Monthly Archives: April 2017

When Microsoft released an artificially intelligent chatbot named Tay on Twitter last March, things took a predictably disastrous turn. Within 24 hours, the bot was spewing racist, neo-Nazi rants, much of which it picked up by incorporating the language of Twitter users who interacted with it.

Unfortunately, new research finds that Twitter trolls aren't the only way that AI devices can learn racist language. In fact, any artificial intelligence that learns from human language is likely to come away biased in the same ways that humans are, according to the scientists.

The researchers experimented with a widely used machine-learning system called the Global Vectors for Word Representation (GloVe) and found that every sort of human bias they tested showed up in the artificial system. [Super-Intelligent Machines: 7 Robotic Futures]

"It was astonishing to see all the results that were embedded in these models," said Aylin Caliskan, a postdoctoral researcher in computer science at Princeton University. Even AI devices that are "trained" on supposedly neutral texts like Wikipedia or news articles came to reflect common human biases, she told Live Science.

GloVe is a tool used to extract associations from texts in this case, a standard corpus of language pulled from the World Wide Web.

Psychologists have long known that the human brain makes associations between words based on their underlying meanings. A tool called the Implicit Association Test uses reaction times to demonstrate these associations: People see a word like "daffodil" alongside pleasant or unpleasant concepts like "pain" or "beauty" and have to quickly associate the terms using a key press. Unsurprisingly, flowers are more quickly associated with positive concepts; while weapons, for example, are more quickly associated with negative concepts.

The IAT can be used to reveal unconscious associations people make about social or demographic groups, as well. For example, some IATs that are available on the Project Implicit website find that people are more likely to automatically associate weapons with black Americans and harmless objects with white Americans.

There are debates about what these results mean, researchers have said. Do people make these associations because they hold personal, deep-seated social biases they aren't aware of, or do they absorb them from language that is statistically more likely to put negative words in close conjunction with ethnic minorities, the elderly and other marginalized groups?

Caliskan and her colleagues developed an IAT for computers, which they dubbed the WEAT, for Word-Embedding Association Test. This test measured the strength of associations between words as represented by GloVe, much as the IAT measures the strength of word associations in the human brain.

For every association and stereotype tested, the WEAT returned the same results as the IAT. The machine-learning tool reproduced human associations between flowers and pleasant words; insects and unpleasant words; musical instruments and pleasant words; and weapons and unpleasant words. In a more troubling finding, it saw European-American names as more pleasant than African-American names. It also associated male names more readily with career words, and female names more readily with family words. Men were more closely associated with math and science, and women with the arts. Names associated with old people were more unpleasant than names associated with young people.

"We were quite surprised that we were able to replicate every single IAT that was performed in the past by millions," Caliskan said.

Using a second method that was similar, the researchers also found that the machine-learning tool was able to accurately represent facts about the world from its semantic associations. Comparing the GloVe word-embedding results with real U.S. Bureau of Labor Statistics data on the percentage of women in occupations, Caliskan found a 90 percent correlation between professions that the GloVe saw as "female" and the actual percentage of women in those professions.

In other words, programs that learn from human language do get "a very accurate representation of the world and culture," Caliskan said, even if that culture like stereotypes and prejudice is problematic. The AI is also bad at understanding context that humans grasp easily. For example, an article about Martin Luther King Jr. being jailed for civil rights protests in Birmingham, Alabama, in 1963 would likely associate a lot of negative words with African-Americans. A human would reasonably interpret the story as one of righteous protest by an American hero; a computer would add another tally to its "black=jail" category.

Retaining accuracy while getting AI tools to understand fairness is a big challenge, Caliskan said. [A Brief History of Artificial Intelligence]

"We don't think that removing bias would necessarily solve these problems, because it's probably going to break the accurate representation of the world," she said.

The new study, published online today (April 12) in the journal Science, is not surprising, said Sorelle Friedler, a computer scientist at Haverford College who was not involved in the research. It is, however, important, she said.

"This is using a standard underlying method that many systems are then built off of," Friedler told Live Science. In other words, biases are likely to infiltrate any AI that uses GloVe, or that learns from human language in general.

Friedler is involved in an emerging field of research called Fairness, Accountability and Transparency in Machine Learning. There are no easy ways to solve these problems, she said. In some cases, programmers might be able to explicitly tell the system to automatically disregard specific stereotypes, she said. In any case involving nuance, humans may need to be looped in to make sure the machine doesn't run amok. The solutions will likely vary, depending on what the AI is designed to do, Caliskan said are they for search applications, for decision making or for something else?

