We investigate the power of censoring techniques, first developed for learning {\em fair representations}, to address domain generalization. We examine {\em adversarial} censoring techniques for learning invariant representations from multiple "studies" (or domains), where each study is drawn according to a distribution on domains. The mapping is used at test time to classify instances from a new domain. In many contexts, such as medical forecasting, domain generalization from studies in populous areas (where data are plentiful), to geographically remote populations (for which no training data exist) provides fairness of a different flavor, not anticipated in previous work on algorithmic fairness. We study an adversarial loss function for $k$ domains and precisely characterize its limiting behavior as $k$ grows, formalizing and proving the intuition, backed by experiments, that observing data from a larger number of domains helps. The limiting results are accompanied by non-asymptotic learning-theoretic bounds. Furthermore, we obtain sufficient conditions for good worst-case prediction performance of our algorithm on previously unseen domains. Finally, we decompose our mappings into two components and provide a complete characterization of invariance in terms of this decomposition. To our knowledge, our results provide the first formal guarantees of these kinds for adversarial invariant domain generalization.

Individual fairness is an intuitive definition of algorithmic fairness that addresses some of the drawbacks of group fairness. Despite its benefits, it depends on a task specific fair metric that encodes our intuition of what is fair and unfair for the ML task at hand, and the lack of a widely accepted fair metric for many ML tasks is the main barrier to broader adoption of individual fairness. In this paper, we present two simple ways to learn fair metrics from a variety of data types. We show empirically that fair training with the learned metrics leads to improved fairness on three machine learning tasks susceptible to gender and racial biases. We also provide theoretical guarantees on the statistical performance of both approaches.

Explainable AI provides insight into the "why" for model predictions, offering potential for users to better understand and trust a model, and to recognize and correct AI predictions that are incorrect. Prior research on human and explainable AI interactions has focused on measures such as interpretability, trust, and usability of the explanation. Whether explainable AI can improve actual human decision-making and the ability to identify the problems with the underlying model are open questions. Using real datasets, we compare and evaluate objective human decision accuracy without AI (control), with an AI prediction (no explanation), and AI prediction with explanation. We find providing any kind of AI prediction tends to improve user decision accuracy, but no conclusive evidence that explainable AI has a meaningful impact. Moreover, we observed the strongest predictor for human decision accuracy was AI accuracy and that users were somewhat able to detect when the AI was correct versus incorrect, but this was not significantly affected by including an explanation. Our results indicate that, at least in some situations, the "why" information provided in explainable AI may not enhance user decision-making, and further research may be needed to understand how to integrate explainable AI into real systems.

Classifiers are often tested on relatively small data sets, which should lead to uncertain performance metrics. Nevertheless, these metrics are usually taken at face value. We present an approach to quantify the uncertainty of classification performance metrics, based on a probability model of the confusion matrix. Application of our approach to classifiers from the scientific literature and a classification competition shows that uncertainties can be surprisingly large and limit performance evaluation. In fact, some published classifiers are likely to be misleading. The application of our approach is simple and requires only the confusion matrix. It is agnostic of the underlying classifier. Our method can also be used for the estimation of sample sizes that achieve a desired precision of a performance metric.

The New York City Council today voted in favor of the Public Oversight of Surveillance Technology (POST) Act, a bill that requires the New York City Police Department (NYPD) to disclose their use of surveillance technologies. The POST Act also mandates that the NYPD develop policies on how it deploys those tools, as well as establish oversight of the department's surveillance programs to ensure they remain compliant. The passage of the POST Act, a three-year-old piece of legislation written with input from local activist organization Surveillance Technology Oversight Project (STOP), comes as cities around the country reexamine law enforcement policies following widespread demonstrations against abuse. Residents and activists on Tuesday urged the Detroit City Council to reject a contract that would extend the city police's use of facial recognition technology. On Wednesday, racial justice and civil liberties groups called on members of the U.S. Congress to end funding for surveillance technology law enforcement is using to spy on demonstrators.

Most animal-inspired robots are designed to move quickly, but Georgia Tech's latest is just the opposite. Their newly developed SlothBot is built to study animals, plants and the overall environment below them by moving as little as possible. It inches along overhead cables only when necessary, charging itself with solar panels to monitor factors like carbon dioxide levels and weather for as long as possible -- possibly for years. It even crawls toward the sunlight to ensure it stays charged. The 3D-printed shell helps SlothBot blend in (at least in areas where sloths live) while sheltering its equipment from the rain.

ABBYY, a Digital Intelligence company, recently launched a global initiative to promote the development of trustworthy artificial intelligence (AI) technology. As AI becomes ubiquitous across consumer and enterprise high-value and large-scale uses and more open source tools become available for digitizing data, the ethical use of accessing and training data is imperative. In a recent study, there were eight themes that kept recirculating. Privacy and accountability were two of the most commonly appearing ethical themes, as was AI safety/security. Transparency/explainability was also a commonly cited goal, with making AI algorithms more explainable being classified as extremely important.

Andrew Pery, is the Ethics Evangelist at ABBYY, a digital intelligence company. They empower organizations to access the valuable, yet often hard to attain, insight into their operations that enables true business transformation. ABBYY recently released a Global Initiative Promoting the Development of Trustworthy Artificial Intelligence. We decided to ask Andrew questions regarding ethics in AI, abuses of AI, and what the AI industry can do about these concerns moving forward. What is it that initially instigated your interest in AI ethics? What initially sparked my interest in AI ethics was a deep interest in the intersection of law and AI technology.

At The Good AI, we strongly believe Artificial Intelligence (AI) should be inclusive and celebrate diversity. However, AI is also the reflector of its creators and this translates into the reproduction of certain biases into AI products related to race, gender or sexual orientation among others. The following article from the MIT Technology Review explains how. In the light of this, the tech industry has an important responsibility towards society, and the death of George Floyd at the hands of a city police officer in Minneapolis, USA on 25 May 2020, -one in a long series of racists attacks against African Americans -, should urge us to take action. We need to make sure we are not perpetuating and letting racism or any other kind of discrimination take roots in our AI systems.

A robot dog from Boston Dynamics is now officially available to purchase. Spot, as the machine has been dubbed, will cost $74,500 (approximately £60,000). The canine droid is only available to customers in the United States at the moment, after they make a $1,000 deposit. It is capable of climbing stairs and crossing rough terrain, with the company sending the mechanical pooch into dangerous environments to carry payloads from place to place or collect data from the site. Users can control spot through its controller, which "easy access" to the robot's body posing, walking gaits, obstacle avoidance, and local navigation. Spot can also be set to follow predefined routes.