors) to deduce aspects of the world. Applications include speech recognition,[110] facial recognition, and object recognition.[111] Computer vision is the ability to analyze visual input. Such input is usually ambiguous; a giant, fifty-meter-tall pedestrian far away may produce exactly the same pixels as a nearby normal-sized pedestrian, requiring the AI to judge the relative likelihood and reasonableness of different interpretations, for example by using its “object model” to assess that fifty-meter pedestrians do not exist.[112]

AI is heavily used in robotics.[113] Advanced robotic arms and other industrial robots, widely used in modern factories, can learn from experience how to move efficiently despite the presence of friction and gear slippage.[114] A modern mobile robot, when given a small, static, and visible environment, can easily determine its location and map its environment; however, dynamic environments, such as (in endoscopy) the interior of a patient’s breathing body, pose a greater challenge. Motion planning is the process of breaking down a movement task into “primitives” such as individual joint movements. Such movement often involves compliant motion, a process where movement requires maintaining physical contact with an object.[116][117] Moravec’s paradox generalizes that low-level sensorimotor skills that humans take for granted are, counterintuitively, difficult to program into a robot; the paradox is named after Hans Moravec, who stated in 1988 that “it is comparatively easy to make computers exhibit adult level performance on intelligence tests or playing checkers, and difficult or impossible to give them the skills of a one-year-old when it comes to perception and mobility”.[118][119] This is attributed to the fact that, unlike checkers, physical dexterity has been a direct target of natural selection for millions of years.[120]

Moravec’s paradox can be extended to many forms of social intelligence.[122][123] Distributed multi-agent coordination of autonomous vehicles remains a difficult problem.[124] Affective computing is an interdisciplinary umbrella that comprises systems which recognize, interpret, process, or simulate human affects. Moderate successes related to affective computing include textual sentiment analysis and, more recently, multimodal affect analysis, wherein AI classifies the affects displayed by a videotaped subject.[128]

In the long run, social skills and an understanding of human emotion and game theory would be valuable to a social agent. Being able to predict the actions of others by understanding their motives and emotional states would allow an agent to make better decisions. Some computer systems mimic human emotion and expressions to appear more sensitive to the emotional dynamics of human interaction, or to otherwise facilitate humancomputer interaction.[129] Similarly, some virtual assistants are programmed to speak conversationally or even to banter humorously; this tends to give naive users an unrealistic conception of how intelligent existing computer agents actually are.[130]

Historically, projects such as the Cyc knowledge base (1984) and the massive Japanese Fifth Generation Computer Systems initiative (19821992) attempted to cover the breadth of human cognition. These early projects failed to escape the limitations of non-quantitative symbolic logic models and, in retrospect, greatly underestimated the difficulty of cross-domain AI. Nowadays, the vast majority of current AI researchers work instead on tractable “narrow AI” applications (such as medical diagnosis or automobile navigation).[131] Many researchers predict that such “narrow AI” work in different individual domains will eventually be incorporated into a machine with artificial general intelligence (AGI), combining most of the narrow skills mentioned in this article and at some point even exceeding human ability in most or all these areas.[17][132] Many advances have general, cross-domain significance. One high-profile example is that DeepMind in the 2010s developed a “generalized artificial intelligence” that could learn many diverse Atari games on its own, and later developed a variant of the system which succeeds at sequential learning.[133][134][135] Besides transfer learning,[136] hypothetical AGI breakthroughs could include the development of reflective architectures that can engage in decision-theoretic metareasoning, and figuring out how to “slurp up” a comprehensive knowledge base from the entire unstructured Web. Some argue that some kind of (currently-undiscovered) conceptually straightforward, but mathematically difficult, “Master Algorithm” could lead to AGI. Finally, a few “emergent” approaches look to simulating human intelligence extremely closely, and believe that anthropomorphic features like an artificial brain or simulated child development may someday reach a critical point where general intelligence emerges.[138][139]

Many of the problems in this article may also require general intelligence, if machines are to solve the problems as well as people do. For example, even specific straightforward tasks, like machine translation, require that a machine read and write in both languages (NLP), follow the author’s argument (reason), know what is being talked about (knowledge), and faithfully reproduce the author’s original intent (social intelligence). A problem like machine translation is considered “AI-complete”, because all of these problems need to be solved simultaneously in order to reach human-level machine performance.

There is no established unifying theory or paradigm that guides AI research. Researchers do not agree about many issues.[140] A few of the most long standing questions that have remained unanswered are these: should artificial intelligence simulate natural intelligence by studying psychology or neurobiology? Or is human biology as irrelevant to AI research as bird biology is to aeronautical engineering?[14] Can intelligent behavior be described using simple, elegant principles (such as logic or optimization)? Or does it necessarily require solving a large number of completely unrelated problems?[15]

In the 1940s and 1950s, a number of researchers explored the connection between neurobiology, information theory, and cybernetics. Some of them built machines that used electronic networks to exhibit rudimentary intelligence, such as W. Grey Walter’s turtles and the Johns Hopkins Beast. Many of these researchers gathered for meetings of the Teleological Society at Princeton University and the Ratio Club in England.[141] By 1960, this approach was largely abandoned, although elements of it would be revived in the 1980s.

When access to digital computers became possible in the middle 1950s, AI research began to explore the possibility that human intelligence could be reduced to symbol manipulation. The research was centered in three institutions: Carnegie Mellon University, Stanford and MIT, and as described below, each one developed its own style of research. John Haugeland named these symbolic approaches to AI “good old fashioned AI” or “GOFAI”.[142] During the 1960s, symbolic approaches had achieved great success at simulating high-level thinking in small demonstration programs. Approaches based on cybernetics or neural networks were abandoned or pushed into the background.[143] Researchers in the 1960s and the 1970s were convinced that symbolic approaches would eventually succeed in creating a machine with artificial general intelligence and considered this the goal of their field.

Economist Herbert Simon and Allen Newell studied human problem-solving skills and attempted to formalize them, and their work laid the foundations of the field of artificial intelligence, as well as cognitive science, operations research and management science. Their research team used the results of psychological experiments to develop programs that simulated the techniques that people used to solve problems. This tradition, centered at Carnegie Mellon University would eventually culminate in the development of the Soar architecture in the middle 1980s.[144][145]

Unlike Simon and Newell, John McCarthy felt that machines did not need to simulate human thought, but should instead try to find the essence of abstract reasoning and problem solving, regardless of whether people used the same algorithms.[14] His laboratory at Stanford (SAIL) focused on using formal logic to solve a wide variety of problems, including knowledge representation, planning and learning.[146] Logic was also the focus of the work at the University of Edinburgh and elsewhere in Europe which led to the development of the programming language Prolog and the science of logic programming.[147]

Researchers at MIT (such as Marvin Minsky and Seymour Papert)[148] found that solving difficult problems in vision and natural language processing required ad-hoc solutions they argued that there was no simple and general principle (like logic) that would capture all the aspects of intelligent behavior. Roger Schank described their “anti-logic” approaches as “scruffy” (as opposed to the “neat” paradigms at CMU and Stanford).[15] Commonsense knowledge bases (such as Doug Lenat’s Cyc) are an example of “scruffy” AI, since they must be built by hand, one complicated concept at a time.[149]

When computers with large memories became available around 1970, researchers from all three traditions began to build knowledge into AI applications.[150] This “knowledge revolution” led to the development and deployment of expert systems (introduced by Edward Feigenbaum), the first truly successful form of AI software.[37] The knowledge revolution was also driven by the realization that enormous amounts of knowledge would be required by many simple AI applications.

By the 1980s progress in symbolic AI seemed to stall and many believed that symbolic systems would never be able to imitate all the processes of human cognition, especially perception, robotics, learning and pattern recognition. A number of researchers began to look into “sub-symbolic” approaches to specific AI problems.[16] Sub-symbolic methods manage to approach intelligence without specific representations of knowledge.

This includes embodied, situated, behavior-based, and nouvelle AI. Researchers from the related field of robotics, such as Rodney Brooks, rejected symbolic AI and focused on the basic engineering problems that would allow robots to move and survive.[151] Their work revived the non-symbolic viewpoint of the early cybernetics researchers of the 1950s and reintroduced the use of control theory in AI. This coincided with the development of the embodied mind thesis in the related field of cognitive science: the idea that aspects of the body (such as movement, perception and visualization) are required for higher intelligence.

Within developmental robotics, developmental learning approaches are elaborated upon to allow robots to accumulate repertoires of novel skills through autonomous self-exploration, social interaction with human teachers, and the use of guidance mechanisms (active learning, maturation, motor synergies, etc.).[152][153]

Interest in neural networks and “connectionism” was revived by David Rumelhart and others in the middle of the 1980s.[156] Neural networks are an example of soft computing — they are solutions to problems which cannot be solved with complete logical certainty, and where an approximate solution is often sufficient. Other soft computing approaches to AI include fuzzy systems, evolutionary computation and many statistical tools. The application of soft computing to AI is studied collectively by the emerging discipline of computational intelligence.[157]

Much of GOFAI got bogged down on ad hoc patches to symbolic computation that worked on their own toy models but failed to generalize to real-world results. However, around the 1990s, AI researchers adopted sophisticated mathematical tools, such as hidden Markov models (HMM), information theory, and normative Bayesian decision theory to compare or to unify competing architectures. The shared mathematical language permitted a high level of collaboration with more established fields (like mathematics, economics or operations research).[d] Compared with GOFAI, new “statistical learning” techniques such as HMM and neural networks were gaining higher levels of accuracy in many practical domains such as data mining, without necessarily acquiring semantic understanding of the datasets. The increased successes with real-world data led to increasing emphasis on comparing different approaches against shared test data to see which approach performed best in a broader context than that provided by idiosyncratic toy models; AI research was becoming more scientific. Nowadays results of experiments are often rigorously measurable, and are sometimes (with difficulty) reproducible.[38][158] Different statistical learning techniques have different limitations; for example, basic HMM cannot model the infinite possible combinations of natural language. Critics note that the shift from GOFAI to statistical learning is often also a shift away from Explainable AI. In AGI research, some scholars caution against over-reliance on statistical learning, and argue that continuing research into GOFAI will still be necessary to attain general intelligence.

