Optimizing prediction accuracy can come at the expense of fairness. Towards minimizing discrimination against a group, fair machine learning algorithms strive to equalize the behavior of a model across different groups, by imposing a fairness constraint on models. However, we show that giving the same importance to groups of different sizes and distributions, to counteract the effect of bias in training data, can be in conflict with robustness. We analyze data poisoning attacks against group-based fair machine learning, with the focus on equalized odds. An adversary who can control sampling or labeling for a fraction of training data, can reduce the test accuracy significantly beyond what he can achieve on unconstrained models. Adversarial sampling and adversarial labeling attacks can also worsen the model's fairness gap on test data, even though the model satisfies the fairness constraint on training data. We analyze the robustness of fair machine learning through an empirical evaluation of attacks on multiple algorithms and benchmark datasets.

Robustness to adversarial attacks has been a major focus in machine learning security [4,12,26], and has been intensively studied in the past few years [8, 15, 32]. However, the theoretical understanding of adversarial robustness is still far from being satisfactory. Research [36] have demonstrated sample complexity may be one of the obstacles in achieving high robustness under standard learning, which is a large challenge since in many real-world applications, labeled examples are few and expensive. To address this challenge, recent works [9, 37] showed that adversarial robustness can be improved by leveraging unlabeled data that come from the same distribution/domain as the original labeled training samples. Nevertheless, that is still limited due to the difficulty to make sure that the unlabeled data are exactly from the same distribution as the labeled data. For example, gathering a large number of unlabeled images that follow the same distribution as CIFAR-10 is challenging, since one would have to carefully match the same lighting conditions, backgrounds, etc. Meanwhile, out-of-domain unlabeled data can be much easier and cheaper to collect. For instance, we used Bing search engine to query a small number of keywords and, within hours, generated a new 500k dataset of noisy CIFAR-10 categories; we call this Cheap-10.

Many important questions about a model cannot be answered just by explaining how much each feature contributes to its output. To answer a broader set of questions, we generalize a popular, mathematically well-grounded explanation technique, Shapley Additive Explanations (SHAP). Our new method - Generalized Shapley Additive Explanations (G-SHAP) - produces many additional types of explanations, including: 1) General classification explanations; Why is this sample more likely to belong to one class rather than another? 2) Intergroup differences; Why do our model's predictions differ between groups of observations? 3) Model failure; Why does our model perform poorly on a given sample? We formally define these types of explanations and illustrate their practical use on real data.

You may be able to help NASA's Curiosity rover drivers better navigate Mars. Using the online tool AI4Mars to label terrain features in pictures downloaded from the Red Planet, you can train an artificial intelligence algorithm to automatically read the landscape. Is that a big rock to the left? AI4Mars, which is hosted on the citizen science website Zooniverse, lets you draw boundaries around terrain and choose one of four labels. Those labels are key to sharpening the Martian terrain-classification algorithm called SPOC (Soil Property and Object Classification). Developed at NASA's Jet Propulsion Laboratory, which has managed all of the agency's Mars rover missions, SPOC labels various terrain types, creating a visual map that helps mission team members determine which paths to take.

The Government is still using a late 20th Century mindset when Britain chose to intervene in the Balkans, Afghanistan and Iraq. "But in 2010, the world began to change," he said. Britain dropped the ball on state competitors, leading to a failure to keep up with Russian and Chinese military technology." He pointed to a defence programme which "isn't enough to allow Britain to make an independent contribution within Nato or any other coalition. "We can't afford even to pay for the things we've kept, and we choose not to see that the list ignores things the Armed Forces knows it will need to go to war at a state level.

Three images from the tool called AI4Mars show different kinds of Martian terrain as seen by NASA's Curiosity rover. By drawing borders around terrain features and assigning one of four labels to them, you can help train an algorithm that will automatically identify terrain types for Curiosity's rover planners. Using an online tool to label Martian terrain types, you can train an artificial intelligence algorithm that could improve the way engineers guide the Curiosity rover. You may be able to help NASA's Curiosity rover drivers better navigate Mars. Using the online tool AI4Mars to label terrain features in pictures downloaded from the Red Planet, you can train an artificial intelligence algorithm to automatically read the landscape.

Introduction to machine learning: An early draft of a proposed textbook.

A population of voters must elect representatives among themselves to decide on a sequence of possibly unforeseen binary issues. Voters care only about the final decision, not the elected representatives. The disutility of a voter is proportional to the fraction of issues, where his preferences disagree with the decision. While an issue-by-issue vote by all voters would maximize social welfare, we are interested in how well the preferences of the population can be approximated by a small committee. We show that a k-sortition (a random committee of k voters with the majority vote within the committee) leads to an outcome within the factor 1+O(1/k) of the optimal social cost for any number of voters n, any number of issues $m$, and any preference profile. For a small number of issues m, the social cost can be made even closer to optimal by delegation procedures that weigh committee members according to their number of followers. However, for large m, we demonstrate that the k-sortition is the worst-case optimal rule within a broad family of committee-based rules that take into account metric information about the preference profile of the whole population.

Understanding predictions made by deep neural networks is notoriously difficult, but also crucial to their dissemination. As all ML-based methods, they are as good as their training data, and can also capture unwanted biases. While there are tools that can help understand whether such biases exist, they do not distinguish between correlation and causation, and might be ill-suited for text-based models and for reasoning about high level language concepts. A key problem of estimating the causal effect of a concept of interest on a given model is that this estimation requires the generation of counterfactual examples, which is challenging with existing generation technology. To bridge that gap, we propose CausaLM, a framework for producing causal model explanations using counterfactual language representation models. Our approach is based on fine-tuning of deep contextualized embedding models with auxiliary adversarial tasks derived from the causal graph of the problem. Concretely, we show that by carefully choosing auxiliary adversarial pre-training tasks, language representation models such as BERT can effectively learn a counterfactual representation for a given concept of interest, and be used to estimate its true causal effect on model performance. A byproduct of our method is a language representation model that is unaffected by the tested concept, which can be useful in mitigating unwanted bias ingrained in the data.

With the growing use of AI in highly consequential domains, the quantification and removal of bias with respect to protected attributes, such as gender, race, etc., is becoming increasingly important. While quantifying bias is essential, sometimes the needs of a business (e.g., hiring) may require the use of certain features that are critical in a way that any bias that can be explained by them might need to be exempted. E.g., a standardized test-score may be a critical feature that should be weighed strongly in hiring even if biased, whereas other features, such as zip code may be used only to the extent that they do not discriminate. In this work, we propose a novel information-theoretic decomposition of the total bias (in a counterfactual sense) into a non-exempt component that quantifies the part of the bias that cannot be accounted for by the critical features, and an exempt component which quantifies the remaining bias. This decomposition allows one to check if the bias arose purely due to the critical features (inspired from the business necessity defense of disparate impact law) and also enables selective removal of the non-exempt component if desired. We arrive at this decomposition through examples that lead to a set of desirable properties (axioms) that any measure of non-exempt bias should satisfy. We demonstrate that our proposed counterfactual measure satisfies all of them. Our quantification bridges ideas of causality, Simpson's paradox, and a body of work from information theory called Partial Information Decomposition. We also obtain an impossibility result showing that no observational measure of non-exempt bias can satisfy all of the desirable properties, which leads us to relax our goals and examine observational measures that satisfy only some of these properties. We then perform case studies to show how one can train models while reducing non-exempt bias.