In humans, implicit attitudes actually don't correlate very strongly with explicit attitudes about social groups. Psychologists have argued about why this is: Are people just keeping mum about their prejudices to avoid stigma? Does the IAT not actually measure prejudice that well? But, it appears that people at least have the ability to reason about right and wrong, with their biased associations, Caliskan said. She and her colleagues think humans will need to be involved and programming code will need to be transparent so that people can make value judgments about the fairness of machines.

"In a biased situation, we know how to make the right decision," Caliskan said, "but unfortunately, machines are not self-aware."

Original article on Live Science.

See the original post:

Bad News: Artificial Intelligence Is Racist, Too - Live Science

BBC News

Artificial intelligence: How to avoid racist algorithms BBC News There is growing concern that many of the algorithms that make decisions about our lives - from what we see on the internet to how likely we are to become victims or instigators of crime - are trained on data sets that do not include a diverse range of ... and more »

Continued here:

Artificial intelligence: How to avoid racist algorithms - BBC News

H

uman intelligence has long powered hospitals and health care. We rely on doctors, nurses, and a variety of other clinicians to solve problems and create new solutions. Advances in artificial intelligence are now making it possible to apply this form of computer-based thinking to health care.

As the chief technology officer for a new state-of-the-art advanced medical learning facility, I have been closely watching developments in artificial intelligence. Here are three areas training, surgical robots, and data mining in which I believe it will begin making a difference sooner rather than later.

Inside their operating rooms, surgeons are the captains of the ship. They possess extensive medical training and the skills to apply it. But they rely on the cooperation and contributions of the entire team to make the most of those skills. Unfortunately, few surgeons get training in how to effectively lead people with different educational and skill backgrounds.

advertisement

Creating environments in which all members of an operating room team can come together to learn and practice communication skills is a significant challenge. There are, of course, standard communication protocols for teamwork in health care. They have been gathered into in a national program known as TeamSTEPPS. However, the opportunity to really learn to communicate in the operating room seldom exists because different players on the team get their education and training via separate professional organizations and events.

New robot no substitute for humans in the operating room

To overcome this problem, my colleagues and I at the Florida Hospital Nicholson Center worked with a game development company called ARA/Virtual Heroes to create a virtual world in which a surgeon can practice team communication and leadership. This game runs on the same type of avatar intelligence underpinning teammates in the Call of Duty games. The automated avatars give audio feedback and guidance to help the surgeon make the right choices. A collection of rules, conditions, and scripts guide the surgeon through a scenario in the operating room and teach him or her which actions and decisions are correct and which ones arent. Game scenarios have decision branches that lead to favorable and unfavorable outcomes. As with most such games, there is just a single path through the scenario that leads to a successful conclusion and a corresponding score derived from making correct and incorrect decisions.

Major advances in robotic surgery let doctors perform many types of complex procedures with more precision, flexibility, and control than is possible with other conventional techniques. Robots like the da Vinci Surgical System provide a platform for translating a surgeons movements into precise actions with advanced instruments. Current robots, however, are not aware of the anatomy they show the surgeon, the procedures they are being used to perform, or what the surgeon intends to do. They are fantastic tools, but they arent yet intelligent assistants.

Future generations of robotic surgery platforms will be more aware of the procedure being performed and use that knowledge and perception to give the surgeon intelligent assistance. Companies like Verb Surgical, a collaboration between Google and Ethicon Endo-Surgery, have indicated that their robot will include machine learning and awareness. That would let it identify potential issues during a procedure. They also plan to link the robot to a cloud supercomputer service like IBMs Watson, so knowledge of thousands of similar procedures will be accessible to both the surgeon and the robot to improve the performance of each operation.

Watson goes to Asia: Hospitals use supercomputer for cancer treatment

Capabilities like those should greatly enhance the level of expertise brought into the operating room of the future, combining the skills and knowledge of the surgeon with the experience of thousands of his or her colleagues and the artificial intelligence of the worlds leading computer scientists.

Hospital systems collect data on thousands of patients each year. But each record slides into multiple disparate and disconnected databases. Hospitals know a great deal about individual patients, but very little about the aggregate health of their populations. Data mining and artificial intelligence have the potential to bring that information together into an integrated whole that can be analyzed to create a valuable picture of the health of any defined population while maintaining the anonymity of the individuals involved.

In 1854, Dr. John Snow proved that cholera was being spread through the water system of London by creating his now-famous death map, which showed the houses of those dying from the disease and which system was delivering their water. This big data analysis of an important health problem, carried out by a single human intelligence, saved countless lives.

