A new whitepapercoauthored by researchers at the Vector Institute for Artificial Intelligence examines the ethics of AI in surgery, making the case that surgery and AI carry similar expectations but diverge with respect to ethical understanding. Surgeons are faced with moral and ethical dilemmas as a matter of course, the paper points out, whereas ethical frameworks in AI have arguably only begun to take shape.

In surgery, AI applications are largely confined to machines performing tasks controlled entirely by surgeons. AI might also be used in a clinical decision support system, and in these circumstances, the burden of responsibility falls on the human designers of the machine or AI system, the coauthors argue.

Privacy is a foremost ethical concern. AI learns to make predictions from large data sets specifically patient data, in the case of surgical systems and its often described as being at odds with privacy-preserving practices. The Royal Free London NHS Foundation Trust, a division of the U.K.s National Health Service based in London, provided Alphabets DeepMind with data on 1.6 million patients without their consent. Separately, Google, whose health data-sharing partnership with Ascension became the subject of scrutiny last November, abandoned plans to publish scans of chest X-rays over concerns that they contained personally identifiable information.

Laws at the state, local, and federal levels aim to make privacy a mandatory part of compliance management. Hundreds of bills that address privacy, cybersecurity, and data breaches are pending or have already been passed in 50 U.S. states, territories, and the District of Columbia. Arguably the most comprehensive of them all the California Consumer Privacy Act was signed into law roughly two years ago. Thats not to mention the national Health Insurance Portability and Accountability Act (HIPAA), which requires companies to seek authorization before disclosing individual health information. And international frameworks like the EUs General Privacy Data Protection Regulation (GDPR) aim to give consumers greater control over personal data collection and use.

But the whitepaper coauthors argue measures adopted to date are limited by jurisdictional interpretations and offer incomplete models of ethics. For instance, HIPAA focuses on health care data from patient records but doesnt cover sources of data generated outside of covered entities, like life insurance companies or fitness band apps. Moreover, while the duty of patient autonomy alludes to a right to explanations of decisions made by AI, frameworks like GDPR only mandate a right to be informed and appear to lack language stating well-defined safeguards against AI decision making.

Beyond this, the coauthors sound the alarm about the potential effects of bias on AI surgical systems. Training data bias, which concerns the quality and representativeness of data used to train an AI system, could dramatically affect a preoperative risk stratification prior to surgery. Underrepresentation of demographics might also cause inaccurate assessments, driving flawed decisions such as whether a patient is treated first or offered extensive ICU resources. And contextual bias, which occurs when an algorithm is employed outside the context of its training, could result in a system ignoring nontrivial caveats like whether a surgeon is right- or left-handed.

Methods to mitigate this bias exist, including ensuring variance in the data set, applying sensitivity to overfitting on training data, and having humans-in-the-loop to examine new data as its deployed. The coauthors advocate the use of these measures and of transparency broadly to prevent patient autonomy from being undermined. Already, an increasing reliance on automated decision-making tools has reduced the opportunity of meaningful dialogue between the healthcare provider and patient, they wrote. If machine learning is in its infancy, then the subfield tasked with making its inner workings explainable is so embryonic that even its terminology has yet to recognizably form. However, several fundamental properties of explainability have started to emerge [that argue] machine learning should be simultaneous, decomposable, and algorithmically transparent.

Despite AIs shortcomings, particularly in the context of surgery, the coauthors argue the harms AI can prevent outweigh the adoption cons. For example, in thyroidectomy, theres risk of permanent hypoparathyroidism and recurrent nerve injury. It might take thousands of procedures with a new method to observe statistically significant changes, which an individual surgeon might never observe at least not in a short time frame. However, a repository of AI-based analytics aggregating these thousands of cases from hundreds of sites would be able to discern and communicate those significant patterns.

The continued technological advancement in AI will sow rapid increases in the breadths and depths of their duties. Extrapolating from the progress curve, we can predict that machines will become more autonomous, the coauthors wrote. The rise in autonomy necessitates an increased focus on the ethical horizon that we need to scrutinize Like ethical decision-making in current practice, machine learning will not be effective if it is merely designed carefully by committee it requires exposure to the real world.

Read the original post:

Researchers examine the ethical implications of AI in surgical settings - VentureBeat

On July 8, 2020, the House Financial Services Committees Taskforce on Artificial Intelligence held a hearing entitled Exposure Notification and Contact Tracing: How AI Helps Localities Reopen Safely and Researchers Find a Cure.

In his opening remarks, Congressman Bill Foster (D-IL), chairman of the task force, stated that the hearing would discuss the essential tradeoffs that the coronavirus disease 2019 (COVID-19) pandemic was forcing on the public between life, liberty, privacy and the pursuit of happiness. Chairman Foster noted that what he called invasive artificial intelligence (AI) surveillance may save lives, but would come at a tremendous cost to personal liberty. He said that contact tracing apps that use back-end AI, which combines raw data collected from voluntarily participating COVID-19-positive patients, may adequately address privacy concerns while still capturing similar health and economic benefits as more intrusive monitoring.

Congressman Barry Loudermilk (R-GA) discussed how digital contact tracing could be more effective than manual contact tracing, but noted that it must have strong participation from people 40-60 percent adoption rate overall to be effective. He said that citizens would need to trust that their privacy would not be violated. To help establish this trust, he suggested, people would need to be able to easily determine what data would be collected, who would have access to the data and how the data would be used.

Four panelists testified at this hearing. Below is a summary of each panelists testimony, followed by an overview of some of the post-testimony questions that committee members raised:

Brian McClendon, the CEO and co-founder of the CVKey Project, discussed how privacy, disclosure and opt-in data collection impact the ability to identify and isolate those infected with COVID-19. AI and machine learning require large amounts of data. He stated that while the most valuable data to combat COVID-19 can be found in the contact-tracing interviews of infected and exposed people, difficulties exist in capturing this information. For example, attempted phone calls to reach exposed individuals may go unanswered because people often do not pick up calls from unknown numbers. Mobile apps, he said, offer a way to conduct contact tracing with greater accuracy and coverage. Mr. McClendon discussed two ways that such apps could work: (1) using GPS location or (2) via low-energy Bluetooth. For the latter, Mr. McClendon explained a method developed by two large technology companies: when a user of a digital contact tracing app tests positive for COVID-19, he or she then chooses to opt in to upload non-personally identifiable information to a state-run cloud server, which would then determine whether potential exposures have occurred and provide in-app notifications to such users.

Krutika Kuppalli, M.D., an infectious diseases physician, discussed how using contact tracing can help impede the spread of infectious diseases. She noted that it is important to remember ethical considerations involving public health information, data protection and data privacy when using these technologies.

Andre M. Perry, a fellow at the Brookings Institution, began his presentation by discussing how COVID-19 has disproportionately affected Black and Latino populations, reflecting historical inequalities and structural racism. Mr. Perry identified particular concerns regarding AI and contact tracing as they pertain to structural racism and bias. These tools, he stated, are not neutral and can either exacerbate or mitigate structural racism. To address such bias, he suggested, contact tracing should include people who have generally been excluded from systems that have provided better health and economic outcomes. Further, the use of AI tools in the healthcare arena presents the same risk as in other fields: the AI is only as good as the programmers who design it. Bias in programming can lead to flaws in technology and amplify biases in the real world. Mr. Perry stated that greater recruitment and investment with Black-owned tech firms, rigorous reviews and testing for bias and more engagement with local communities is required.