AI has developed a large number of tools to solve the most difficult problems in computer science. A few of the most general of these methods are discussed below.

Many problems in AI can be solved in theory by intelligently searching through many possible solutions:[167] Reasoning can be reduced to performing a search. For example, logical proof can be viewed as searching for a path that leads from premises to conclusions, where each step is the application of an inference rule.[168] Planning algorithms search through trees of goals and subgoals, attempting to find a path to a target goal, a process called means-ends analysis.[169] Robotics algorithms for moving limbs and grasping objects use local searches in configuration space.[114] Many learning algorithms use search algorithms based on optimization.

Simple exhaustive searches[170] are rarely sufficient for most real world problems: the search space (the number of places to search) quickly grows to astronomical numbers. The result is a search that is too slow or never completes. The solution, for many problems, is to use “heuristics” or “rules of thumb” that prioritize choices in favor of those that are more likely to reach a goal, and to do so in a shorter number of steps. In some search methodologies heuristics can also serve to entirely eliminate some choices that are unlikely to lead to a goal (called “pruning the search tree”). Heuristics supply the program with a “best guess” for the path on which the solution lies.[171] Heuristics limit the search for solutions into a smaller sample size.

A very different kind of search came to prominence in the 1990s, based on the mathematical theory of optimization. For many problems, it is possible to begin the search with some form of a guess and then refine the guess incrementally until no more refinements can be made. These algorithms can be visualized as blind hill climbing: we begin the search at a random point on the landscape, and then, by jumps or steps, we keep moving our guess uphill, until we reach the top. Other optimization algorithms are simulated annealing, beam search and random optimization.[172]

Evolutionary computation uses a form of optimization search. For example, they may begin with a population of organisms (the guesses) and then allow them to mutate and recombine, selecting only the fittest to survive each generation (refining the guesses). Classic evolutionary algorithms include genetic algorithms, gene expression programming, and genetic programming.[173] Alternatively, distributed search processes can coordinate via swarm intelligence algorithms. Two popular swarm algorithms used in search are particle swarm optimization (inspired by bird flocking) and ant colony optimization (inspired by ant trails).[174][175]

Logic[176] is used for knowledge representation and problem solving, but it can be applied to other problems as well. For example, the satplan algorithm uses logic for planning[177] and inductive logic programming is a method for learning.[178]

Several different forms of logic are used in AI research. Propositional logic[179] involves truth functions such as “or” and “not”. First-order logic[180] adds quantifiers and predicates, and can express facts about objects, their properties, and their relations with each other. Fuzzy set theory assigns a “degree of truth” (between 0 and 1) to vague statements such as “Alice is old” (or rich, or tall, or hungry) that are too linguistically imprecise to be completely true or false. Fuzzy logic is successfully used in control systems to allow experts to contribute vague rules such as “if you are close to the destination station and moving fast, increase the train’s brake pressure”; these vague rules can then be numerically refined within the system. Fuzzy logic fails to scale well in knowledge bases; many AI researchers question the validity of chaining fuzzy-logic inferences.[e][182][183]

Default logics, non-monotonic logics and circumscription[90] are forms of logic designed to help with default reasoning and the qualification problem. Several extensions of logic have been designed to handle specific domains of knowledge, such as: description logics;[78] situation calculus, event calculus and fluent calculus (for representing events and time);[79] causal calculus;[80] belief calculus;[184] and modal logics.[81]

Overall, qualitiative symbolic logic is brittle and scales poorly in the presence of noise or other uncertainty. Exceptions to rules are numerous, and it is difficult for logical systems to function in the presence of contradictory rules.[186]

Many problems in AI (in reasoning, planning, learning, perception and robotics) require the agent to operate with incomplete or uncertain information. AI researchers have devised a number of powerful tools to solve these problems using methods from probability theory and economics.[187]

Bayesian networks[188] are a very general tool that can be used for a large number of problems: reasoning (using the Bayesian inference algorithm),[189] learning (using the expectation-maximization algorithm),[f][191] planning (using decision networks)[192] and perception (using dynamic Bayesian networks).[193] Probabilistic algorithms can also be used for filtering, prediction, smoothing and finding explanations for streams of data, helping perception systems to analyze processes that occur over time (e.g., hidden Markov models or Kalman filters).[193] Compared with symbolic logic, formal Bayesian inference is computationally expensive. For inference to be tractable, most observations must be conditionally independent of one another. Complicated graphs with diamonds or other “loops” (undirected cycles) can require a sophisticated method such as Markov Chain Monte Carlo, which spreads an ensemble of random walkers throughout the Bayesian network and attempts to converge to an assessment of the conditional probabilities. Bayesian networks are used on XBox Live to rate and match players; wins and losses are “evidence” of how good a player is. AdSense uses a Bayesian network with over 300 million edges to learn which ads to serve.

A key concept from the science of economics is “utility”: a measure of how valuable something is to an intelligent agent. Precise mathematical tools have been developed that analyze how an agent can make choices and plan, using decision theory, decision analysis,[195] and information value theory.[96] These tools include models such as Markov decision processes,[196] dynamic decision networks,[193] game theory and mechanism design.[197]

The simplest AI applications can be divided into two types: classifiers (“if shiny then diamond”) and controllers (“if shiny then pick up”). Controllers do, however, also classify conditions before inferring actions, and therefore classification forms a central part of many AI systems. Classifiers are functions that use pattern matching to determine a closest match. They can be tuned according to examples, making them very attractive for use in AI. These examples are known as observations or patterns. In supervised learning, each pattern belongs to a certain predefined class. A class can be seen as a decision that has to be made. All the observations combined with their class labels are known as a data set. When a new observation is received, that observation is classified based on previous experience.[198]

A classifier can be trained in various ways; there are many statistical and machine learning approaches. The decision tree[199] is perhaps the most widely used machine learning algorithm. Other widely used classifiers are the neural network,[201] k-nearest neighbor algorithm,[g][203] kernel methods such as the support vector machine (SVM),[h][205] Gaussian mixture model[206] and the extremely popular naive Bayes classifier.[i][208] Classifier performance depends greatly on the characteristics of the data to be classified, such as the dataset size, the dimensionality, and the level of noise. Model-based classifiers perform well if the assumed model is an extremely good fit for the actual data. Otherwise, if no matching model is available, and if accuracy (rather than speed or scalability) is the sole concern, conventional wisdom is that discriminative classifiers (especially SVM) tend to be more accurate than model-based classifiers such as “naive Bayes” on most practical data sets.[209]

Neural networks, or neural nets, were inspired by the architecture of neurons in the human brain. A simple “neuron” N accepts input from multiple other neurons, each of which, when activated (or “fired”), cast a weighted “vote” for or against whether neuron N should itself activate. Learning requires an algorithm to adjust these weights based on the training data; one simple algorithm (dubbed “fire together, wire together”) is to increase the weight between two connected neurons when the activation of one triggers the successful activation of another. The net forms “concepts” that are distributed among a subnetwork of shared[j] neurons that tend to fire together; a concept meaning “leg” might be coupled with a subnetwork meaning “foot” that includes the sound for “foot”. Neurons have a continuous spectrum of activation; in addition, neurons can process inputs in a nonlinear way rather than weighing straightforward votes. Modern neural nets can learn both continuous functions and, surprisingly, digital logical operations. Neural networks’ early successes included predicting the stock market and (in 1995) a mostly self-driving car.[k] In the 2010s, advances in neural networks using deep learning thrust AI into widespread public consciousness and contributed to an enormous upshift in corporate AI spending; for example, AI-related M&A in 2017 was over 25 times as large as in 2015.[212][213]

The study of non-learning artificial neural networks[201] began in the decade before the field of AI research was founded, in the work of Walter Pitts and Warren McCullouch. Frank Rosenblatt invented the perceptron, a learning network with a single layer, similar to the old concept of linear regression. Early pioneers also include Alexey Grigorevich Ivakhnenko, Teuvo Kohonen, Stephen Grossberg, Kunihiko Fukushima, Christoph von der Malsburg, David Willshaw, Shun-Ichi Amari, Bernard Widrow, John Hopfield, Eduardo R. Caianiello, and others.

The main categories of networks are acyclic or feedforward neural networks (where the signal passes in only one direction) and recurrent neural networks (which allow feedback and short-term memories of previous input events). Among the most popular feedforward networks are perceptrons, multi-layer perceptrons and radial basis networks.[214] Neural networks can be applied to the problem of intelligent control (for robotics) or learning, using such techniques as Hebbian learning (“fire together, wire together”), GMDH or competitive learning.[215]

Today, neural networks are often trained by the backpropagation algorithm, which had been around since 1970 as the reverse mode of automatic differentiation published by Seppo Linnainmaa,[216][217] and was introduced to neural networks by Paul Werbos.[218][219][220]

Hierarchical temporal memory is an approach that models some of the structural and algorithmic properties of the neocortex.[221]

In short, most neural networks use some form of gradient descent on a hand-created neural topology. However, some research groups, such as Uber, argue that simple neuroevolution to mutate new neural network topologies and weights may be competitive with sophisticated gradient descent approaches. One advantage of neuroevolution is that it may be less prone to get caught in “dead ends”.[222]

Deep learning is any artificial neural network that can learn a long chain of causal links. For example, a feedforward network with six hidden layers can learn a seven-link causal chain (six hidden layers + output layer) and has a “credit assignment path” (CAP) depth of seven. Many deep learning systems need to be able to learn chains ten or more causal links in length.[223] Deep learning has transformed many important subfields of artificial intelligence, including computer vision, speech recognition, natural language processing and others.[224][225][223]