The databases in modern hospital systems contain information that may identify the causes of disease for thousands of different issues on command. They can address questions like which medical services are best suited to which communities in the city, or where are new disease outbreaks originating, or which communities would benefit from which health education programs? Government is increasingly holding local health care providers responsible for the health of the populations they serve. They expect these providers to deliver clinics, vaccinations, and screenings. However, you cant know what to provide if you dont know the demographic and health makeup of your population.

We know so little about the aggregate health of communities because there arent enough minds and hours to collect and analyze massive datasets. Big data and artificial intelligence will bring computer minds to these problems and significantly improve our ability to offer effective health care to individuals and communities.

Roger Smith, PhD, is chief technology officer for the Florida Hospital Nicholson Center and a graduate faculty member at the University of Central Florida.

Roger Smith can be reached at FHNC.Info@flhosp.org Follow Roger on Twitter @@NCGlobal

View post:

Artificial intelligence: coming soon to a hospital near you - STAT

Forrester Research says intelligent agents (cognitive marketing, artificial intelligence, machine learning, chatbots, etc.) will eliminate 6% of U.S. jobs by 2021. But machine learning and artificial intelligence (AI) are already making inroads in marketing areas usually reserved for humans.

Is it time to push the panic button? Will marketer jobs as we know them today become obsolete in five, 10 or 15 years?

No on the panic button, but some marketing roles will clearly be at risk in the future.

Artificial intelligence and machine learning will bring significant changes to marketing, but those changes will most likely come through automating repetitive tasks that machines can do more efficiently.

This will free up marketers to work on activities where the human touch is more reliable, such as planning, strategy, context and correction.

What AI means for marketers

You've seen all the hype over self-driving cars, right? Marketing AI is similar. Both autonomous-driving technology for cars and marketing AI will handle the most routine, repeatable tasks -- but humans, at least for the near future, will set and correct the destination and keep their hands on or near the wheel.

A couple of definitions:

Machine learning is moving into the email technology mainstream, as you might have read about in industry publications like MediaPost -- or heard when you attend conferences where early adopter companies and technology vendors have been talking about it for a couple of years.

Subject lines are an excellent proving ground for AI and machine learning, because computers can figure out which combination of words, phrases and images in a subject line works best to meet the goals you set for a specific kind of email.

Machine learning can select content assets in emails for hyper-personalization, such as choosing which offers to send to different segments based on behavior and other data. More than dynamic content, these assets represent what works best for the brand.

Time to start sending out resumes?

Again, not yet.

AI/machine learning will likely take over jobs with these characteristics:

See why subject-line writing is a natural fit? Now, the jobs most at risk:

Even if your job is on this list, you don't need to panic unless it's the only thing you do. But for most marketers, these tasks take up most of their time but contribute the least to job satisfaction.

How much more could you get done if AI took over some or all of those tasks? Maybe you could finally have more time to develop new strategic plans or programs, integrate email with other channels, and more.

Embrace it now; profit from it later

One of my favorite quotes is from German economist Rudi Dornbusch: "In economics, things take longer to happen than you think they will, and then they happen faster than you thought they could."

In other words, these vast changes will probably take longer than we think to change the marketing landscape, but they are coming, and when they take hold -- as with the transformation to mobile-first digital marketing -- it will happen quickly.

Take time to understand the possibilities and get ahead of the changes, and look at your own work to see where you might need to upgrade your own skills so you don't get replaced by a robot.

Until next time, take it up a notch!

advertisement

advertisement

See more here:

Will Artificial Intelligence Make Email Marketers Obsolete? - MediaPost Communications

The first generation of police body cameras was introduced in 2005 in Great Britain. Their primary task was, and still is, to record police interactions with the public as well as to gather evidence at crime scenes. During a typical day, every police officer captures hours and hours of footage. To separate the irrelevant data from incriminating evidence, every recording needs to be reviewed and edited.

Now imagine sifting through hundreds of hours of police footage, looking for a specific piece of information, such as a verbal exchange between an officer and a suspect, a clear shot of license plates, or a suspect entering a particular building or meeting a particular person. Its a rather time-consuming task, so a lot of this footage ends up archived to be reviewed later, forfeiting potentially valuable or time-sensitive evidence.

The solution could be the use of artificial intelligence. Machine-learning algorithms within an AI system can be trained to differentiate between people and objects, recognize various events, such as car chases and shootings, and even identify people caught on tape through its facial-recognition abilities. It also can do some of the officers paperwork.

Once AI recognizes patterns, it tags them appropriately, enabling police officers to browse for a particular scene within a video using keywords, just like via online search engines.

It also can do some of paperwork that consumes a good chunk of police officers time. AI can transcribe verbal interactions as well as handle images and then enter both in reports. With AI working hard as officers personal secretary, he or she will have more time to concentrate on more important matters.