Ramesh Raskar, a professor at MIT and the founder of the PathCheck Foundation, emphasized three elements during his presentation: (1) how to augment manual contact tracing with apps; (2) how to make sure apps are privacy-preserving, inclusive, trustworthy, and built using open-source methods and nonprofits; and (3) the creation of a National Pandemic Response Service. Regarding inclusivity, Mr. Raskar noted that Congress should actively require that solutions be accessible broadly and generally; contact tracing cannot be effective only for segments of the population that have access to the latest technology.

Post-testimony questions

Chairman Foster asked about limits of privacy-preserving techniques by providing an example of a person who had been isolated for a week, then interacted with only one other person, and then later received a notification of exposure: such a person likely will know the identity of the infected person. Mr. Raskar replied that data protection has different layers: confidentiality, anonymity, and then privacy. In public health scenarios, Mr. Raskar stated that today, we only care about confidentiality and not anonymity or privacy (eventually, he commented, you will have to meet a doctor).

If we were to implement a federal contact tracing program, Representative Loudermilk asked, how would we ensure citizens that they can know what data will be used and collected, and who has access? Mr. McClendon responded that under the approach developed by the two large technology companies, data is random and stored on a personal phone until the user opts in to upload random numbers to the server. The notification determination is made on the phone and the state provides the messages. The state will not know who the exposed person is until that person opts in by calling the manual contact tracing team.

Representative Maxine Waters (D-CA) asked what developers of a mobile contact tracing technology should consider to ensure that minority communities are not further disadvantaged. Mr. Perry reiterated that AI technologies have not been tested, created, or vetted by persons of color, which has led to various biases.

Congressman Sean Casten (D-IL) asked whether AI used in contact tracing is solely backward-looking or could predict future hotspots. Mr. McClendon replied that to predict the future, you need to know the past. Manual contact tracing interviews, where an infected or exposed person describes where he or she has been, would provide significant data to include in a machine-learning algorithm, enabling tracers to predict where a hotspot might occur in the future. However, privacy issues and technological incompatibility (e.g., county and state tools that are not compatible with each other) mean that a lot of data is currently siloed and even inaccessible, impeding the ability for AI to look forward.

Read more:

Task Force on Artificial Intelligence - hearing to discuss use of AI in contact tracing - Lexology

New carbon emission tracking technology will quantify emissions of greenhouse gas, holding the energy industry accountable for its CO2 output. Backed by Google, this cutting-edge initiative will be known as Climate TRACE (Tracking Real-Time Atmospheric Carbon Emissions).

Advanced AI and machine learning now make it possible to trace greenhouse gas (GHG) emissions from factories, power plants and more. By using image processing algorithms to detect carbon emissions from power plants, AI technology makes use of the growing global satellite network to develop a more comprehensive global database of power plant activity. Because most countries self-report emissions and manually compile results, scientists often rely on data that is several years out of date. Moreover, companies often underreport carbon emissions, rendering existing data inaccurate.

Climate TRACE addresses these issues by partnering with other leaders in sustainability practicesincluding former U.S. Vice President Al Gore, WattTime, CarbonPlan, Carbon Tracker, Earthrise Alliance, Hudson Carbon, OceanMind, Rocky Mountain Institute, Blue Sky Analytics and Hypervine. The Climate TRACE coalition aims to help countries in meeting Paris Agreement targets and place the world on a path to sustainability.

The carbon tracking efforts of Climate TRACE will result in a conglomeration of data to be made available to the public, which may assist plaintiffs in climate liability cases and lead to enhanced enforcement of environmental laws. The slow pace of international climate negotiations has led to an increase in lawsuits demanding action on global warming. As of this year, 1,600 climate-related lawsuits have been filed worldwide, including 1,200 lawsuits in the United States alone. Currently, climate liability cases rely predominantly on a database run by the Carbon Disclosure Project and the Climate Accountability Institute. This database, initially released in 2013 as the Carbon Majors Report, attempts to link carbon pollution to emitters. The 2013 report pinpointed 100 producers responsible for 71% of global industrial GHG emissions. Its 2017 report, for instance, indicated that 25 corporate and state producing entities account for 51% of global industrial GHG emissions. While the Carbon Majors Report has assisted in determining the largest carbon emitters on a global scale, Climate TRACE will provide more frequent and accurate monitoring of pollutants.

Data from Climate TRACE will also help hold countries accountable to the Paris Climate Agreement, expanding upon European efforts to monitor global warming. Early last year, a space budget increase put Europe in the lead to monitor carbon from space using satellite technology. In December 2019, member governments awarded the European Space Agency $12.5 billion. This substantial increase allowed the ESA to devote $1.8 billion to Copernicus, a satellite technology program which continuously tracks Earths atmosphere. The program allowed Europe to analyze human carbon emissions regularly. With Copernicus, the ESA became the only space agency to monitor pledges made under the Paris Climate Agreement. The Climate TRACE coalitionwith members spanning across three continentswill make carbon monitoring a global effort.

Climate TRACE has created a working prototype that is currently in its developmental stages. The coalition intends to release its first version of the AI project by the summer of 2021.

[View source.]

See the original post:

Artificial Intelligence and Satellite Technology to Enhance Carbon Tracking Measures - JD Supra

The amount of uncertainty and complexity the recent economic difficulties have introduced into the business landscape has left many businesses reeling. While trying to adjust to the new normal, businesses are pressured to find new efficiencies and discover previously untapped sources of economic opportunity, makingA.I. and machine learning modelsmore important than ever to making critical and often timely business decisions.

The time for A.I. experimentation is over. We have arrived at the point where A.I. has to produce results and drive real revenue, while safeguarding the business from all of the potential risks that can jeopardize the bottom line. This expectation only becomes more challenging at a time when data is changing by the hour and previous historical patterns are not reliable. Furthermore, the complexities compound as businesses decide to rely more on A.I. in these trying times as away to stay ahead of the competition.

Newly emerging best practices, commonly referred to as MLOps (ML Operations), underpinned by a new layer of technologies with the same name, are the missing piece of the puzzle for many organizations looking to fast-track and scale their A.I. capabilities without putting their businesses at risk during this time of economic uncertainty. With MLOps technology and practices in place, businesses can bridge the inherent gap between data and operations teams, and get a scalable and governed means to deploy and manage A.I. applications in real-world production environments.

MLOps can be broken down into four key areas of the process required to derive value from machine learning, that must be well-resourced and well-understood to work in your business:

With MLOps, the goal is to make model deployment easy, regardless of which platform or language those models were created in, or where they need to eventually be deployed. MLOps essentially serves as an abstraction layer of automation whereby data teams point their models to, and where they can become managed by MLOps or Ops teams, while providing role-based visibility and actionability based on the needs of your organization.

The notion of removing ownership from the data teams as pertains to production environments, while providing them with the required visibility, allows for taking a lot of work off their plates, freeing them up to conduct their jobswhich is solving complex business problems using data.

To ensure the visibility and removal of unnecessary risk resulting from models going haywire, MLOps solutions need to deploy unique monitoring that is designed from the ground-up to monitor ML models. Such monitoring includes data drift, concept drift, feature importance, model accuracy, as well as overall service health, coupled with proactive alerting to be sent to various stakeholders using a variety of channels such as email, Slack and PagerDuty (based on severity and role). With MLOps monitoring in place, teams can deploy and manage thousands of models, and businesses will be ready to scale production A.I.

MLOps recognizes that models need to be updated frequently and seamlessly. Model lifecycle management supports the testing and warm-up of replacement models, A/B testing of new models against older versions, seamless rollout of updates, failover procedures, and full version control for simple rollback to prior model versions, all wrapped in designable approval workflows.