According to one overview,[226] the expression “Deep Learning” was introduced to the Machine Learning community by Rina Dechter in 1986[227] and gained traction after Igor Aizenberg and colleagues introduced it to Artificial Neural Networks in 2000.[228] The first functional Deep Learning networks were published by Alexey Grigorevich Ivakhnenko and V. G. Lapa in 1965.[229][pageneeded] These networks are trained one layer at a time. Ivakhnenko’s 1971 paper[230] describes the learning of a deep feedforward multilayer perceptron with eight layers, already much deeper than many later networks. In 2006, a publication by Geoffrey Hinton and Ruslan Salakhutdinov introduced another way of pre-training many-layered feedforward neural networks (FNNs) one layer at a time, treating each layer in turn as an unsupervised restricted Boltzmann machine, then using supervised backpropagation for fine-tuning. Similar to shallow artificial neural networks, deep neural networks can model complex non-linear relationships. Over the last few years, advances in both machine learning algorithms and computer hardware have led to more efficient methods for training deep neural networks that contain many layers of non-linear hidden units and a very large output layer.[232]

Deep learning often uses convolutional neural networks (CNNs), whose origins can be traced back to the Neocognitron introduced by Kunihiko Fukushima in 1980.[233] In 1989, Yann LeCun and colleagues applied backpropagation to such an architecture. In the early 2000s, in an industrial application CNNs already processed an estimated 10% to 20% of all the checks written in the US.[234] Since 2011, fast implementations of CNNs on GPUs have won many visual pattern recognition competitions.[223]

CNNs with 12 convolutional layers were used in conjunction with reinforcement learning by Deepmind’s “AlphaGo Lee”, the program that beat a top Go champion in 2016.[235]

Early on, deep learning was also applied to sequence learning with recurrent neural networks (RNNs)[236] which are in theory Turing complete[237] and can run arbitrary programs to process arbitrary sequences of inputs. The depth of an RNN is unlimited and depends on the length of its input sequence; thus, an RNN is an example of deep learning.[223] RNNs can be trained by gradient descent[238][239][240] but suffer from the vanishing gradient problem.[224][241] In 1992, it was shown that unsupervised pre-training of a stack of recurrent neural networks can speed up subsequent supervised learning of deep sequential problems.[242]

Numerous researchers now use variants of a deep learning recurrent NN called the long short-term memory (LSTM) network published by Hochreiter & Schmidhuber in 1997.[243] LSTM is often trained by Connectionist Temporal Classification (CTC).[244] At Google, Microsoft and Baidu this approach has revolutionised speech recognition.[245][246][247] For example, in 2015, Google’s speech recognition experienced a dramatic performance jump of 49% through CTC-trained LSTM, which is now available through Google Voice to billions of smartphone users.[248] Google also used LSTM to improve machine translation,[249] Language Modeling[250] and Multilingual Language Processing.[251] LSTM combined with CNNs also improved automatic image captioning[252] and a plethora of other applications.

AI, like electricity or the steam engine, is a general purpose technology. There is no consensus on how to characterize which tasks AI tends to excel at.[253] While projects such as AlphaZero have succeeded in generating their own knowledge from scratch, many other machine learning projects require large training datasets.[254][255] Researcher Andrew Ng has suggested, as a “highly imperfect rule of thumb”, that “almost anything a typical human can do with less than one second of mental thought, we can probably now or in the near future automate using AI.”[256] Moravec’s paradox suggests that AI lags humans at many tasks that the human brain has specifically evolved to perform well.[257]

Games provide a well-publicized benchmark for assessing rates of progress. AlphaGo around 2016 brought the era of classical board-game benchmarks to an close. Games of imperfect knowledge provide new challenges to AI in the area of game theory.[258][259] E-sports such as StarCraft continue to provide additional public benchmarks.[260][261] There are many competitions and prizes, such as the Imagenet Challenge, to promote research in artificial intelligence. The main areas of competition include general machine intelligence, conversational behavior, data-mining, robotic cars, and robot soccer as well as conventional games.[citation needed]

The “imitation game” (an interpretation of the 1950 Turing test that assesses whether a computer can imitate a human) is nowadays considered too exploitable to be a meaningful benchmark.[262] A derivative of the Turing test is the Completely Automated Public Turing test to tell Computers and Humans Apart (CAPTCHA). As the name implies, this helps to determine that a user is an actual person and not a computer posing as a human. In contrast to the standard Turing test, CAPTCHA is administered by a machine and targeted to a human as opposed to being administered by a human and targeted to a machine. A computer asks a user to complete a simple test then generates a grade for that test. Computers are unable to solve the problem, so correct solutions are deemed to be the result of a person taking the test. A common type of CAPTCHA is the test that requires the typing of distorted letters, numbers or symbols that appear in an image undecipherable by a computer.

Proposed “universal intelligence” tests aim to compare how well machines, humans, and even non-human animals perform on problem sets that are generic as possible. At an extreme, the test suite can contain every possible problem, weighted by Kolmogorov complexity; unfortunately, these problem sets tend to be dominated by impoverished pattern-matching exercises where a tuned AI can easily exceed human performance levels.[264][265]

AI is relevant to any intellectual task. Modern artificial intelligence techniques are pervasive and are too numerous to list here. Frequently, when a technique reaches mainstream use, it is no longer considered artificial intelligence; this phenomenon is described as the AI effect.

High-profile examples of AI include autonomous vehicles (such as drones and self-driving cars), medical diagnosis, creating art (such as poetry), proving mathematical theorems, playing games (such as Chess or Go), search engines (such as Google search), online assistants (such as Siri), image recognition in photographs, spam filtering, prediction of judicial decisions[268] and targeting online advertisements.[269][270]

With social media sites overtaking TV as a source for news for young people and news organisations increasingly reliant on social media platforms for generating distribution,[271] major publishers now use artificial intelligence (AI) technology to post stories more effectively and generate higher volumes of traffic.[272]

Artificial intelligence is breaking into the healthcare industry by assisting doctors. According to Bloomberg Technology, Microsoft has developed AI to help doctors find the right treatments for cancer.[273] There is a great amount of research and drugs developed relating to cancer. In detail, there are more than 800 medicines and vaccines to treat cancer. This negatively affects the doctors, because there are too many options to choose from, making it more difficult to choose the right drugs for the patients. Microsoft is working on a project to develop a machine called “Hanover”. Its goal is to memorize all the papers necessary to cancer and help predict which combinations of drugs will be most effective for each patient. One project that is being worked on at the moment is fighting myeloid leukemia, a fatal cancer where the treatment has not improved in decades. Another study was reported to have found that artificial intelligence was as good as trained doctors in identifying skin cancers.[274] Another study is using artificial intelligence to try and monitor multiple high-risk patients, and this is done by asking each patient numerous questions based on data acquired from live doctor to patient interactions.[275]

According to CNN, there was a recent study by surgeons at the Children’s National Medical Center in Washington which successfully demonstrated surgery with an autonomous robot. The team supervised the robot while it performed soft-tissue surgery, stitching together a pig’s bowel during open surgery, and doing so better than a human surgeon, the team claimed.[276] IBM has created its own artificial intelligence computer, the IBM Watson, which has beaten human intelligence (at some levels). Watson not only won at the game show Jeopardy! against former champions,[277] but, was declared a hero after successfully diagnosing a women who was suffering from leukemia.[278]

Advancements in AI have contributed to the growth of the automotive industry through the creation and evolution of self-driving vehicles. As of 2016, there are over 30 companies utilizing AI into the creation of driverless cars. A few companies involved with AI include Tesla, Google, and Apple.[279]

Many components contribute to the functioning of self-driving cars. These vehicles incorporate systems such as braking, lane changing, collision prevention, navigation and mapping. Together, these systems, as well as high performance computers, are integrated into one complex vehicle.[280]

Recent developments in autonomous automobiles have made the innovation of self-driving trucks possible, though they are still in the testing phase. The UK government has passed legislation to begin testing of self-driving truck platoons in 2018.[281] Self-driving truck platoons are a fleet of self-driving trucks following the lead of one non-self-driving truck, so the truck platoons aren’t entirely autonomous yet. Meanwhile, the Daimler, a German automobile corporation, is testing the Freightliner Inspiration which is a semi-autonomous truck that will only be used on the highway.[282]

One main factor that influences the ability for a driver-less automobile to function is mapping. In general, the vehicle would be pre-programmed with a map of the area being driven. This map would include data on the approximations of street light and curb heights in order for the vehicle to be aware of its surroundings. However, Google has been working on an algorithm with the purpose of eliminating the need for pre-programmed maps and instead, creating a device that would be able to adjust to a variety of new surroundings.[283] Some self-driving cars are not equipped with steering wheels or brake pedals, so there has also been research focused on creating an algorithm that is capable of maintaining a safe environment for the passengers in the vehicle through awareness of speed and driving conditions.[284]

Another factor that is influencing the ability for a driver-less automobile is the safety of the passenger. To make a driver-less automobile, engineers must program it to handle high risk situations. These situations could include a head on collision with pedestrians. The car’s main goal should be to make a decision that would avoid hitting the pedestrians and saving the passengers in the car. But there is a possibility the car would need to make a decision that would put someone in danger. In other words, the car would need to decide to save the pedestrians or the passengers.[285] The programing of the car in these situations is crucial to a successful driver-less automobile.

Financial institutions have long used artificial neural network systems to detect charges or claims outside of the norm, flagging these for human investigation. The use of AI in banking can be traced back to 1987 when Security Pacific National Bank in US set-up a Fraud Prevention Task force to counter the unauthorised use of debit cards. Programs like Kasisto and Moneystream are using AI in financial services.

Banks use artificial intelligence systems today to organize operations, maintain book-keeping, invest in stocks, and manage properties. AI can react to changes overnight or when business is not taking place.[286] In August 2001, robots beat humans in a simulated financial trading competition.[287] AI has also reduced fraud and financial crimes by monitoring behavioral patterns of users for any abnormal changes or anomalies.[288]

The use of AI machines in the market in applications such as online trading and decision making has changed major economic theories.[289] For example, AI based buying and selling platforms have changed the law of supply and demand in that it is now possible to easily estimate individualized demand and supply curves and thus individualized pricing. Furthermore, AI machines reduce information asymmetry in the market and thus making markets more efficient while reducing the volume of trades. Furthermore, AI in the markets limits the consequences of behavior in the markets again making markets more efficient. Other theories where AI has had impact include in rational choice, rational expectations, game theory, Lewis turning point, portfolio optimization and counterfactual thinking.

