This paper introduces the Shepherd Test, a new conceptual test for assessing the moral and relational dimensions of superintelligent artificial agents. The test is inspired by human interactions with animals, where ethical considerations about care, manipulation, and consumption arise in contexts of asymmetric power and self-preservation. We argue that AI crosses an important, and potentially dangerous, threshold of intelligence when it exhibits the ability to manipulate, nurture, and instrumentally use less intelligent agents, while also managing its own survival and expansion goals. This includes the ability to weigh moral trade-offs between self-interest and the well-being of subordinate agents. The Shepherd Test thus challenges traditional AI evaluation paradigms by emphasizing moral agency, hierarchical behavior, and complex decision-making under existential stakes. We argue that this shift is critical for advancing AI governance, particularly as AI systems become increasingly integrated into multi-agent environments. We conclude by identifying key research directions, including the development of simulation environments for testing moral behavior in AI, and the formalization of ethical manipulation within multi-agent systems.

As chatbots driven by large language models (LLMs) are increasingly deployed in everyday contexts, their ability to recover from errors through effective apologies is critical to maintaining user trust and satisfaction. In a preregistered study with Prolific workers ( N = 162), we examine user preferences for three types of apologies ( rote , explanatory , and empathic ) issued in response to three categories of common LLM mistakes ( bias , unfounded fabrication , and factual errors ). We designed a pairwise experiment in which participants evaluated chatbot responses consisting of an initial error, a subsequent apology, and a resolution. Explanatory apologies were generally preferred, but this varied by context and user. In the bias scenario, empathic apologies were favored for acknowledging emotional impact, while hallucinations, though seen as serious, elicited no clear preference, reflecting user uncertainty. Our findings show the complexity of effective apology in AI systems. We discuss key insights such as personalization and calibration that future systems must navigate to meaningfully repair trust.

Developing value-aligned agents is a complex undertaking and an ongoing challenge in the field of AI. Indeed, designing Large Language Models (LLMs) that can balance multiple possibly conflicting moral values based on the context is a problem of paramount importance. In this paper, we propose a system that performs contextual value alignment based on contextual aggregation of possible responses. This aggregation is achieved by integrating a subset of possible LLM responses that are best suited to a user's input while taking into account features extracted about the user's moral preferences. The proposed system trained using the Moral Integrity Corpus displays better alignment to human values than state-of-the-art baselines.

Erik Miehling, Michael Desmond, Karthikeyan Natesan Ramamurthy, Elizabeth M. Daly, Kush R. Varshney, Eitan Farchi, Pierre Dognin, Jesus Rios, Djallel Bouneffouf, Miao Liu, Prasanna Sattigeri. Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2025.

Graph neural networks are powerful graph representation learners in which node representations are highly influenced by features of neighboring nodes. Prior work on individual fairness in graphs has focused only on node features rather than structural issues. However, from the perspective of fairness in high-stakes applications, structural fairness is also important, and the learned representations may be systematically and undesirably biased against unprivileged individuals due to a lack of structural awareness in the learning process. In this work, we propose a pre-processing bias mitigation approach for individual fairness that gives importance to local and global structural features. We mitigate the local structure discrepancy of the graph embedding via a locally fair PageRank method. We address the global structure disproportion between pairs of nodes by introducing truncated singular value decomposition-based pairwise node similarities. Empirically, the proposed pre-processed fair structural features have superior performance in individual fairness metrics compared to the state-of-the-art methods while maintaining prediction performance.

Prior work has explicated the coloniality of artificial intelligence (AI) development and deployment through mechanisms such as extractivism, automation, sociological essentialism, surveillance, and containment. However, that work has not engaged much with alignment: teaching behaviors to a large language model (LLM) in line with desired values, and has not considered a mechanism that arises within that process: moral absolutism---a part of the coloniality of knowledge. Colonialism has a history of altering the beliefs and values of colonized peoples; in this paper, I argue that this history is recapitulated in current LLM alignment practices and technologies. Furthermore, I suggest that AI alignment be decolonialized using three forms of openness: openness of models, openness to society, and openness to excluded knowledges. This suggested approach to decolonial AI alignment uses ideas from the argumentative moral philosophical tradition of Hinduism, which has been described as an open-source religion. One concept used is viśeṣa-dharma, or particular context-specific notions of right and wrong. At the end of the paper, I provide a suggested reference architecture to work toward the proposed framework.