MLOps provides the integrations and capabilities you need to ensure consistent, repeatable and reportable processes for your models in production.Key capabilities include access control for production models and systems, such as integration to LDAP and role-based access control systems (RBAC), as well as approval flows, logging, version storage of each version of each model, and traceability of results for legal and regulatory compliance.

With the right processes, tools and training in place, businesses will be able to reap many benefits from MLOps. Itll provide insight into areas where the data might be skewed.One of the many frustrating parts of running A.I. modelsespecially right nowis that the data is constantly shifting. With MLOps, businesses can quickly identify and act on that information in order to retrain production models on newer data, using the data pipeline, algorithms, and code leveraged to create the original.

Users can also scale production while minimizing risk.Scaling A.I. across the enterprise is easier said than done. There can be numerous roadblocks that stand in the way, such as lack of communication between the IT and data science teams, or lack of visibility into A.I. outcomes. With MLOps, you can support multiple types of machine learning models created by different tools, as well as software dependencies needed by models.

Sivan Metzger is Managing Director, MLOps and Governance at DataRobot.

Membership has its benefits. Sign up for a free Dice profile, add your resume, discover great career insights and set your tech career in motion. Register now

Visit link:

Scaling AI While Navigating the Current Uncertainty - Dice Insights

App developers, students and researchers are using the transformative power of AI technologies to develop people's emotional connection to video games.

Since the 2001 introduction of the first AI digital helper Cortana in Halo, technology and AI have become pivotal to gameplay. With all the buzz around the release of a new iteration of the popular GPT-3 video game tool, IT developers are more in tune than ever with the needs of creative deployments of popular AI technology. The future of AI in video games lies in the ability of the technology to increase the human connection.

Since the dawn of chatbots and digital assistant creation, one critique has been universal: the helper is not human-like enough. This issue spans enterprises, and IT developers and startups are now developing AI that is human-like, emotional and responsive.

Christian Selchau-Hansen, CEO of enterprise software company Formation and former manager of product at social game development company Zynga, said that one of the major uses of AI in video games is the implementation of generative adversarial network (GAN) technology, image recognition and replication in character design. The ability of an algorithm to read emotion, generate emotion from text and accurately portray emotion enables a heightened level of gameplay.

"Whether it's GPT-3 or the processing and techniques of developments like deepfakes the good things that come from [these developments] are more immersive worlds," Selchau-Hansen said.

"For people to be able to interact with more immersive and complex characters, and not just have the ability to interact with them -- but create new responses based on interactions through facial expressions, language, dialogue and actions," Selchau-Hansen said.

Danny Tomsett, CEO of UneeQ, a digital assistant platform creator, said emotional connection is the creation of a feeling between you and the story or character, and AI allows for the closest representation of visual humans.

Visual representations of humans are not as good as meeting in real life, but a model that can see your emotion and vice versa lets you respond dynamically, Tomsett said.

When looking toward the future of gameplay, Selchau-Hansen imagines a world where you have something akin to a physics engine during the game -- one that controls gravity, wind-resistance and thermal conductivity -- but for emotional interactions.

"You could have an emotional engine where your interactions with a [character] can make them sad, confused, scared, jealous -- and their dialogue would spring from those emotions," Selchau-Hansen said.

The gamification of AI has been a driver of technology, with iterations of DeepBlue and AlphaGo teaching developers that perhaps the most important part of augmenting gameplay is the ability to find the spot between competition and demolition. Gamers want to be challenged but still have a chance to win because their competitors are making human-like decisions.

This idea of competition between humans and computers, a friendly tussle between players, is central to creating brand loyalty -- returning players need to be challenged with dynamic, human-like bots on the other side of the game.

Creating brand loyalty in gaming is also about eschewing flat, two-dimensional, text-based digital interfaces to unlock the power of emotion and story, Tomsett said.

Another crossover between AI and gameplay is the ability to personalize. Much like marketing campaigns and personalized promotions, the future of AI in the video game industry depends on monetizing the emotional connection between the game and the consumer. Algorithms collect data from the game -- what the player collects, what quests they follow, what skins they use -- and suggest and alter additional downloads that have the highest chance of winning over the player.

From gameplay to retail to IT personalization, AI is being used to create and strengthen the idea of product value. That value -- monetary, recreational or business-related -- is offered to the consumer to increase the likelihood of brand loyalty, Selchau-Hansen said.

While the future of AI in video games would naturally point to automation and generated text, the AI-generated video games now testing the fringes of current gaming technology also highlight their limits.

Independent designers are toying with open-source technology to use natural language generation to create virtual games without a gaming studio. Developer Nick Walton's AI Dungeon storytelling game throws you into the development of the decision tree -- your choices change the outcome and help train the game for future players. This interactive virtual role-playing game is modeled on Open-AI's machine learning-based GPT-2 natural language generator. Walton tuned more than 117 parameters and crafted neural networks to output this unique story text.

But the game reflects many of the major issues of language generation. The game is a chaotic story as the program cannot tell what you know or if you have seen a character before. Some of the language is nonsensical. There is no human emotion or human decision making.

Michael Cook, a research fellow at the Royal Academy of Engineering, developed Angelina, an AI digital assistant who is trained to develop intelligently designed videogames.

Angelina is designed to make games based on simple theme inputs and is the first system to make 3D games within the game design engine Unity. Despite the nonsensical gameplay, somewhat comical instability and terrible UX, games by Angelina are an interesting foray into what it means to train an AI or machine learning system -- it's a peek into the mechanics of how to train computational creativity. When you input a word or phrase, Angelina accesses a word association database to create a framework for creation. A "secret" theme leads to word associations like "crypt," "dark," "hidden" and "dungeon," but it can also lead to a tangled web of characters, color and ineffective jump-scares.

It's clear that the future of AI in video games lies somewhere between generated text and finely crafted human emotion to wrangle consumers.

Here is the original post:

Future of AI in video games focuses on the human connection - TechTarget

Q2 2020 saw 458 deals worth $7.2 billion

The fortunes of startups that leverage artificial intelligence have soared dramatically in recent years.

These AI-powered startups have seen quarterly investment totals rise from a few hundred rounds and a few billion dollars each quarter to 1,245 rounds and $17.3 billion in the second and third quarters of 2019, according to data from CB Insights. The rise in dollars chasing AI startups has been huge, demonstrating strong venture capital interest in the cohort.

But in recent quarters, the trend has slowed as VC deals for AI-powered startups fell off.

The Exchange explores startups, markets and money. You can read it every morning on Extra Crunch, or get The Exchange newsletter every Saturday.

A new report from the business-data company looking at the second quarter of venture capital results for global AI startups shows historically strong but declining investing rates for the upstart firms. During a pandemic and widespread recession, this is not a complete surprise; other areas of VC investment have also fallen in recent quarters. This is The Exchanges second look at quarterly data in the startup category, something partially spurred by our interest in the economics of the startups that make up the group.

The scale of decline is notable, however, as is the national breakdown of VC investment into AI. (The United States is doing better than you probably guessed, if you have only listened to politicians lately.)

Lets unpack the latest results, determine how investing patterns have changed by stage and examine how different countries compare when it comes to deal and dollar volume for AI-powered startups.

In the second quarter of 2020, global investment into AI startups fell to 458 deals worth $7.2 billion. According to the CB Insights dataset, the deal volume is the lowest for 12 quarters, or since Q2 2017 when 387 investments into AI startups were worth $4.7 billion.

Go here to see the original:

Investment in AI startups slips to three-year low - TechCrunch

Defense

Eric Schmidt speaks at a March 2020 meeting of the Defense Innovation Board in Austin, Texas. (DOD photo by EJ Hersom).