In video games, artificial intelligence is routinely used to generate dynamic purposeful behavior in non-player characters (NPCs). In addition, well-understood AI techniques are routinely used for pathfinding. Some researchers consider NPC AI in games to be a “solved problem” for most production tasks. Games with more atypical AI include the AI director of Left 4 Dead (2008) and the neuroevolutionary training of platoons in Supreme Commander 2 (2010).[290][291] In 2018 researchers at Cornwall University have done something that will fundamentally change the process of developing new video games. They created a pair of competing neural networks (Generative Adversarial Network, GAN) and trained them on the example of the very first game-joker, DOOM. During the training, neural networks defined the basic principles of building the levels of this game and after that they became able to generate new levels without the slightest help from the people[292].

Worldwide annual military spending on robotics rose from 5.1 billion USD in 2010 to 7.5 billion USD in 2015.[293][294] Military drones capable of autonomous action are widely considered a useful asset. In 2017, Vladimir Putin stated that “Whoever becomes the leader in (artificial intelligence) will become the ruler of the world”.[295][296] Many artificial intelligence researchers seek to distance themselves from military applications of AI.[297]

For financial statements audit, AI makes continuous audit possible. AI tools could analyze many sets of different information immediately. The potential benefit would be the overall audit risk will be reduced, the level of assurance will be increased and the time duration of audit will be reduced.[298]

There are three philosophical questions related to AI:

Can a machine be intelligent? Can it “think”?

Widespread use of artificial intelligence could have unintended consequences that are dangerous or undesirable. Scientists from the Future of Life Institute, among others, described some short-term research goals to see how AI influences the economy, the laws and ethics that are involved with AI and how to minimize AI security risks. In the long-term, the scientists have proposed to continue optimizing function while minimizing possible security risks that come along with new technologies.[309]

Machines with intelligence have the potential to use their intelligence to make ethical decisions. Research in this area includes “machine ethics”, “artificial moral agents”, and the study of “malevolent vs. friendly AI”.

The development of full artificial intelligence could spell the end of the human race. Once humans develop artificial intelligence, it will take off on its own and redesign itself at an ever-increasing rate. Humans, who are limited by slow biological evolution, couldn’t compete and would be superseded.

Read more:

Artificial intelligence – Wikipedia

Benefits & Risks of Artificial Intelligence – Future of …

Many AI researchers roll their eyes when seeing this headline:Stephen Hawking warns that rise of robots may be disastrous for mankind. And as many havelost count of how many similar articles theyveseen.Typically, these articles are accompanied by an evil-looking robot carrying a weapon, and they suggest we should worry about robots rising up and killing us because theyve become conscious and/or evil.On a lighter note, such articles are actually rather impressive, because they succinctly summarize the scenario that AI researchers dontworry about. That scenario combines as many as three separate misconceptions: concern about consciousness, evil, androbots.

If you drive down the road, you have a subjective experience of colors, sounds, etc. But does a self-driving car have a subjective experience? Does it feel like anything at all to be a self-driving car?Although this mystery of consciousness is interesting in its own right, its irrelevant to AI risk. If you get struck by a driverless car, it makes no difference to you whether it subjectively feels conscious. In the same way, what will affect us humans is what superintelligent AIdoes, not how it subjectively feels.

The fear of machines turning evil is another red herring. The real worry isnt malevolence, but competence. A superintelligent AI is by definition very good at attaining its goals, whatever they may be, so we need to ensure that its goals are aligned with ours. Humans dont generally hate ants, but were more intelligent than they are so if we want to build a hydroelectric dam and theres an anthill there, too bad for the ants. The beneficial-AI movement wants to avoid placing humanity in the position of those ants.

The consciousness misconception is related to the myth that machines cant have goals.Machines can obviously have goals in the narrow sense of exhibiting goal-oriented behavior: the behavior of a heat-seeking missile is most economically explained as a goal to hit a target.If you feel threatened by a machine whose goals are misaligned with yours, then it is precisely its goals in this narrow sense that troubles you, not whether the machine is conscious and experiences a sense of purpose.If that heat-seeking missile were chasing you, you probably wouldnt exclaim: Im not worried, because machines cant have goals!

I sympathize with Rodney Brooks and other robotics pioneers who feel unfairly demonized by scaremongering tabloids,because some journalists seem obsessively fixated on robots and adorn many of their articles with evil-looking metal monsters with red shiny eyes. In fact, the main concern of the beneficial-AI movement isnt with robots but with intelligence itself: specifically, intelligence whose goals are misaligned with ours. To cause us trouble, such misaligned superhuman intelligence needs no robotic body, merely an internet connection this may enable outsmarting financial markets, out-inventing human researchers, out-manipulating human leaders, and developing weapons we cannot even understand. Even if building robots were physically impossible, a super-intelligent and super-wealthy AI could easily pay or manipulate many humans to unwittingly do its bidding.

The robot misconception is related to the myth that machines cant control humans. Intelligence enables control: humans control tigers not because we are stronger, but because we are smarter. This means that if we cede our position as smartest on our planet, its possible that we might also cede control.

Continued here:

Benefits & Risks of Artificial Intelligence – Future of …

What is Artificial Intelligence (AI)? – Definition from …

Artificial intelligence is a branch of computer science that aims to create intelligent machines. It has become an essential part of the technology industry.

Research associated with artificial intelligence is highly technical and specialized. The core problems of artificial intelligence include programming computers for certain traits such as:

Knowledge engineering is a core part of AI research. Machines can often act and react like humans only if they have abundant information relating to the world. Artificial intelligence must have access to objects, categories, properties and relations between all of them to implement knowledge engineering. Initiating common sense, reasoning and problem-solving power in machines is a difficult and tedious task.

Machine learning is also a core part of AI. Learning without any kind of supervision requires an ability to identify patterns in streams of inputs, whereas learning with adequate supervision involves classification and numerical regressions. Classification determines the category an object belongs to and regression deals with obtaining a set of numerical input or output examples, thereby discovering functions enabling the generation of suitable outputs from respective inputs. Mathematical analysis of machine learning algorithms and their performance is a well-defined branch of theoretical computer science often referred to as computational learning theory.

Machine perception deals with the capability to use sensory inputs to deduce the different aspects of the world, while computer vision is the power to analyze visual inputs with a few sub-problems such as facial, object and gesture recognition.

Robotics is also a major field related to AI. Robots require intelligence to handle tasks such as object manipulation and navigation, along with sub-problems of localization, motion planning and mapping.

Read more:

What is Artificial Intelligence (AI)? – Definition from …

Hedonism | Internet Encyclopedia of Philosophy

The term “hedonism,” from the Greek word (hdon) for pleasure, refers to several related theories about what is good for us, how we should behave, and what motivates us to behave in the way that we do. All hedonistic theories identify pleasure and pain as the only important elements of whatever phenomena they are designed to describe. If hedonistic theories identified pleasure and pain as merely two important elements, instead of the only important elements of what they are describing, then they would not be nearly as unpopular as they all are. However, the claim that pleasure and pain are the only things of ultimate importance is what makes hedonism distinctive and philosophically interesting.

Philosophical hedonists tend to focus on hedonistic theories of value, and especially of well-being (the good life for the one living it). As a theory of value, hedonism states that all and only pleasure is intrinsically valuable and all and only pain is intrinsically not valuable. Hedonists usually define pleasure and pain broadly, such that both physical and mental phenomena are included. Thus, a gentle massage and recalling a fond memory are both considered to cause pleasure and stubbing a toe and hearing about the death of a loved one are both considered to cause pain. With pleasure and pain so defined, hedonism as a theory about what is valuable for us is intuitively appealing. Indeed, its appeal is evidenced by the fact that nearly all historical and contemporary treatments of well-being allocate at least some space for discussion of hedonism. Unfortunately for hedonism, the discussions rarely endorse it and some even deplore its focus on pleasure.

This article begins by clarifying the different types of hedonistic theories and the labels they are often given. Then, hedonisms ancient origins and its subsequent development are reviewed. The majority of this article is concerned with describing the important theoretical divisions within Prudential Hedonism and discussing the major criticisms of these approaches.

When the term “hedonism” is used in modern literature, or by non-philosophers in their everyday talk, its meaning is quite different from the meaning it takes when used in the discussions of philosophers. Non-philosophers tend to think of a hedonist as a person who seeks out pleasure for themselves without any particular regard for their own future well-being or for the well-being of others. According to non-philosophers, then, a stereotypical hedonist is someone who never misses an opportunity to indulge of the pleasures of sex, drugs, and rock n roll, even if the indulgences are likely to lead to relationship problems, health problems, regrets, or sadness for themselves or others. Philosophers commonly refer to this everyday understanding of hedonism as “Folk Hedonism.” Folk Hedonism is a rough combination of Motivational Hedonism, Hedonistic Egoism, and a reckless lack of foresight.

When philosophers discuss hedonism, they are most likely to be referring to hedonism about value, and especially the slightly more specific theory, hedonism about well-being. Hedonism as a theory about value (best referred to as Value Hedonism) holds that all and only pleasure is intrinsically valuable and all and only pain is intrinsically disvaluable. The term “intrinsically” is an important part of the definition and is best understood in contrast to the term “instrumentally.” Something is intrinsically valuable if it is valuable for its own sake. Pleasure is thought to be intrinsically valuable because, even if it did not lead to any other benefit, it would still be good to experience. Money is an example of an instrumental good; its value for us comes from what we can do with it (what we can buy with it). The fact that a copious amount of money has no value if no one ever sells anything reveals that money lacks intrinsic value. Value Hedonism reduces everything of value to pleasure. For example, a Value Hedonist would explain the instrumental value of money by describing how the things we can buy with money, such as food, shelter, and status-signifying goods, bring us pleasure or help us to avoid pain.