Legal financial obligations (LFOs) such as court fees and fines are commonly levied on individuals who are convicted of crimes. It is expected that LFO amounts should be similar across social, racial, and geographic subpopulations convicted of the same crime. This work analyzes the distribution of LFOs in Jefferson County, Alabama and highlights disparities across different individual and neighborhood demographic characteristics. Data-driven discovery methods are used to detect subpopulations that experience higher LFOs than the overall population of offenders. Critically, these discovery methods do not rely on pre-specified groups and can assist scientists and researchers investigate socially-sensitive hypotheses in a disciplined way. Some findings, such as individuals who are Black, live in Black-majority neighborhoods, or live in low-income neighborhoods tending to experience higher LFOs, are commensurate with prior expectation. However others, such as high LFO amounts in worthless instrument (bad check) cases experienced disproportionately by individuals living in affluent majority-white neighborhoods, are more surprising. More broadly than the specific findings, the methodology is shown to identify structural weaknesses that undermine the goal of equal justice under law that can be addressed through policy interventions.

We introduce equi-tuning, a novel fine-tuning method that transforms (potentially non-equivariant) pretrained models into group equivariant models while incurring minimum L_2 loss between the feature representations of the pretrained and the equivariant models. Large pretrained models can be equi-tuned for different groups to satisfy the needs of various downstream tasks. Equi-tuned models benefit from both group equivariance as an inductive bias and semantic priors from pretrained models. We provide applications of equi-tuning on three different tasks: image classification, compositional generalization in language, and fairness in natural language generation (NLG). We also provide a novel group-theoretic definition for fairness in NLG. The effectiveness of this definition is shown by testing it against a standard empirical method of fairness in NLG. We provide experimental results for equi-tuning using a variety of pretrained models: Alexnet, Resnet, VGG, and Densenet for image classification; RNNs, GRUs, and LSTMs for compositional generalization; and GPT2 for fairness in NLG. We test these models on benchmark datasets across all considered tasks to show the generality and effectiveness of the proposed method.

The use of machine learning models in consequential decision making often exacerbates societal inequity, in particular yielding disparate impact on members of marginalized groups defined by race and gender. The area under the ROC curve (AUC) is widely used to evaluate the performance of a scoring function in machine learning, but is studied in algorithmic fairness less than other performance metrics. Due to the pairwise nature of the AUC, defining an AUC-based group fairness metric is pairwise-dependent and may involve both intra-group and inter-group AUCs. Importantly, considering only one category of AUCs is not sufficient to mitigate unfairness in AUC optimization. In this paper, we propose a minimax learning and bias mitigation framework that incorporates both intra-group and inter-group AUCs while maintaining utility. Based on this Rawlsian framework, we design an efficient stochastic optimization algorithm and prove its convergence to the minimum group-level AUC. We conduct numerical experiments on both synthetic and real-world datasets to validate the effectiveness of the minimax framework and the proposed optimization algorithm.

As artificial intelligence and machine learning algorithms become increasingly prevalent in society, multiple stakeholders are calling for these algorithms to provide explanations. At the same time, these stakeholders, whether they be affected citizens, government regulators, domain experts, or system developers, have different explanation needs. To address these needs, in 2019, we created AI Explainability 360, an open source software toolkit featuring ten diverse and state-of-the-art explainability methods and two evaluation metrics. This paper examines the impact of the toolkit with several case studies, statistics, and community feedback. The different ways in which users have experienced AI Explainability 360 have resulted in multiple types of impact and improvements in multiple metrics, highlighted by the adoption of the toolkit by the independent LF AI & Data Foundation. The paper also describes the flexible design of the toolkit, examples of its use, and the significant educational material and documentation available to its users.

In consequential decision-making applications, mitigating unwanted biases in machine learning models that yield systematic disadvantage to members of groups delineated by sensitive attributes such as race and gender is one key intervention to strive for equity. Focusing on demographic parity and equality of opportunity, in this paper we propose an algorithm that improves the fairness of a pre-trained classifier by simply dropping carefully selected training data points. We select instances based on their influence on the fairness metric of interest, computed using an infinitesimal jackknife-based approach. The dropping of training points is done in principle, but in practice does not require the model to be refit. Crucially, we find that such an intervention does not substantially reduce the predictive performance of the model but drastically improves the fairness metric. Through careful experiments, we evaluate the effectiveness of the proposed approach on diverse tasks and find that it consistently improves upon existing alternatives.