Axon Enterprise Inc. AAXN, -1.69% formerly Taser International and best known for its Taser electroshock weapon, has since diversified into technology products for police officers. Its Axon AI division is now working to create a system that will use enhanced image/video/audio processing and integrate AI and deep-learning capabilities.

As you can see, there are many reasons why using AI and body cameras in police work is a good idea.

However, theres another side to this story.

AIs facial-recognition ability raises privacy concerns, because it allows the police to track down any individual captured on the footage simply by running the video through the algorithm.

The Pentagon's Strategic Capabilities Office aims to "get the military ready to fight tomorrow's war." SCO Director William Roper talked to MarketWatch about how drones, innovation and big data will shape the future of warfare.

This footage be abused by police departments as well as other officials. In addition, hackers, criminals and others could get their hands on it since the data are stored in the cloud. Although faces can be redacted from the videos as its being uploaded, its only a matter of time before systems evolve enough to enable dynamic facial recognition, making these videos even more powerful tools in the hands of individuals or organizations that seek to establish mass surveillance a dream come true for any authoritarian regime.

Finally, some may ask what guarantee there is to prevent for-profit companies from acquiring and reselling police footage on the black market or on the dark web. Criminals can use that information to learn how police operate, for example.

Still I believe these cameras should exist since their benefits outweigh their disadvantages. Could the same be said in five, 10 or 15 years? Only time will tell. In the meantime, you could start wearing these while running errands to trick facial-recognition systems.

See the original post here:

Opinion: How artificial intelligence could transform police body cameras - MarketWatch

Thursday afternoon event free and open to public

SOUTH KINGSTOWN, R.I. A father and son team who have succeeded in technology entrepreneurship and who are both graduates of the University of Rhode Island will discuss artificial intelligence and robotics Thursday at 4:30 p.m. during an event that is free and open to the public, URI announced.

The annual innovation and entrepreneurship lecture features Thomas J. Chisholm, class of 1968, and 1994 graduate T. Brett Chisholm. Successful in other businesses, the two now are co-founders and heads of NeuraFlash, which provides intelligent, personalized and predictive capabilities to customers utilizing Artificial Intelligence, according to the university.

Thursdays discussion, Artificial Intelligence, Machine Learning, Bots, and the Evolution of Great Customer Service, will be moderated by Professor Nancy Forster-Holt. The event will be in the Galanti Lounge at of URIs Robert L. Carothers Library and Learning Commons, 15 Lippitt Road on the Kingston Campus.

The annual Anthony J. Risica Lecture Series features prominent leaders in engineering and business [sharing] their knowledge with the campus community and public, according to URI. The series was established in 2003 with a generous gift from Anthony Risica, a University of Rhode Island engineering alumnus.

Read more here:

URI graduates to discuss Artificial Intelligence, robotics - The Providence Journal

Amazon.com Inc. is embracing artificial intelligence to deliver goods more quickly, enhance its voice-activated Alexa assistant and create new tools sold to others through its cloud-computing division, Chief Executive Officer Jeff Bezos said in his annual shareholder letter.

Changes ushered in by artificial intelligence and machine learning will help the companies that embrace them and put up barriers for those who dont, the worlds second-richest man wrote in a 1,700-word letter released Wednesday.

Bezos repeated familiar themes, such as the need to operate a business like its always Day 1 to keep a startup mentality and the ability to act quickly on limited information to stay ahead, what he calls high-velocity decision making. His emphasis on artificial intelligence and machine learning was the most concrete indication of areas in which the e-commerce giant will continue to invest.

Machine learning is the science of getting computers to act without being programmed, and is used in autonomous cars, speech-recognition and internet search engines. The technology has influenced high-profile projects at Amazon such as drone delivery, its popular Echo voice-activated speaker and the new cashier-less Amazon Go convenience store unveiled late last year in Seattle, Bezos wrote.

Exclusive insights on technology around the world.

Get Fully Charged, from Bloomberg Technology.

But much of what we do with machine learning happens beneath the surface, he wrote. Machine learning drives our algorithms for demand forecasting, product search ranking, product and deals recommendations, merchandising placements, fraud detection, translations, and much more. Though less visible, much of the impact of machine learning will be of this type -- quietly but meaningfully improving core operations.

Amazon Web Services, the companys cloud-computing division, will offer affordable tools so clients can incorporate artificial intelligence and machine learning into their own operations. Such tools have already been used by to detect diseases and increase crop yields, Bezos wrote.

Watch this space, Bezos said. Much more to come.

See more here:

Bezos Says Artificial Intelligence to Fuel Amazon's Success - Bloomberg