Hedonism as a theory about well-being (best referred to as Prudential Hedonism) is more specific than Value Hedonism because it stipulates what the value is for. Prudential Hedonism holds that all and only pleasure intrinsically makes peoples lives go better for them and all and only pain intrinsically makes their lives go worse for them. Some philosophers replace “people” with “animals” or “sentient creatures,” so as to apply Prudential Hedonism more widely. A good example of this comes from Peter Singers work on animals and ethics. Singer questions why some humans can see the intrinsic disvalue in human pain, but do not also accept that it is bad for sentient non-human animals to experience pain.

When Prudential Hedonists claim that happiness is what they value most, they intend happiness to be understood as a preponderance of pleasure over pain. An important distinction between Prudential Hedonism and Folk Hedonism is that Prudential Hedonists usually understand that pursuing pleasure and avoiding pain in the very short-term is not always the best strategy for achieving the best long-term balance of pleasure over pain.

Prudential Hedonism is an integral part of several derivative types of hedonistic theory, all of which have featured prominently in philosophical debates of the past. Since Prudential Hedonism plays this important role, the majority of this article is dedicated to Prudential Hedonism. First, however, the main derivative types of hedonism are briefly discussed.

Motivational Hedonism (more commonly referred to by the less descriptive label, “Psychological Hedonism”) is the theory that the desires to encounter pleasure and to avoid pain guide all of our behavior. Most accounts of Motivational Hedonism include both conscious and unconscious desires for pleasure, but emphasize the latter. Epicurus, William James, Sigmund Freud, Jeremy Bentham, John Stuart Mill, and (on one interpretation) even Charles Darwin have all argued for varieties of Motivational Hedonism. Bentham used the idea to support his theory of Hedonistic Utilitarianism (discussed below). Weak versions of Motivational Hedonism hold that the desires to seek pleasure and avoid pain often or always have some influence on our behavior. Weak versions are generally considered to be uncontroversially true and not especially useful for philosophy.

Philosophers have been more interested in strong accounts of Motivational Hedonism, which hold that all behavior is governed by the desires to encounter pleasure and to avoid pain (and only those desires). Strong accounts of Motivational Hedonism have been used to support some of the normative types of hedonism and to argue against non-hedonistic normative theories. One of the most notable mentions of Motivational Hedonism is Platos Ring of Gyges example in The Republic. Platos Socrates is discussing with Glaucon how men would react if they were to possess a ring that gives its wearer immense powers, including invisibility. Glaucon believes that a strong version of Motivational Hedonism is true, but Socrates does not. Glaucon asserts that, emboldened with the power provided by the Ring of Gyges, everyone would succumb to the inherent and ubiquitous desire to pursue their own ends at the expense of others. Socrates disagrees, arguing that good people would be able to overcome this desire because of their strong love of justice, fostered through philosophising.

Strong accounts of Motivational Hedonism currently garner very little support for similar reasons. Many examples of seemingly-pain-seeking acts performed out of a sense of duty are well-known from the soldier who jumps on a grenade to save his comrades to that time you rescued a trapped dog only to be (predictably) bitten in the process. Introspective evidence also weighs against strong accounts of Motivational Hedonism; many of the decisions we make seem to be based on motives other than seeking pleasure and avoiding pain. Given these reasons, the burden of proof is considered to be squarely on the shoulders of anyone wishing to argue for a strong account of Motivational Hedonism.

Value Hedonism, occasionally with assistance from Motivational Hedonism, has been used to argue for specific theories of right action (theories that explain which actions are morally permissible or impermissible and why). The theory that happiness should be pursued (that pleasure should be pursued and pain should be avoided) is referred to as Normative Hedonism and sometimes Ethical Hedonism. There are two major types of Normative Hedonism, Hedonistic Egoism and Hedonistic Utilitarianism. Both types commonly use happiness (defined as pleasure minus pain) as the sole criterion for determining the moral rightness or wrongness of an action. Important variations within each of these two main types specify either the actual resulting happiness (after the act) or the predicted resulting happiness (before the act) as the moral criterion. Although both major types of Normative Hedonism have been accused of being repugnant, Hedonistic Egoism is considered the most offensive.

Hedonistic Egoism is a hedonistic version of egoism, the theory that we should, morally speaking, do whatever is most in our own interests. Hedonistic Egoism is the theory that we ought, morally speaking, to do whatever makes us happiest that is whatever provides us with the most net pleasure after pain is subtracted. The most repugnant feature of this theory is that one never has to ascribe any value whatsoever to the consequences for anyone other than oneself. For example, a Hedonistic Egoist who did not feel saddened by theft would be morally required to steal, even from needy orphans (if he thought he could get away with it). Would-be defenders of Hedonistic Egoism often point out that performing acts of theft, murder, treachery and the like would not make them happier overall because of the guilt, the fear of being caught, and the chance of being caught and punished. The would-be defenders tend to surrender, however, when it is pointed out that a Hedonistic Egoist is morally obliged by their own theory to pursue an unusual kind of practical education; a brief and possibly painful training period that reduces their moral emotions of sympathy and guilt. Such an education might be achieved by desensitising over-exposure to, and performance of, torture on innocents. If Hedonistic Egoists underwent such an education, their reduced capacity for sympathy and guilt would allow them to take advantage of any opportunities to perform pleasurable, but normally-guilt-inducing, actions, such as stealing from the poor.

Hedonistic Egoism is very unpopular amongst philosophers, not just for this reason, but also because it suffers from all of the objections that apply to Prudential Hedonism.

Hedonistic Utilitarianism is the theory that the right action is the one that produces (or is most likely to produce) the greatest net happiness for all concerned. Hedonistic Utilitarianism is often considered fairer than Hedonistic Egoism because the happiness of everyone involved (everyone who is affected or likely to be affected) is taken into account and given equal weight. Hedonistic Utilitarians, then, tend to advocate not stealing from needy orphans because to do so would usually leave the orphan far less happy and the (probably better-off) thief only slightly happier (assuming he felt no guilt). Despite treating all individuals equally, Hedonistic Utilitarianism is still seen as objectionable by some because it assigns no intrinsic moral value to justice, friendship, truth, or any of the many other goods that are thought by some to be irreducibly valuable. For example, a Hedonistic Utilitarian would be morally obliged to publicly execute an innocent friend of theirs if doing so was the only way to promote the greatest happiness overall. Although unlikely, such a situation might arise if a child was murdered in a small town and the lack of suspects was causing large-scale inter-ethnic violence. Some philosophers argue that executing an innocent friend is immoral precisely because it ignores the intrinsic values of justice, friendship, and possibly truth.

Hedonistic Utilitarianism is rarely endorsed by philosophers, but mainly because of its reliance on Prudential Hedonism as opposed to its utilitarian element. Non-hedonistic versions of utilitarianism are about as popular as the other leading theories of right action, especially when it is the actions of institutions that are being considered.

Perhaps the earliest written record of hedonism comes from the Crvka, an Indian philosophical tradition based on the Barhaspatya sutras. The Crvka persisted for two thousand years (from about 600 B.C.E.). Most notably, the Crvka advocated scepticism and Hedonistic Egoism that the right action is the one that brings the actor the most net pleasure. The Crvka acknowledged that some pain often accompanied, or was later caused by, sensual pleasure, but that pleasure was worth it.

The Cyrenaics, founded by Aristippus (c. 435-356 B.C.E.), were also sceptics and Hedonistic Egoists. Although the paucity of original texts makes it difficult to confidently state all of the justifications for the Cyrenaics positions, their overall stance is clear enough. The Cyrenaics believed pleasure was the ultimate good and everyone should pursue all immediate pleasures for themselves. They considered bodily pleasures better than mental pleasures, presumably because they were more vivid or trustworthy. The Cyrenaics also recommended pursuing immediate pleasures and avoiding immediate pains with scant or no regard for future consequences. Their reasoning for this is even less clear, but is most plausibly linked to their sceptical views perhaps that what we can be most sure of in this uncertain existence is our current bodily pleasures.

Epicurus (c. 341-271 B.C.E.), founder of Epicureanism, developed a Normative Hedonism in stark contrast to that of Aristippus. The Epicureanism of Epicurus is also quite the opposite to the common usage of Epicureanism; while we might like to go on a luxurious “Epicurean” holiday packed with fine dining and moderately excessive wining, Epicurus would warn us that we are only setting ourselves up for future pain. For Epicurus, happiness was the complete absence of bodily and especially mental pains, including fear of the Gods and desires for anything other than the bare necessities of life. Even with only the limited excesses of ancient Greece on offer, Epicurus advised his followers to avoid towns, and especially marketplaces, in order to limit the resulting desires for unnecessary things. Once we experience unnecessary pleasures, such as those from sex and rich food, we will then suffer from painful and hard to satisfy desires for more and better of the same. No matter how wealthy we might be, Epicurus would argue, our desires will eventually outstrip our means and interfere with our ability to live tranquil, happy lives. Epicureanism is generally egoistic, in that it encourages everyone to pursue happiness for themselves. However, Epicureans would be unlikely to commit any of the selfish acts we might expect from other egoists because Epicureans train themselves to desire only the very basics, which gives them very little reason to do anything to interfere with the affairs of others.

With the exception of a brief period discussed below, Hedonism has been generally unpopular ever since its ancient beginnings. Although criticisms of the ancient forms of hedonism were many and varied, one in particular was heavily cited. In Philebus, Platos Socrates and one of his many foils, Protarchus in this instance, are discussing the role of pleasure in the good life. Socrates asks Protarchus to imagine a life without much pleasure but full of the higher cognitive processes, such as knowledge, forethought and consciousness and to compare it with a life that is the opposite. Socrates describes this opposite life as having perfect pleasure but the mental life of an oyster, pointing out that the subject of such a life would not be able to appreciate any of the pleasure within it. The harrowing thought of living the pleasurable but unthinking life of an oyster causes Protarchus to abandon his hedonistic argument. The oyster example is now easily avoided by clarifying that pleasure is best understood as being a conscious experience, so any sensation that we are not consciously aware of cannot be pleasure.