Defense technology experts think adding a military academy could be the solution to the U.S. government's tech talent gap.

"The canonical view is that the government cannot hire these people because they will get paid more in private industry," said Eric Schmidt, former Google chief and current chair of the Defense Department's Innovation Advisory Board, during a July 29 Brookings Institution virtual event.

"My experience is that people are patriotic and that you have a large number of people -- and this I think is missed in the dialogue -- a very large number of people who want to serve the country that they love. And the reason that they're not doing it is there's no program that makes sense to them."

Schmidt's comments come as the National Security Commission on Artificial Intelligence, of which he chairs, issued its second quarterly report with recommendations to Congress on how the U.S. government can invest in and implement AI technology.

One key recommendation: A national digital service academy, to act like the civilian equivalent of a military service academy to train technical talent. That institution would be paired with an effort to establish a national reserve digital corps to serve on a rotational basis.

Robert Work, former deputy secretary of defense who is now NSCAI's vice chair, said the academy would bring in people who want to serve in government and would graduate students to serve as full time federal employees at GS-7 to GS-11 pay grade. Members of the digital corps would five years at 38 days a year helping government agencies figure out how to best implement AI.

For the military, the commission wants to focus on creating a clear way to test existing service members' skills and better gauge the abilities of incoming recruits and personnel.

"We think we have a lot of talent inside the military that we just aren't aware of," Work said.

To remedy that, Work said the commission recommends a grading, via a programming proficiency test, to identify government and military workers that have software development experience. The recommendations also include a computational thinking component to the armed services' vocational aptitude battery to better identify incoming talent.

"I suspect that if we can convince the Congress to make this real and the president signs off hopefully then not only will we be successful but we'll discover that we need 10 times more. The people are there and the talent is available," Schmidt said.

About the Author

Lauren C. Williams is a staff writer at FCW covering defense and cybersecurity.

Prior to joining FCW, Williams was the tech reporter for ThinkProgress, where she covered everything from internet culture to national security issues. In past positions, Williams covered health care, politics and crime for various publications, including The Seattle Times.

Williams graduated with a master's in journalism from the University of Maryland, College Park and a bachelor's in dietetics from the University of Delaware. She can be contacted at [emailprotected], or follow her on Twitter @lalaurenista.

Click here for previous articles by Wiliams.

View original post here:

Could a new academy solve the AI talent problem? - FCW.com

The Information Commissioners Office (ICO) has published an 80-page guidance document for companies and other organisations about using artificial intelligence (AI) in line with data protection principles.

The guidance is the culmination of two years research and consultation by Reuben Binns, an associate professor in the department of Computer Science at the University of Oxford, and the ICOs AI team.

The guidance covers what the ICO thinks is best practice for data protection-compliant AI, as well as how we interpret data protection law as it applies to AI systems that process personal data. The guidance is not a statutory code. It contains advice on how to interpret relevant law as it applies to AI, and recommendations on good practice for organisational and technical measures to mitigate the risks to individuals that AI may cause or exacerbate.

It seeks to provide a framework for auditing AI, focusing on best practices for data protection compliance whether you design your own AI system, or implement one from a third party.

It embodies, it says, auditing tools and procedures that we will use in audits and investigations; detailed guidance on AI and data protection; and a toolkit designed to provide further practical support to organisations auditing the compliance of their own AI systems.

It is also an interactive document which invites further communication with the ICO.

This guidance is said to be aimed at two audiences: those with a compliance focus, such as data protection officers (DPOs), general counsel, risk managers, senior management, and the ICO's own auditors; and technology specialists, including machine learning experts, data scientists, software developers and engineers, and cyber security and IT risk managers.

It points out two security risks that can be exacerbated by AI, namely the loss or misuse of the large amounts of personal data often required to train AI systems; and software vulnerabilities to be introduced as a result of the introduction of new AI-related code and infrastructure.

For, as the guidance document points out, the standard practices for developing and deploying AI involve, by necessity, processing large amounts of data. There is therefore an inherent risk that this fails to comply with the data minimisation principle.

This, according to the GDPR [the EU General Data Protection Regulation] as glossed by former Computer Weekly journalist Warwick Ashford, requires organisations not to hold data for any longer than absolutely necessary, and not to change the use of the data from the purpose for which it was originally collected, while at the same time they must delete any data at therequest of the data subject.

While the guidance document notes that data protection and AI ethics overlap, it does not seek to provide generic ethical or design principles for your use of AI.

What is AI, in the eyes of the ICO? We use the umbrella term AI because it has become a standard industry term for a range of technologies. One prominent area of AI is machine learning, which is the use of computational techniques to create (often complex) statistical models using (typically) large quantities of data. Those models can be used to make classifications or predictions about new data points. While not all AI involves ML, most of the recent interest in AI is driven by ML in some way, whether in image recognition, speech-to-text, or classifying credit risk.

This guidance therefore focuses on the data protection challenges that ML-based AI may present, while acknowledging that other kinds of AI may give rise to other data protection challenges.

Of particular interest to the ICO is the concept of explainability in AI. The guidance goes on: in collaboration with the Alan Turing Institute we have produced guidance on how organisations can best explain their use of AI to individuals. This resulted in the Explaining decisions made with AI guidance, which was published in May 2020.

The guidance contains commentary about the distinction between a controller and a processor. It says organisations that determine the purposes and means of processing will be controllers regardless of how they are described in any contract about processing services.

This could be potentially relevant to the controversy surrounding the involvement of US data analytics company Palantirs in the NHS Data Store project, where has been repeatedly stressed that the provider is merely a processor and not a controller which is the NHS in that contractual relationship.

The guidance also discusses such matters as bias in data sets leading to AIs making biased decisions, and offers this advice, among other pointers: In cases of imbalanced training data, it may be possible to balance it out by adding or removing data about under/overrepresented subsets of the population (eg adding more data points on loan applications from women).

In cases where the training data reflects past discrimination, you could either modify the data, change the learning process, or modify the model after training.

Simon McDougall, deputy commissioner of regulatory innovation and technology at the ICO,said of the guidance: Understanding how to assess compliance with data protection principles can be challenging in the context of AI. From the exacerbated, and sometimes novel, security risks that come from the use of AI systems, to the potential for discrimination and bias in the data. It is hard for technology specialists and compliance experts to navigate their way to compliant and workable AI systems.

The guidance contains recommendations on best practice and technical measures that organisations can use to mitigate those risks caused or exacerbated by the use of this technology. It is reflective of current AI practices and is practically applicable.

Continue reading here:

ICO launches guidance on AI and data protection - ComputerWeekly.com

The drilling optimizer employs artificial intelligence to help mitigate drilling dysfunction and improve performance.

You have access to this full article to experience the outstanding content available to SPE members and JPT subscribers.

To ensure continued access to JPT's content, please Sign In, JOIN SPE, or Subscribe to JPT

Ever since the first commercial well was spudded, operators have looked for ways to drill wells faster without sacrificing safety or incurring huge costs. While saving time and money through efficient drilling is not a new concept, the more recent adoption of drilling optimization and automation services has certainly become one of the biggest drivers to achieving those goals.

As the current downturn has shown limited signs of recovery, it has continued to evolve in ways never imagined, and the effects are taking their toll in every facet of the oil and gas industry. While rigs and drilling equipment can be set aside to ride out the storm, what about the drilling teams who are working on the rigs and in remote operation centers? As these teams are being removed from the field, the expectation is that many of them wont return for a myriad of reasons. So, what happens when that experience is lost?