Interpretable and explainable machine learning has seen a recent surge of interest. We focus on safety as a key motivation behind the surge and make the relationship between interpretability and safety more quantitative. Toward assessing safety, we introduce the concept of maximum deviation via an optimization problem to find the largest deviation of a supervised learning model from a reference model regarded as safe. We then show how interpretability facilitates this safety assessment. For models including decision trees, generalized linear and additive models, the maximum deviation can be computed exactly and efficiently. For tree ensembles, which are not regarded as interpretable, discrete optimization techniques can still provide informative bounds. For a broader class of piecewise Lipschitz functions, we leverage the multi-armed bandit literature to show that interpretability produces tighter (regret) bounds on the maximum deviation. We present case studies, including one on mortgage approval, to illustrate our methods and the insights about models that may be obtained from deviation maximization.

The use of machine learning (ML) in high-stakes societal decisions has encouraged the consideration of fairness throughout the ML lifecycle. Although data integration is one of the primary steps to generate high-quality training data, most of the fairness literature ignores this stage. In this work, we consider fairness in the integration component of data management, aiming to identify features that improve prediction without adding any bias to the dataset. We work under the causal fairness paradigm. Without requiring the underlying structural causal model a priori, we propose an approach to identify a sub-collection of features that ensure fairness of the dataset by performing conditional independence tests between different subsets of features. We use group testing to improve the complexity of the approach. We theoretically prove the correctness of the proposed algorithm and show that sublinear conditional independence tests are sufficient to identify these variables. A detailed empirical evaluation is performed on real-world datasets to demonstrate the efficacy and efficiency of our technique.

Stochastic gradient descent ascent (SGDA) and its variants have been the workhorse for solving minimax problems. However, in contrast to the well-studied stochastic gradient descent (SGD) with differential privacy (DP) constraints, there is little work on understanding the generalization (utility) of SGDA with DP constraints. In this paper, we use the algorithmic stability approach to establish the generalization (utility) of DP-SGDA in different settings. In particular, for the convex-concave setting, we prove that the DP-SGDA can achieve an optimal utility rate in terms of the weak primal-dual population risk in both smooth and non-smooth cases. To our best knowledge, this is the first-ever-known result for DP-SGDA in the non-smooth case. We further provide its utility analysis in the nonconvex-strongly-concave setting which is the first-ever-known result in terms of the primal population risk. The convergence and generalization results for this nonconvex setting are new even in the non-private setting. Finally, numerical experiments are conducted to demonstrate the effectiveness of DP-SGDA for both convex and nonconvex cases.

Several strands of research have aimed to bridge the gap between artificial intelligence (AI) and human decision-makers in AI-assisted decision-making, where humans are the consumers of AI model predictions and the ultimate decision-makers in high-stakes applications. However, people's perception and understanding are often distorted by their cognitive biases, such as confirmation bias, anchoring bias, availability bias, to name a few. In this work, we use knowledge from the field of cognitive science to account for cognitive biases in the human-AI collaborative decision-making setting, and mitigate their negative effects on collaborative performance. To this end, we mathematically model cognitive biases and provide a general framework through which researchers and practitioners can understand the interplay between cognitive biases and human-AI accuracy. We then focus specifically on anchoring bias, a bias commonly encountered in human-AI collaboration. We implement a time-based de-anchoring strategy and conduct our first user experiment that validates its effectiveness in human-AI collaborative decision-making. With this result, we design a time allocation strategy for a resource-constrained setting that achieves optimal human-AI collaboration under some assumptions. We, then, conduct a second user experiment which shows that our time allocation strategy with explanation can effectively de-anchor the human and improve collaborative performance when the AI model has low confidence and is incorrect.