Normative and Motivational Hedonism were both at their most popular during the heyday of Empiricism in the 18th and 19th Centuries. Indeed, this is the only period during which any kind of hedonism could be considered popular at all. During this period, two Hedonistic Utilitarians, Jeremy Bentham (1748-1832) and his protg John Stuart Mill (1806-1873), were particularly influential. Their theories are similar in many ways, but are notably distinct on the nature of pleasure.

Bentham argued for several types of hedonism, including those now referred to as Prudential Hedonism, Hedonistic Utilitarianism, and Motivational Hedonism (although his commitment to strong Motivational Hedonism eventually began to wane). Bentham argued that happiness was the ultimate good and that happiness was pleasure and the absence of pain. He acknowledged the egoistic and hedonistic nature of peoples motivation, but argued that the maximization of collective happiness was the correct criterion for moral behavior. Benthams greatest happiness principle states that actions are immoral if they are not the action that appears to maximise the happiness of all the people likely to be affected; only the action that appears to maximise the happiness of all the people likely to be affected is the morally right action.

Bentham devised the greatest happiness principle to justify the legal reforms he also argued for. He understood that he could not conclusively prove that the principle was the correct criterion for morally right action, but also thought that it should be accepted because it was fair and better than existing criteria for evaluating actions and legislation. Bentham thought that his Hedonic Calculus could be applied to situations to see what should, morally speaking, be done in a situation. The Hedonic Calculus is a method of counting the amount of pleasure and pain that would likely be caused by different actions. The Hedonic Calculus required a methodology for measuring pleasure, which in turn required an understanding of the nature of pleasure and specifically what aspects of pleasure were valuable for us.

Benthams Hedonic Calculus identifies several aspects of pleasure that contribute to its value, including certainty, propinquity, extent, intensity, and duration. The Hedonic Calculus also makes use of two future-pleasure-or-pain-related aspects of actions fecundity and purity. Certainty refers to the likelihood that the pleasure or pain will occur. Propinquity refers to how long away (in terms of time) the pleasure or pain is. Fecundity refers to the likelihood of the pleasure or pain leading to more of the same sensation. Purity refers to the likelihood of the pleasure or pain leading to some of the opposite sensation. Extent refers to the number of people the pleasure or pain is likely to affect. Intensity refers to the felt strength of the pleasure or pain. Duration refers to how long the pleasure or pain are felt for. It should be noted that only intensity and duration have intrinsic value for an individual. Certainty, propinquity, fecundity, and purity are all instrumentally valuable for an individual because they affect the likelihood of an individual feeling future pleasure and pain. Extent is not directly valuable for an individuals well-being because it refers to the likelihood of other people experiencing pleasure or pain.

Benthams inclusion of certainty, propinquity, fecundity, and purity in the Hedonic Calculus helps to differentiate his hedonism from Folk Hedonism. Folk Hedonists rarely consider how likely their actions are to lead to future pleasure or pain, focussing instead on the pursuit of immediate pleasure and the avoidance of immediate pain. So while Folk Hedonists would be unlikely to study for an exam, anyone using Benthams Hedonic Calculus would consider the future happiness benefits to themselves (and possibly others) of passing the exam and then promptly begin studying.

Most importantly for Benthams Hedonic Calculus, the pleasure from different sources is always measured against these criteria in the same way, that is to say that no additional value is afforded to pleasures from particularly moral, clean, or culturally-sophisticated sources. For example, Bentham held that pleasure from the parlor game push-pin was just as valuable for us as pleasure from music and poetry. Since Benthams theory of Prudential Hedonism focuses on the quantity of the pleasure, rather than the source-derived quality of it, it is best described as a type of Quantitative Hedonism.

Benthams indifferent stance on the source of pleasures led to others disparaging his hedonism as the philosophy of swine. Even his student, John Stuart Mill, questioned whether we should believe that a satisfied pig leads a better life than a dissatisfied human or that a satisfied fool leads a better life than a dissatisfied Socrates results that Benthams Quantitative Hedonism seems to endorse.

Like Bentham, Mill endorsed the varieties of hedonism now referred to as Prudential Hedonism, Hedonistic Utilitarianism, and Motivational Hedonism. Mill also thought happiness, defined as pleasure and the avoidance of pain, was the highest good. Where Mills hedonism differs from Benthams is in his understanding of the nature of pleasure. Mill argued that pleasures could vary in quality, being either higher or lower pleasures. Mill employed the distinction between higher and lower pleasures in an attempt to avoid the criticism that his hedonism was just another philosophy of swine. Lower pleasures are those associated with the body, which we share with other animals, such as pleasure from quenching thirst or having sex. Higher pleasures are those associated with the mind, which were thought to be unique to humans, such as pleasure from listening to opera, acting virtuously, and philosophising. Mill justified this distinction by arguing that those who have experienced both types of pleasure realise that higher pleasures are much more valuable. He dismissed challenges to this claim by asserting that those who disagreed lacked either the experience of higher pleasures or the capacity for such experiences. For Mill, higher pleasures were not different from lower pleasures by mere degree; they were different in kind. Since Mills theory of Prudential Hedonism focuses on the quality of the pleasure, rather than the amount of it, it is best described as a type of Qualitative Hedonism.

George Edward Moore (1873-1958) was instrumental in bringing hedonisms brief heyday to an end. Moores criticisms of hedonism in general, and Mills hedonism in particular, were frequently cited as good reasons to reject hedonism even decades after his death. Indeed, since G. E. Moore, hedonism has been viewed by most philosophers as being an initially intuitive and interesting family of theories, but also one that is flawed on closer inspection. Moore was a pluralist about value and argued persuasively against the Value Hedonists central claim that all and only pleasure is the bearer of intrinsic value. Moores most damaging objection against Hedonism was his heap of filth example. Moore himself thought the heap of filth example thoroughly refuted what he saw as the only potentially viable form of Prudential Hedonism that conscious pleasure is the only thing that positively contributes to well-being. Moore used the heap of filth example to argue that Prudential Hedonism is false because pleasure is not the only thing of value.

In the heap of filth example, Moore asks the reader to imagine two worlds, one of which is exceedingly beautiful and the other a disgusting heap of filth. Moore then instructs the reader to imagine that no one would ever experience either world and asks if it is better for the beautiful world to exist than the filthy one. As Moore expected, his contemporaries tended to agree that it would be better if the beautiful world existed. Relying on this agreement, Moore infers that the beautiful world is more valuable than the heap of filth and, therefore, that beauty must be valuable. Moore then concluded that all of the potentially viable theories of Prudential Hedonism (those that value only conscious pleasures) must be false because something, namely beauty, is valuable even when no conscious pleasure can be derived from it.

Moores heap of filth example has rarely been used to object to Prudential Hedonism since the 1970s because it is not directly relevant to Prudential Hedonism (it evaluates worlds and not lives). Moores other objections to Prudential Hedonism also went out of favor around the same time. The demise of these arguments was partly due to mounting objections against them, but mainly because arguments more suited to the task of refuting Prudential Hedonism were developed. These arguments are discussed after the contemporary varieties of hedonism are introduced below.

Several contemporary varieties of hedonism have been defended, although usually by just a handful of philosophers or less at any one time. Other varieties of hedonism are also theoretically available but have received little or no discussion. Contemporary varieties of Prudential Hedonism can be grouped based on how they define pleasure and pain, as is done below. In addition to providing different notions of what pleasure and pain are, contemporary varieties of Prudential Hedonism also disagree about what aspect or aspects of pleasure are valuable for well-being (and the opposite for pain).

The most well-known disagreement about what aspects of pleasure are valuable occurs between Quantitative and Qualitative Hedonists. Quantitative Hedonists argue that how valuable pleasure is for well-being depends on only the amount of pleasure, and so they are only concerned with dimensions of pleasure such as duration and intensity. Quantitative Hedonism is often accused of over-valuing animalistic, simple, and debauched pleasures.

Qualitative Hedonists argue that, in addition to the dimensions related to the amount of pleasure, one or more dimensions of quality can have an impact on how pleasure affects well-being. The quality dimensions might be based on how cognitive or bodily the pleasure is (as it was for Mill), the moral status of the source of the pleasure, or some other non-amount-related dimension. Qualitative Hedonism is criticised by some for smuggling values other than pleasure into well-being by misleadingly labelling them as dimensions of pleasure. How these qualities are chosen for inclusion is also criticised for being arbitrary or ad hoc by some because inclusion of these dimensions of pleasure is often in direct response to objections that Quantitative Hedonism cannot easily deal with. That is to say, the inclusion of these dimensions is often accused of being an exercise in plastering over holes, rather than deducing corollary conclusions from existing theoretical premises. Others have argued that any dimensions of quality can be better explained in terms of dimensions of quantity. For example, they might claim that moral pleasures are no higher in quality than immoral pleasures, but that moral pleasures are instrumentally more valuable because they are likely to lead to more moments of pleasure or less moments of pain in the future.

Hedonists also have differing views about how the value of pleasure compares with the value of pain. This is not a practical disagreement about how best to measure pleasure and pain, but rather a theoretical disagreement about comparative value, such as whether pain is worse for us than an equivalent amount of pleasure is good for us. The default position is that one unit of pleasure (sometimes referred to as a Hedon) is equivalent but opposite in value to one unit of pain (sometimes referred to as a Dolor). Several Hedonistic Utilitarians have argued that reduction of pain should be seen as more important than increasing pleasure, sometimes for the Epicurean reason that pain seems worse for us than an equivalent amount of pleasure is good for us. Imagine that a magical genie offered for you to play a game with him. The game consists of you flipping a fair coin. If the coin lands on heads, then you immediately feel a burst of very intense pleasure and if it lands on tails, then you immediately feel a burst of very intense pain. Is it in your best interests to play the game?