The exodus of seasoned crews, otherwise known as the great crew change, has been discussed for several years, but recent conditions could expedite the process. Considering the recent shutdown of rigs and the loss of personnel, the question remains whether we will see a noticeable gap in knowledge and experience once crews return to the drilling rigs in full force.

The lack of individual skills can be offset over time with hands-on experience, but a drilling crew needs to operate at the highest level possible, preferably with few to no gaps in experience. To assist the drilling process, NOVs M/D Totco division recently launched its KAIZEN intelligent drilling optimization application, which performs as an adaptive autodriller. The system features continuous learning capabilities, enabling it to provide proactive drilling dysfunction mitigation while maximizing rate of penetration (ROP) and optimizing mechanical specific energy. It also reduces human dependence in the drilling process, lowering the risk of slow or incorrect responses to drilling dysfunction. In turn, the system assesses wellbore conditions and drilling performance, then automatically applies appropriate parameters to mitigate those dysfunctions.

When faced with distinct interbedded formations, drillers often encounter drilling dysfunction due to varying formations, and optimal setpoints are required to identify and proactively mitigate dysfunction.

While drillers are inundated with large amounts of data, the system takes the human dependence away and employs artificial intelligence (AI) to continuously optimize the drilling process. Utilizing an array of machine-learning algorithms and a digital twin that is updated each second, the AI system builds a store of knowledge that the drilling application leverages to make more accurate and timely decisions. This automated parameter application approach enables the system to remove distractions from the driller so their focus can be on critical items such as keeping the crew safe and the well under control, while the system instantly responds to changing conditions and provides optimal weight on bit (WOB) and revolutions per minute (rev/min) setpoints.

The AI and machine-learning feature stores thousands of hours of processed drilling data. This capability allows the system to recommend surface parameters that deliver the best expected performance as well as select the correct dataset to mitigate changes detected in drilling dynamic behaviors.

Additionally, the systems digital twin uses an advanced, physics-based model to analyze several key dysfunctions in real time by way of distributed stress modeling, torque and drag modeling, and critical rev/min calculations. The result is a system that is capable of providing guidance to the parameter search engine, identifying which parameters to avoid. Drillstring buckling, downhole vibration (torsional, axial, and lateral), and mud motor stalls can all be diagnosed, and thus, mitigated in real time.

The systems optimization functions can be run in either advisory or control mode. The advisory mode sends recommended rev/min and WOB setpoints for the driller to implement, while control mode sends those same setpoints directly to the autodriller.

Advisory mode is typically run if a rigs control system is not fully compatible with the control mode. When a rig has a control system that is compatible with the KAIZEN command requests, control mode is activated, and the system consistently achieves a higher level of performance by continuously and automatically adjusting the setpoints.

The KAIZEN system has been successfully integrated into multiple control systems, including NOVs NOVOS reflexive drilling system. The application is currently compatible in closed-loop mode with approximately 50% of the active North American rig fleet.

The drilling systems installation comprises a minor update to the rig control layer to accept the systems commands, along with the physical installation of a KAIZEN system box. The rig must be running NOVs RigSense electronic drilling recorder, and the system box is simply plugged into both the rigs control network and the RigSense system. The system is then accessible on the RigSense screen and operated through the control system interface.

The use of KAIZEN control mode on the NOVOS control system has recently shown immediate improvements in the field.

A drilling contractor operating in the Marcellus Shale sought a solution to complete each hole section in a single-bit run while improving the ROP and reducing drilling time with less damage to the bit. The target formation showed shale with interbedded limestone as well as concentrations of iron pyrite and siderite, making for a challenging well.

The intelligent drilling optimizer applied its continuous learning capabilities, successfully demonstrating the ability to optimize the drilling program by way of dysfunction mitigation and parameter optimization. It was noted that the system handled formation changes exceptionally well and exhibited creative solutions to solve downhole vibration, resulting in smooth drilling. This allowed the system to exploit this parameter map and drill at the limit of efficiency, and the system performed those functions consistently.

The use of the drilling systems control mode on the NOVOS system showed immediate improvements. All three wells that used the system were compared against the customer-provided benchmark well; the system collectively saved the customer 38.6 drilling hours. The cumulative hourly savings translated to an average of $37,518 per well for a total savings of $112,554 over the three wells, based on a $70,000/day spread rate.

The need for greater efficiency through optimized drilling systems has pushed service companies into new and exciting areas of product development, with features like automation and AI leading the way. As the KAIZEN systems performance has shown, the application of automated control systems delivers cumulative operational benefits that continue to increase independent of the crews experience levels. As the industry looks to optimize processes in all areas of operations, intelligent systems will become the accepted, pragmatic approach to achieving safer and more efficient performance at the wellsite.

Here is the original post:

Adaptive Drilling Application Uses AI To Enhance On-Bottom Drilling Performance - Journal of Petroleum Technology

Some tech companies make a splash when they launch, others seem to bellyflop.

Genderify, a new service that promised to identify someones gender by analyzing their name, email address, or username with the help AI, looks firmly to be in the latter camp. The company launched on Product Hunt last week, but picked up a lot of attention on social media as users discovered biases and inaccuracies in its algorithms.

Type the name Meghan Smith into Genderify, for example, and the service offers the assessment: Male: 39.60%, Female: 60.40%. Change that name to Dr. Meghan Smith, however, and the assessment changes to: Male: 75.90%, Female: 24.10%. Other names prefixed with Dr produce similar results while inputs seem to generally skew male. Test@test.com is said to be 96.90 percent male, for example, while Mrs Joan smith is 94.10 percent male.

The outcry against the service has been so great that Genderify tells The Verge its shutting down altogether. If the community dont want it, maybe it was fair, said a representative via email. Genderify.com has been taken offline and its free API is no longer accessible.

Although these sorts of biases appear regularly in machine learning systems, the thoughtlessness of Genderify seems to have surprised many experts in the field. The response from Meredith Whittaker, co-founder of the AI Now Institute, which studies the impact of AI on society, was somewhat typical. Are we being trolled? she asked. Is this a psyop meant to distract the tech+justice world? Is it cringey tech April fools day already?

The problem is not that Genderify made assumptions about someones gender based on their name. People do this all the time, and sometimes make mistakes in the process. Thats why its polite to find out how people self-identify and how they want to be addressed. The problem with Genderify is that it automated these assumptions; applying them at scale while sorting individuals into a male/female binary (and so ignoring individuals who identify as non-binary) while reinforcing gender stereotypes in the process (such as: if youre a doctor youre probably a man).

The potential harm of this depends on how and where Genderify was applied. If the service was integrated into a medical chatbot, for example, its assumptions about users genders might have led to the chatbot issuing misleading medical advice.

Thankfully, Genderify didnt seem to be aiming to automate this sort of system, but was primarily designed to be a marketing tool. As Genderifys creator, Arevik Gasparyan, said on Product Hunt: Genderify can obtain data that will help you with analytics, enhancing your customer data, segmenting your marketing database, demographic statistics, etc.

In the same comment section, Gasparyan acknowledged the concerns of some users about bias and ignoring non-binary individuals, but didnt offer any concrete answers.

One user asked: Lets say I choose to identify as neither Male or Female, how do you approach this? How do you avoid gender discrimination? How are you tackling gender bias? To which Gasparyan replied that the service makes its decisions based on already existing binary name/gender databases, and that the company was actively looking into ways of improving the experience for transgender and non-binary visitors by separating the concepts of name/username/email from gender identity. Its a confusing answer given that the entire premise of Genderify is that this data is a reliable proxy for gender identity.