Another area of disagreement between some Hedonists is whether pleasure is entirely internal to a person or if it includes external elements. Internalism about pleasure is the thesis that, whatever pleasure is, it is always and only inside a person. Externalism about pleasure, on the other hand, is the thesis that, pleasure is more than just a state of an individual (that is, that a necessary component of pleasure lies outside of the individual). Externalists about pleasure might, for example, describe pleasure as a function that mediates between our minds and the environment, such that every instance of pleasure has one or more integral environmental components. The vast majority of historic and contemporary versions of Prudential Hedonism consider pleasure to be an internal mental state.

Perhaps the least known disagreement about what aspects of pleasure make it valuable is the debate about whether we have to be conscious of pleasure for it to be valuable. The standard position is that pleasure is a conscious mental state, or at least that any pleasure a person is not conscious of does not intrinsically improve their well-being.

The most common definition of pleasure is that it is a sensation, something that we identify through our senses or that we feel. Psychologists claim that we have at least ten senses, including the familiar, sight, hearing, smell, taste, and touch, but also, movement, balance, and several sub-senses of touch, including heat, cold, pressure, and pain. New senses get added to the list when it is understood that some independent physical process underpins their functioning. The most widely-used examples of pleasurable sensations are the pleasures of eating, drinking, listening to music, and having sex. Use of these examples has done little to help Hedonism avoid its debauched reputation.

It is also commonly recognised that our senses are physical processes that usually involve a mental component, such as the tickling feeling when someone blows gently on the back of your neck. If a sensation is something we identify through our sense organs, however, it is not entirely clear how to account for abstract pleasures. This is because abstract pleasures, such as a feeling of accomplishment for a job well done, do not seem to be experienced through any of the senses in the standard lists. Some Hedonists have attempted to resolve this problem by arguing for the existence of an independent pleasure sense and by defining sensation as something that we feel (regardless of whether it has been mediated by sense organs).

Most Hedonists who describe pleasure as a sensation will be Quantitative Hedonists and will argue that the pleasure from the different senses is the same. Qualitative Hedonists, in comparison, can use the framework of the senses to help differentiate between qualities of pleasure. For example, a Qualitative Hedonist might argue that pleasurable sensations from touch and movement are always lower quality than the others.

Hedonists have also defined pleasure as intrinsically valuable experience, that is to say any experiences that we find intrinsically valuable either are, or include, instances of pleasure. According to this definition, the reason that listening to music and eating a fine meal are both intrinsically pleasurable is because those experiences include an element of pleasure (along with the other elements specific to each activity, such as the experience of the texture of the food and the melody of the music). By itself, this definition enables Hedonists to make an argument that is close to perfectly circular. Defining pleasure as intrinsically valuable experience and well-being as all and only experiences that are intrinsically valuable allows a Hedonist to all but stipulate that Prudential Hedonism is the correct theory of well-being. Where defining pleasure as intrinsically valuable experience is not circular is in its stipulation that only experiences matter for well-being. Some well-known objections to this idea are discussed below.

Another problem with defining pleasure as intrinsically valuable experience is that the definition does not tell us very much about what pleasure is or how it can be identified. For example, knowing that pleasure is intrinsically valuable experience would not help someone to work out if a particular experience was intrinsically or just instrumentally valuable. Hedonists have attempted to respond to this problem by explaining how to find out whether an experience is intrinsically valuable.

One method is to ask yourself if you would like the experience to continue for its own sake (rather than because of what it might lead to). Wanting an experience to continue for its own sake reveals that you find it to be intrinsically valuable. While still making a coherent theory of well-being, defining intrinsically valuable experiences as those you want to perpetuate makes the theory much less hedonistic. The fact that what a person wants is the main criterion for something having intrinsic value, makes this kind of theory more in line with preference satisfaction theories of well-being. The central claim of preference satisfaction theories of well-being is that some variant of getting what one wants, or should want, under certain conditions is the only thing that intrinsically improves ones well-being.

Another method of fleshing out the definition of pleasure as intrinsically valuable experience is to describe how intrinsically valuable experiences feel. This method remains a hedonistic one, but seems to fall back into defining pleasure as a sensation.

It has also been argued that what makes an experience intrinsically valuable is that you like or enjoy it for its own sake. Hedonists arguing for this definition of pleasure usually take pains to position their definition in between the realms of sensation and preference satisfaction. They argue that since we can like or enjoy some experiences without concurrently wanting them or feeling any particular sensation, then liking is distinct from both sensation and preference satisfaction. Liking and enjoyment are also difficult terms to define in more detail, but they are certainly easier to recognise than the rather opaque “intrinsically valuable experience.”

Merely defining pleasure as intrinsically valuable experience and intrinsically valuable experiences as those that we like or enjoy still lacks enough detail to be very useful for contemplating well-being. A potential method for making this theory more useful would be to draw on the cognitive sciences to investigate if there is a specific neurological function for liking or enjoying. Cognitive science has not reached the point where anything definitive can be said about this, but a few neuroscientists have experimental evidence that liking and wanting (at least in regards to food) are neurologically distinct processes in rats and have argued that it should be the same for humans. The same scientists have wondered if the same processes govern all of our liking and wanting, but this question remains unresolved.

Most Hedonists who describe pleasure as intrinsically valuable experience believe that pleasure is internal and conscious. Hedonists who define pleasure in this way may be either Quantitative or Qualitative Hedonists, depending on whether they think that quality is a relevant dimension of how intrinsically valuable we find certain experiences.

One of the most recent developments in modern hedonism is the rise of defining pleasure as a pro-attitude a positive psychological stance toward some object. Any account of Prudential Hedonism that defines pleasure as a pro-attitude is referred to as Attitudinal Hedonism because it is a persons attitude that dictates whether anything has intrinsic value. Positive psychological stances include approving of something, thinking it is good, and being pleased about it. The object of the positive psychological stance could be a physical object, such as a painting one is observing, but it could also be a thought, such as “my country is not at war,” or even a sensation. An example of a pro-attitude towards a sensation could be being pleased about the fact that an ice cream tastes so delicious.

Fred Feldman, the leading proponent of Attitudinal Hedonism, argues that the sensation of pleasure only has instrumental value it only brings about value if you also have a positive psychological stance toward that sensation. In addition to his basic Intrinsic Attitudinal Hedonism, which is a form of Quantitative Hedonism, Feldman has also developed many variants that are types of Qualitative Hedonism. For example, Desert-Adjusted Intrinsic Attitudinal Hedonism, which reduces the intrinsic value a pro-attitude has for our well-being based on the quality of deservedness (that is, on the extent to which the particular object deserves a pro-attitude or not). For example, Desert-Adjusted Intrinsic Attitudinal Hedonism might stipulate that sensations of pleasure arising from adulterous behavior do not deserve approval, and so assign them no value.

Defining pleasure as a pro-attitude, while maintaining that all sensations of pleasure have no intrinsic value, makes Attitudinal Hedonism less obviously hedonistic as the versions that define pleasure as a sensation. Indeed, defining pleasure as a pro-attitude runs the risk of creating a preference satisfaction account of well-being because being pleased about something without feeling any pleasure seems hard to distinguish from having a preference for that thing.

The most common argument against Prudential Hedonism is that pleasure is not the only thing that intrinsically contributes to well-being. Living in reality, finding meaning in life, producing noteworthy achievements, building and maintaining friendships, achieving perfection in certain domains, and living in accordance with religious or moral laws are just some of the other things thought to intrinsically add value to our lives. When presented with these apparently valuable aspects of life, Hedonists usually attempt to explain their apparent value in terms of pleasure. A Hedonist would argue, for example, that friendship is not valuable in and of itself, rather it is valuable to the extent that it brings us pleasure. Furthermore, to answer why we might help a friend even when it harms us, a Hedonist will argue that the prospect of future pleasure from receiving reciprocal favors from our friend, rather than the value of friendship itself, should motivate us to help in this way.

Those who object to Prudential Hedonism on the grounds that pleasure is not the only source of intrinsic value use two main strategies. In the first strategy, objectors make arguments that some specific value cannot be reduced to pleasure. In the second strategy, objectors cite very long lists of apparently intrinsically valuable aspects of life and then challenge hedonists with the prolonged and arduous task of trying to explain how the value of all of them can be explained solely by reference to pleasure and the avoidance of pain. This second strategy gives good reason to be a pluralist about value because the odds seem to be against any monistic theory of value, such as Prudential Hedonism. The first strategy, however, has the ability to show that Prudential Hedonism is false, rather than being just unlikely to be the best theory of well-being.

The most widely cited argument for pleasure not being the only source of intrinsic value is based on Robert Nozicks experience machine thought-experiment. Nozicks experience machine thought-experiment was designed to show that more than just our experiences matter to us because living in reality also matters to us. This argument has proven to be so convincing that nearly every single book on ethics that discusses hedonism rejects it using only this argument or this one and one other.

In the thought experiment, Nozick asks us to imagine that we have the choice of plugging in to a fantastic machine that flawlessly provides an amazing mix of experiences. Importantly, this machine can provide these experiences in a way that, once plugged in to the machine, no one can tell that their experiences are not real. Disregarding considerations about responsibilities to others and the problems that would arise if everyone plugged in, would you plug in to the machine for life? The vast majority of people reject the choice to live a much more pleasurable life in the machine, mostly because they agree with Nozick that living in reality seems to be important for our well-being. Opinions differ on what exactly about living in reality is so much better for us than the additional pleasure of living in the experience machine, but the most common response is that a life that is not lived in reality is pointless or meaningless.

Since this argument has been used so extensively (from the mid 1970s onwards) to dismiss Prudential Hedonism, several attempts have been made to refute it. Most commonly, Hedonists argue that living an experience machine life would be better than living a real life and that most people are simply mistaken to not want to plug in. Some go further and try to explain why so many people choose not to plug in. Such explanations often point out that the most obvious reasons for not wanting to plug in can be explained in terms of expected pleasure and avoidance of pain. For example, it might be argued that we expect to get pleasure from spending time with our real friends and family, but we do not expect to get as much pleasure from the fake friends or family we might have in the experience machine. These kinds of attempts to refute the experience machine objection do little to persuade non-Hedonists that they have made the wrong choice.

A more promising line of defence for the Prudential Hedonists is to provide evidence that there is a particular psychological bias that affects most peoples choice in the experience machine thought experiment. A reversal of Nozicks thought experiment has been argued to reveal just such a bias. Imagine that a credible source tells you that you are actually in an experience machine right now. You have no idea what reality would be like. Given the choice between having your memory of this conversation wiped and going to reality, what would be best for you to choose? Empirical evidence on this choice shows that most people would choose to stay in the experience machine. Comparing this result with how people respond to Nozicks experience machine thought experiment reveals the following: In Nozicks experience machine thought experiment people tend to choose a real and familiar life over a more pleasurable life and in the reversed experience machine thought experiment people tend to choose a familiar life over a real life. Familiarity seems to matter more than reality, undermining the strength of Nozicks original argument. The bias thought to be responsible for this difference is the status quo bias an irrational preference for the familiar or for things to stay as they are.

Regardless of whether Nozicks experience machine thought experiment is as decisive a refutation of Prudential Hedonism as it is often thought to be, the wider argument (that living in reality is valuable for our well-being) is still a problem for Prudential Hedonists. That our actions have real consequences, that our friends are real, and that our experiences are genuine seem to matter for most of us regardless of considerations of pleasure. Unfortunately, we lack a trusted methodology for discerning if these things should matter to us. Perhaps the best method for identifying intrinsically valuable aspects of lives is to compare lives that are equal in pleasure and all other important ways, except that one aspect of one of the lives is increased. Using this methodology, however, seems certain to lead to an artificial pluralist conclusion about what has value. This is because any increase in a potentially valuable aspect of our lives will be viewed as a free bonus. And, most people will choose the life with the free bonus just in case it has intrinsic value, not necessarily because they think it does have intrinsic value.

The main traditional line of criticism against Prudential Hedonism is that not all pleasure is valuable for well-being, or at least that some pleasures are less valuable than others because of non-amount-related factors. Some versions of this criticism are much easier for Prudential Hedonists to deal with than others depending on where the allegedly disvaluable aspect of the pleasure resides. If the disvaluable aspect is experienced with the pleasure itself, then both Qualitative and Quantitative varieties of Prudential Hedonism have sufficient answers to these problems. If, however, the disvaluable aspect of the pleasure is never experienced, then all types of Prudential Hedonism struggle to explain why the allegedly disvaluable aspect is irrelevant.

Examples of the easier criticisms to deal with are that Prudential Hedonism values, or at least overvalues, perverse and base pleasures. These kinds of criticisms tend to have had more sway in the past and doubtless encouraged Mill to develop his Qualitative Hedonism. In response to the charge that Prudential Hedonism mistakenly values pleasure from sadistic torture, sating hunger, copulating, listening to opera, and philosophising all equally, Qualitative Hedonists can simply deny that it does. Since pleasure from sadistic torture will normally be experienced as containing the quality of sadism (just as the pleasure from listening to good opera is experienced as containing the quality of acoustic excellence), the Qualitative Hedonist can plausibly claim to be aware of the difference in quality and allocate less value to perverse or base pleasures accordingly.

Prudential Hedonists need not relinquish the Quantitative aspect of their theory in order to deal with these criticisms, however. Quantitative Hedonists, can simply point out that moral or cultural values are not necessarily relevant to well-being because the investigation of well-being aims to understand what the good life for the one living it is and what intrinsically makes their life go better for them. A Quantitative Hedonist can simply respond that a sadist that gets sadistic pleasure from torturing someone does improve their own well-being (assuming that the sadist never feels any negative emotions or gets into any other trouble as a result). Similarly, a Quantitative Hedonist can argue that if someone genuinely gets a lot of pleasure from porcine company and wallowing in the mud, but finds opera thoroughly dull, then we have good reason to think that having to live in a pig sty would be better for her well-being than forcing her to listen to opera.

Much more problematic for both Quantitative and Qualitative Hedonists, however, are the more modern versions of the criticism that not all pleasure is valuable. The modern versions of this criticism tend to use examples in which the disvaluable aspect of the pleasure is never experienced by the person whose well-being is being evaluated. The best example of these modern criticisms is a thought experiment devised by Shelly Kagan. Kagans deceived businessman thought experiment is widely thought to show that pleasures of a certain kind, namely false pleasures, are worth much less than true pleasures.

Kagan asks us to imagine the life of a very successful businessman who takes great pleasure in being respected by his colleagues, well-liked by his friends, and loved by his wife and children until the day he died. Then Kagan asks us to compare this life with one of equal length and the same amount of pleasure (experienced as coming from exactly the same sources), except that in each case the businessman is mistaken about how those around him really feel. This second (deceived) businessman experiences just as much pleasure from the respect of his colleagues and the love of his family as the first businessman. The only difference is that the second businessman has many false beliefs. Specifically, the deceived businessmans colleagues actually think he is useless, his wife doesnt really love him, and his children are only nice to him so that he will keep giving them money. Given that the deceived businessman never knew of any of these deceptions and his experiences were never negatively impacted by the deceptions indirectly, which life do you think is better?

Nearly everyone thinks that the deceived businessman has a worse life. This is a problem for Prudential Hedonists because the pleasure is quantitatively equal in each life, so they should be equally good for the one living it. Qualitative Hedonism does not seem to be able to avoid this criticism either because the falsity of the pleasures experienced by the deceived businessman is a dimension of the pleasure that he never becomes aware of. Theoretically, an externalist and qualitative version of Attitudinal Hedonism could include the falsity dimension of an instance of pleasure even if the falsity dimension never impacts the consciousness of the person. However, the resulting definition of pleasure bears little resemblance to what we commonly understand pleasure to be and also seems to be ad hoc in its inclusion of the truth dimension but not others. A dedicated Prudential Hedonist of any variety can always stubbornly stick to the claim that the lives of the two businessmen are of equal value, but that will do little to convince the vast majority to take Prudential Hedonism more seriously.

Another major line of criticism used against Prudential Hedonists is that they have yet to come up with a meaningful definition of pleasure that unifies the seemingly disparate array of pleasures while remaining recognisable as pleasure. Some definitions lack sufficient detail to be informative about what pleasure actually is, or why it is valuable, and those that do offer enough detail to be meaningful are faced with two difficult tasks.

The first obstacle for a useful definition of pleasure for hedonism is to unify all of the diverse pleasures in a reasonable way. Phenomenologically, the pleasure from reading a good book is very different to the pleasure from bungee jumping, and both of these pleasures are very different to the pleasure of having sex. This obstacle is unsurpassable for most versions of Quantitative Hedonism because it makes the value gained from different pleasures impossible to compare. Not being able to compare different types of pleasure results in being unable to say if a life is better than another in most even vaguely realistic cases. Furthermore, not being able to compare lives means that Quantitative Hedonism could not be usefully used to guide behavior since it cannot instruct us on which life to aim for.

Attempts to resolve the problem of unifying the different pleasures while remaining within a framework of Quantitative Hedonism, usually involve pointing out something that is constant in all of the disparate pleasures and defining that particular thing as pleasure. When pleasure is defined as a strict sensation, this strategy fails because introspection reveals that no such sensation exists. Pleasure defined as the experience of liking or as a pro-attitude does much better at unifying all of the diverse pleasures. However, defining pleasure in these ways makes the task of filling in the details of the theory a fine balancing act. Liking or pro-attitudes must be described in such a way that they are not solely a sensation or best described as a preference satisfaction theory. And they must perform this balancing act while still describing a scientifically plausible and conceptually coherent account of pleasure. Most attempts to define pleasure as liking or pro-attitudes seem to disagree with either the folk conception of what pleasure is or any of the plausible scientific conceptions of how pleasure functions.

Most varieties of Qualitative Hedonism do better at dealing with the problem of diverse pleasures because they can evaluate different pleasures according to their distinct qualities. Qualitative Hedonists still need a coherent method for comparing the different pleasures with each other in order to be more than just an abstract theory of well-being, however. And, it is difficult to construct such a methodology in a way that avoids counter examples, while still describing a scientifically plausible and conceptually coherent account of pleasure.

The second obstacle is creating a definition of pleasure that retains at least some of the core properties of the common understanding of the term pleasure. As mentioned, many of the potential adjustments to the main definitions of pleasure are useful for avoiding one or more of the many objections against Prudential Hedonism. The problem with this strategy is that the more adjustments that are made, the more apparent it becomes that the definition of pleasure is not recognisable as the pleasure that gave Hedonism its distinctive intuitive plausibility in the first place. When an instance of pleasure is defined simply as when someone feels good, its intrinsic value for well-being is intuitively obvious. However, when the definition of pleasure is stretched, so as to more effectively argue that all valuable experiences are pleasurable, it becomes much less recognisable as the concept of pleasure we use in day-to-day life and its intrinsic value becomes much less intuitive.

The future of hedonism seems bleak. The considerable number and strength of the arguments against Prudential Hedonisms central principle (that pleasure and only pleasure intrinsically contributes positively to well-being and the opposite for pain) seem insurmountable. Hedonists have been creative in their definitions of pleasure so as to avoid these objections, but more often than not find themselves defending a theory that is not particularly hedonistic, realistic or both.

Perhaps the only hope that Hedonists of all types can have for the future is that advances in cognitive science will lead to a better understanding of how pleasure works in the brain and how biases affect our judgements about thought experiments. If our improved understanding in these areas confirms a particular theory about what pleasure is and also provides reasons to doubt some of the widespread judgements about the thought experiments that make the vast majority of philosophers reject hedonism, then hedonism might experience at least a partial revival. The good news for Hedonists is that at least some emerging theories and results from cognitive science do appear to support some aspects of hedonism.

Dan WeijersEmail: danweijers@gmail.comVictoria University of WellingtonNew Zealand

Excerpt from:

Hedonism | Internet Encyclopedia of Philosophy