The company told The Verge that the service was very similar to existing companies who use databases of names to guess an individuals gender, though none of them use AI.

We understand that our model will never provide ideal results, and the algorithm needs significant improvements, but our goal was to build a self-learning AI that will not be biased as any existing solutions, said a representative via email. And to make it work, we very much relied on the feedback of transgender and non-binary visitors to help us improve our gender detection algorithms as best as possible for the LGBTQ+ community.

Update Wednesday July 29, 12:42PM ET: Story has been updated to confirm that Genderify has been shut down and to add additional comment from a representative of the firm.

Read more:

Service that uses AI to identify gender based on names looks incredibly biased - The Verge

Dan Morris, AI for Earth program director, says the most important result from the hackathon was that AI for Earth taught The Ocean Cleanup a lot about machine learning. The real value was teaching them through interaction with data scientists and engineers at Microsoft, he says.

This year, The Ocean Cleanup was named an AI for Earth grantee for its work.

Using the AI for Earth grant, weve been able to set up and run the machine learning models, De Vries says. Having the resources at our fingertips has greatly accelerated the technical progress, by taking away practical concerns and letting us focus on the development.

It allowed us to develop the vision that this is something we can do, not just for one river, but eventually for rivers across the globe.

Robin de Vries, right, of The Ocean Cleanup works with a Microsoft Global Hackathon team member in 2019.

The Ocean Cleanup is highly admired, particularly in the Netherlands, where the organization has been a symbol of pride for years, even before they became more well-known internationally, says Harry van Geijn, a digital adviser for Microsoft in the Netherlands. Van Geijn is among the Microsoft staffers there who have volunteered to help The Ocean Cleanup when it comes to computer and related support.

While its staff is relatively small with around 100 employees, they have this cause that they pursue with great tenacity and in an extremely professional way, van Geijn says. So much so that When I ask around for someone at Microsoft Netherlands to do something for The Ocean Cleanup, half the company raises their hand to say, I want to volunteer for that.

Drew Wilkinson at the 2019 Microsoft Global Hackathon in Redmond, Washington.

Wilkinson, who grew up in the hot, dry climate of the Arizona desert, spent time at sea as a volunteer for the Sea Shepherd Conservation Society, a nonprofit, marine wildlife conservation organization.

In 2018 at Microsoft, he and another coworker started an employee group, Microsofts Worldwide Sustainability Community, which has grown to more than 3,000 members globally. The group focuses on ways employees can help the company be more environmentally sustainable. Wilkinson now is a community program manager for the Worldwide Communities Program, which includes the employee group he co-founded.

Wilkinson sees the issue of plastics in the ocean as a pretty solvable problem and is excited about the work that has been done, the work that he spurred with an email.

Im not a scientist, but it doesnt take a lot of science to understand that our fate on the land is very much tied to the ocean, he says. The ocean is the planets life support system. Without a healthy ocean, we dont stand a chance either.

Top image: Some of the plastic and trash picked up onto the conveyor belt of The Ocean Cleanups Interceptor 002 on the Klang River in Malaysia. Photo credit: The Ocean Cleanup.

Go here to read the rest:

Microsoft Hackathon leads to AI and sustainability collaboration to rid plastic from rivers and the ocean - Stories - Microsoft

Two of the biggest names in biometrics have issued a joint White Paper detailing the benefits of AI-driven face authentication.

The paper, entitled Enhancing Trust with AI-Driven Biometrics, is jointly presented by FaceTec and Jumio, ID verification and authentication specialists leading the current digital onboarding trend with remote enrollment, face authentication and biometric liveness technology. Artificial Intelligence plays an enormously important role in their solutions, and accordingly, its a major focus of the paper.

Available as an e-book, the paper builds on the idea of the trust anchor explored in the previously published White Paper Trusted Identity From Start to Finish: essentially, the idea is that strong authentication starts with the foundation of an anchor document that provides reliable identity assurance. Establishing a link between the end users face and this anchor a passport or a drivers license, for example allows for the creation of a trust chain in which authentication can reliably be performed at any point going forward.

This is pretty straightforward when it comes to selfie authentication, since an end users face can easily be matched to an accompanying document. But for the highest level of security and end user convenience, sophisticated AI is required AI that can quickly map an individual in three dimensions and in a range of lighting and other environmental conditions, and that wont be thrown off by subtle differences between the end user and the image in their accompanying document.

Enhancing Trust with AI-Driven Biometrics delves into what this kind of AI looks like and how it works, including the importance of liveness detection capabilities that allow AI systems to spot fake artifacts that are meant to trick the authentication system. The paper also takes a bit of time to address broader societal concerns about AI and facial recognition, and to differentiate the more controversial systems from the kind of AI that enables convenient and secure selfie authentication.

Thanks to large-scale data breaches, the dark web, and sophisticated crime rings, the notion of relying on static databases for identity proofing many of which have also been compromised simplyno longer makes practical sense, said Dean Nicolls, Vice President of Global Marketing for Jumio, commenting on the publication.

Admittedly, switching from database pings to AI-powered identity verification does represent a pretty significant sea-change, he added. But, now is the time to dive into the deep end and adopt stronger, biometric-based approaches to identity verification and ongoing user authentication. This eBook provides some important insights about the mainstreaming of biometrics and AI and why modern enterprises are abandoning traditional ways of identity proofing new and existing customers, and deploying modern AI-based approaches that deliver significantly higher levels of identity assurance.

The paper comes at a crucial time for education in the biometrics and identity-adjacent industries, particularly around facial recognition. Growing controversy over the use of facial recognition by law enforcement is necessitating the purposeful demarcation between surveillance applications and the face-based authentication discussed in the paper.

With digital onboarding for new accounts now commonplace, and hundreds of governments working on digital identity initiatives, even the average consumer inherently knows the difference between Face Authentication they perform with their phone and surveillance software running on security cameras they dont control, said Kevin Alan Tussy, CEO of FaceTec. The former protects our privacy; the latter threatens it.

FaceTec provides the fundamental biometric technology that has allowed digital transformation to happen securely by providing our patented 3D Liveness Detection and world-leading 3D Face Matching accuracy, while continuously educating through articles and White Papers, and atLiveness.com.

At a time when this kind of technology is being embraced more widely than ever, the paper is well worth a read for those who are new to the topic of AI-driven biometrics.

(Originally posted on FindBiometrics)

See the article here:

What Are the Benefits of AI-Enhanced Biometrics? - Mobile ID World

Despite efforts throughout the United States last spring to suppress the spread of the novel coronavirus, states across the country have experienced spikes in the past several weeks. The number of confirmed COVID-19 cases in the nation has climbed to more than 3.5 million since the start of the pandemic.

Public officials in many states, including California, have now started to roll back the reopening process to help curb the spread of the virus. Eventually, state and local policymakers will be faced with deciding for a second time when and how to reopen their communities. A pair of researchers in UC Santa Barbaras College of Engineering, Xifeng Yan and Yu-Xiang Wang, have developed a novel forecasting model, inspired by artificial intelligence (AI) techniques, to provide timely information at a more localized level that officials and anyone in the public can use in their decision-making processes.

We are all overwhelmed by the data, most of which is provided at national and state levels, said Yan, an associate professor who holds the Venkatesh Narayanamurti Chair in Computer Science. Parents are more interested in what is happening in their school district and if its safe for their kids to go to school in the fall. However, there are very few websites providing that information. We aim to provide forecasting and explanations at a localized level with data that is more useful for residents and decision makers.

The forecasting project, Interventional COVID-19 Response Forecasting in Local Communities Using Neural Domain Adaption Models, received a Rapid Response Research (RAPID) grant for nearly $200,000 from the National Science Foundation (NSF).

The challenges of making sense of messy data are precisely the type of problems that we deal with every day as computer scientists working in AI and machine learning, said Wang, an assistant professor of computer science and holder of the Eugene Aas Chair. We are compelled to lend our expertise to help communities make informed decisions.

Yan and Wang developed an innovative forecasting algorithm based on a deep learning model called Transformer. The model is driven by an attention mechanism that intuitively learns how to forecast by learning what time period in the past to look at and what data is the most important and relevant.

If we are trying to forecast for a specific region, like Santa Barbara County, our algorithm compares the growth curves of COVID-19 cases across different regions over a period of time to determine the most-similar regions. It then weighs these regions to forecast cases in the target region, explained Yan.

In addition to COVID-19 data, the algorithm also draws information from the U.S. Census to factor in hyper-local details when calibrating the forecast for a local community.

The census data is very informative because it implicitly captures the culture, lifestyle, demographics and types of businesses in each local community, said Wang. When you combine that with COVID-19 data available by region, it helps us transfer the knowledge learned from one region to another, which will be useful for communities that want data on the effectiveness of interventions in order to make informed decisions.

The researchers models showed that, during the recent spike, Santa Barbara County experienced spread similar to what Mecklenburg, Wake, and Durham counties in North Carolina saw in late March and early April. Using those counties to forecast future cases in Santa Barbara County, the researchers attention-based model outperformed the most commonly used epidemiological models: the SIR (susceptible, infected, recovered) model, which describes the flow of individuals through three mutually exclusive stages; and the autoregressive model, which makes predictions based solely on a series of data points displayed over time. The AI-based model had a mean absolute percentage error (MAPE) of 0.030, compared with 0.11 for the SIR model and 0.072 with autoregression. The MAPE is a common measure of prediction accuracy in statistics.

Yan and Wang say their model forecasts more accurately because it eliminates key weaknesses associated with current models. Census data provides fine-grained details missing in existing simulation models, while the attention mechanism leverages the substantial amounts of data now available publicly.

Humans, even trained professionals, are not able to process the massive data as effectively as computer algorithms, said Wang. Our research provides tools for automatically extracting useful information from the data to simplify the picture, rather than making it more complicated.

The project, conducted in collaboration with Dr. Richard Beswick and Dr. Lynn Fitzgibbons from Cottage Hospital in Santa Barbara, will be presented later this month during the Computing Research Association (CRA) Virtual Conference. Formed in 1972 as a forum for department chairs of computer sciences departments across the country, the CRAs membership has grown to include more than 200 organizations active in computing research.

Yan and Wangs research efforts will not stop there. They plan to make their model and forecasts available to the public via a website and to collect enough data to forecast for communities across the country. We hope to forecast for every community in the country because we believe that when people are well informed with local data, they will make well-informed decisions, said Yan.

They also hope their algorithm can be used to forecast what could happen if a particular intervention is implemented at a specific time.

Because our research focuses on more fundamental aspects, the developed tools can be applied to a variety of factors, added Yan. Hopefully, the next time we are in such a situation, we will be better equipped to make the right decisions at the right time.

Read more:

COVID-19 Forecasts by AI - The UCSB Current

With the pandemic still raging, restaurants are struggling to stay in business. One way restaurants can help stave off permanent closure is to make sure whatever money they have now is being spent properly and not going to waste. One way to do that is to measure the food being used in meals and the food going to waste.

Phood is a company that uses a combination of scales, computer vision and artificial intelligence (AI) to help restaurants, cafeterias and other eateries better understand and optimize how their food inventory is being used.

There are three parts to the Phood system: a scale, a camera and a software backend. Food is placed on the scale either before going into a dish (to see how much is being used to make meals) or afterwards (to see how much waste is being generated). Theres a camera mounted above the scale that uses AI to automatically identify what each food item is.

Phoods system also integrates with a restaurants existing POS and inventory management software to track how much of a particular item is being used and who supplied it. Based on that information, restaurants can then realign both production and ordering to reduce waste. So if a cafeteria or restaurant winds up with too many leftover mixed vegetables, that point is highlighted in a Phood dashboard so the manager can take appropriate action (make less or order more).

On its website, Phood claims that its solution can help reduce food waste by 42 percent. I spoke with Phood Founder, Luc Dang, by phone this week and he said Phood can provide a cost savings of 8 10 percent. In the thin margin world of restaurants, those savings can go a long way.

Phood, which began using computer vision and AI in its product last year and has raised $100,000 in seed funding, isnt the only company fighting food waste in this manner. Winnow, which raised $12 million last year, uses a similar scale, computer vision and AI approach. LeanPath does much the same thing to help change behavior in the kitchen (e.g., less wasteful chopping of veggies or trimming of meat).

During these unpredictable times when the future of just about every eatery hangs in the balance, using a tool like Phood could not only help close down food waste, but also play its part in helping keep restaurants open.

Related

Read the rest here:

Phood Fights Food Waste with Scales, Computer Vision and AI - The Spoon

In 2018, Canton organized a risk-a-thon in which people from across Facebook spent three days competing to find the most striking way to trip up those systems. Some teams found weaknesses that Canton says convinced him the company needed to make its AI systems more robust.

One team at the contest showed that using different languages within a post could befuddle Facebooks automated hate-speech filters. A second discovered the attack used in early 2019 to spread porn on Instagram, but it wasnt considered an immediate priority to fix at the time. We forecast the future, Canton says. That inspired me that this should be my day job.

In the past year, Cantons team has probed Facebooks moderation systems. It also began working with another research team inside the company that has built a simulated version of Facebook called WW that can be used as a virtual playground to safely study bad behavior. One project is examining the circulation of posts offering goods banned on the social network, such as recreational drugs.

The red teams weightiest project aims to better understand deepfakes, imagery generated using AI that looks like it was captured with a camera. The results show that preventing AI trickery isnt easy.

Deepfake technology is becoming easier to access and has been used for targeted harassment. When Cantons group formed last year, researchers had begun to publish ideas for how to automatically filter out deepfakes. But he found some results suspicious. There was no way to measure progress, he says. Some people were reporting 99 percent accuracy, and we were like That is not true.

Facebooks AI red team launched a project called the Deepfakes Detection Challenge to spur advances in detecting AI-generated videos. It paid 4,000 actors to star in videos featuring a variety of genders, skin tones, and ages. After Facebook engineers turned some of the clips into deepfakes by swapping peoples faces around, developers were challenged to create software that could spot the simulacra.

The results, released last month, show that the best algorithm could spot deepfakes not in Facebooks collection only 65 percent of the time. That suggests Facebook isnt likely to be able to reliably detect deepfakes soon. Its a really hard problem, and its not solved, Canton says.

Cantons team is now examining the robustness of Facebook's misinformation detectors and political ad classifiers. Were trying to think very broadly about the pressing problems in the upcoming elections, he says.

Most companies using AI in their business dont have to worry as Facebook does about being accused of skewing a presidential election. But Ram Shankar Siva Kumar, who works on AI security at Microsoft, says they should still worry about people messing with their AI models. He contributed to a paper published in March that found 22 of 25 companies queried did not secure their AI systems at all. The bulk of security analysts are still wrapping their head around machine learning, he says. Phishing and malware on the box is still their main thing.

Last fall Microsoft released documentation on AI security developed in partnership with Harvard that the company uses internally to guide its security teams. It discusses threats such as model stealing, where an attacker sends repeated queries to an AI service and uses the responses to build a copy that behaves similarly. That stolen copy can either be put to work directly or used to discover flaws that allow attackers to manipulate the original, paid service.

Battista Biggio, a professor at the University of Cagliari who has been publishing studies on how to trick machine-learning systems for more than a decade, says the tech industry needs to start automating AI security checks.

Companies use batteries of preprogrammed tests to check for bugs in conventional software before it is deployed. Biggio says improving the security of AI systems in use will require similar tools, potentially building on attacks he and others have demonstrated in academic research.

That could help address the gap Kumar highlights between the numbers of deployed machine-learning algorithms and the workforce of people knowledgeable about their potential vulnerabilities. However, Biggio says biological intelligence will still be needed, since adversaries will keep inventing new tricks. The human in the loop is still going to be an important component, he says.

More Great WIRED Stories

Originally posted here:

Facebooks Red Team Hacks Its Own AI Programs - WIRED

While there's plenty to be critical about when it comes to retiring U.S. Rep. Will Hurd his records on the environment and health care, for example it's a fair bet at least some of his constituents will miss his bipartisanship.

After all, the San Antonio-area Republican co-wonAlleghenyCollege's 2018Prize for Civility in Public Life for his 30-hour "bipartisan road trip" with Beto O'Rourke, back when when the latter was just another Texas congressman and not yet a Democratic superstar.

Apparently, even in the waning months of his term, Hurd has kept up that spirit of reaching across the aisle.

The former CIA intelligence officer recentlyworked with U.S. Rep. Robin Kelly, D-Illinois, to author a detailed report on how to keep the U.S. from falling behind China on artificial intelligence. That's important, the pair argue, because AI has big implications for defense and national security.

Among the two House members' suggestions: getting the federal government to devote more money to deploying safe AI and cutting off Chinas access to AI-specific microchips.

The techie bible Wired Magazine was impressed enough with the pair's work that it devoted some serious real estate to letting them delve into their plan. Turns out Hurd and Kelly are alsodrafting a congressional resolution on their AI concerns and plan to introduce similar legislation.

Some of that I hope we get done in this Congress, and others can be taken and run with in the next Congress, Hurd told the mag.

Stay on top of San Antonio news and views. Sign up for our Weekly Headlines Newsletter.

Read the original:

San Antonio GOP Congressman Will Hurd Reaches Across the Aisle on Artificial Intelligence - San Antonio Current

Nvidia uses the new Ampere A100 GPU and the Selene Supercomputer to break MLPerf performance records

Artificial Intelligence (AI) training numbers based on the suite of the new MLPerf 0.7 benchmark performance numbers were released today (7/29/20) and once again, Nvidia takes the performance crown. Eight companies submitted numbers on systems based on both AMD and Intel CPU processors and using a variety of AI accelerators from Google, Huawei, and Nvidia. The increase in peak performance for each MLPerf benchmark by the leading platform was 2.5x or more. The new benchmark also added new tests for additional emerging AI workloads.

As a brief background, MLPerf is an organization established to develop benchmarks for effectively and consistently testing systems running a wide range of AI workloads, including training and inference processing. The organization has gained wide industry support from semiconductor and IP companies, tools vendors, systems vendors, and the research and academic communities. First launched in 2018, updates and new benchmarking results have been announced for training about once a year, even though the goal is once a quarter.

The benefit of the MLPerf benchmark is not only seeing the advancements by each vendor, but the overall advancements of the industry, especially as new workloads are added. For the latest training version 0.7, new workloads were added for Natural Language Processing (NLP) using Bidirectional Encoder Representations from Transformers (BERT), recommendation systems using Deep Learning Recommendation Machines (DRLM), and reinforcement learning using Minigo. Note that using Minigo for re-enforced may also serve as a baseline for AI gaming applications. The benchmark results are reported as either commercially available (on-premise or in the cloud), preview (products coming to the market in the next six months, or research and development (systems still in earlier development). The most important near-term results are those that are commercially available or in preview. There is an open division, but that had no material impact on the overall result.

The companies and institutions that submitted results included Alibaba, Dell EMC, Fujitsu, Google, Inspur, Intel, Nvidia, and the Shenzhen Institute of Advanced Technology. The largest number of submissions came from Nvidia, which is not surprising given that the company recently built its own supercomputer (ranked #7 in the TOP500 supercomputer list and #2 in the Green500 supercomputer list), which is based on its latest Ampere A100 GPUs. This system, called Selene allows the company considerable flexibility in test different workloads and system configurations. In the MLPerf test results, the number of GPU accelerators range from two to 2048 in the commercially available category and 4096 in the research and development category.

All of the systems were based on AMD and Intel CPUs paired with one of the following accelerators: the Google TPU v3, the Google TPU v4, the Huawei Ascend910, the Nvidia Tesla V100 (in various configurations), or the Nvidia Ampere A100. Noticeably absent were the chip startups like Cerebras, Esperanto, Groq, Graphcore, Habana (an Intel company), and SambaNova. This is especially surprising because all of these companies are listed as contributors or supporters of MLPerf. There is a long list of other AI chips startups that are also not represented. Intel submitted performance numbers but only in the preview category for its upcoming Xeon Platinum processors, not for its recently acquired Habana AI accelerators. With only Intel submitting processor-only numbers, there is nothing to compare them to and the performance is well below the systems using accelerators. It is also worth noting that Google and Nvidia were the only companies that submitted performance numbers for all the different benchmark categories, but Google only submitted complete benchmark numbers for the TPU v4, which is in the preview category.

Each benchmark is ranked in terms of the execution time of the benchmark. Because of the high number of system configurations, the best way to compare the result is to normalize the execution time to each AI accelerator by dividing the execution time by the number of accelerators. This is not perfect because the performance per accelerator does typically increase with the number of accelerators and/or some workloads appear to have optimized performance around certain system configuration, but the results appear relatively consistent even when comparing the performance numbers of systems with relatively similar numbers of accelerators. The clear winner was Nvidia. Nvidia-based systems dominated all eight benchmarks for commercially available solutions. If considering all categories, including preview, the Google TPU v4 had the fastest per accelerator execution time for recommendations.

The platforms with the top performance results for each MLPerf 0.7 benchmark

Overall, the benchmarks increased from 2.5x to 3.3x from the 0.6 version benchmark categories, which include image classification, object detection, and translation. Interestingly, Nvidias previous generation GPU, the Tesla V100 scored best in three categories non-recurrent translation, recommendation, and reinforcement learning, the latter two being new MLPerf categories. This is not completely surprising because the Ampere had significant changes in the architecture that will also improve performance in inference processing. It will be interesting to see how the Ampere A100 systems score in the next generation of inference benchmarks that should be released later this year. Another development to note is the emergence of AMD Epyc processors in the top performance benchmarks because of their presence in the new Nvidia DGX A100 systems and DGX SuperPods with Nvidias new Ampere A100 accelerators.

Summary of the top MLPerf benchmark results and the performance improvements from version 0.6 to ... [+] versions 0.7

Nvidia continues to lead the pack, not just because of its lead in GPUs, but also its leadership in complete systems, software, libraries, trained models, and other tools for AI developers. Yet, every other company offering AI chips and solutions offers comparisons to Nvidia without the supporting benchmark numbers. MLPerf is not perfect. The results should be published more than once a year and the results should include an efficiency ranking (performance/watt) for the system configurations, two points the organization is working to achieve. However, MLPerf was developed as an industry collaboration and represents the best method of evaluating AI platforms. It is time for everyone else to submit MLPerf numbers to support their claims.

Visit link:

Nvidia Trounces Google And Huawei In AI Benchmarks, Startups Nowhere To Be Found - Forbes