Thousands of scientific publications discuss evidence on the efficacy of non-cancer generic drugs being tested for cancer. However, trying to manually identify and extract such evidence is intractable at scale. We introduce a natural language processing pipeline to automate the identification of relevant studies and facilitate the extraction of therapeutic associations between generic drugs and cancers from PubMed abstracts. We annotate datasets of drug-cancer evidence and use them to train models to identify and characterize such evidence at scale. To make this evidence readily consumable, we incorporate the results of the models in a web application that allows users to browse documents and their extracted evidence. Users can provide feedback on the quality of the evidence extracted by our models. This feedback is used to improve our datasets and the corresponding models in a continuous integration system. We describe the natural language processing pipeline in our application and the steps required to deploy services based on the machine learning models.

Prior work on fairness in machine learning has focused on settings where all the information needed about each individual is readily available. However, in many applications, further information may be acquired at a cost. For example, when assessing a customer's creditworthiness, a bank initially has access to a limited set of information but progressively improves the assessment by acquiring additional information before making a final decision. In such settings, we posit that a fair decision maker may want to ensure that decisions for all individuals are made with similar expected error rate, even if the features acquired for the individuals are different. We show that a set of carefully chosen confidence thresholds can not only effectively redistribute an information budget according to each individual's needs, but also serve to address individual and group fairness concerns simultaneously. Finally, using two public datasets, we confirm the effectiveness of our methods and investigate the limitations.

In recent years there has been a considerable amount of research on local post hoc explanations for neural networks. However, work on building interpretable neural architectures has been relatively sparse. In this paper, we present a novel neural architecture, CoFrNet, inspired by the form of continued fractions which are known to have many attractive properties in number theory, such as fast convergence of approximations to real numbers. We show that CoFrNets can be efficiently trained as well as interpreted leveraging their particular functional form. Moreover, we prove that such architectures are universal approximators based on a proof strategy that is different than the typical strategy used to prove universal approximation results for neural networks based on infinite width (or depth), which is likely to be of independent interest. We experiment on nonlinear synthetic functions and are able to accurately model as well as estimate feature attributions and even higher order terms in some cases, which is a testament to the representational power as well as interpretability of such architectures. To further showcase the power of CoFrNets, we experiment on seven real datasets spanning tabular, text and image modalities, and show that they are either comparable or significantly better than other interpretable models and multilayer perceptrons, sometimes approaching the accuracies of state-of-the-art models.

In the current era, people and society have grown increasingly reliant on artificial intelligence (AI) technologies. AI has the potential to drive us towards a future in which all of humanity flourishes. It also comes with substantial risks for oppression and calamity. Discussions about whether we should (re)trust AI have repeatedly emerged in recent years and in many quarters, including industry, academia, healthcare, services, and so on. Technologists and AI researchers have a responsibility to develop trustworthy AI systems. They have responded with great effort to design more responsible AI algorithms. However, existing technical solutions are narrow in scope and have been primarily directed towards algorithms for scoring or classification tasks, with an emphasis on fairness and unwanted bias. To build long-lasting trust between AI and human beings, we argue that the key is to think beyond algorithmic fairness and connect major aspects of AI that potentially cause AI’s indifferent behavior. In this survey, we provide a systematic framework of Socially Responsible AI Algorithms that aims to examine the subjects of AI indifference and the need for socially responsible AI algorithms, define the objectives, and introduce the means by which we may achieve these objectives. We further discuss how to leverage this framework to improve societal well-being through protection, information, and prevention/mitigation. This article appears in the special track on AI & Society.

We introduce a novel event-driven continuous time Bayesian network (ECTBN) representation to model situations where a system's state variables could be influenced by occurrences of events of various types. In this way, the model parameters and graphical structure capture not only potential “causal” dynamics of system evolution but also the influence of event occurrences that may be interventions. We propose a greedy search procedure for structure learning based on the BIC score for a special class of ECTBNs, showing that it is asymptotically consistent and also effective for limited data. We demonstrate the power of the representation by applying it to model paths out of poverty for clients of CityLink Center, an integrated social service provider in Cincinnati, USA. Here the ECTBN formulation captures the effect of classes/counseling sessions on an individual's life outcome areas such as education, transportation, employment and financial education.

More than 200 generic drugs approved by the U.S. Food and Drug Administration for non-cancer indications have shown promise for treating cancer. Due to their long history of safe patient use, low cost, and widespread availability, repurposing of these drugs represents a major opportunity to rapidly improve outcomes for cancer patients and reduce healthcare costs. In many cases, there is already evidence of efficacy for cancer, but trying to manually extract such evidence from the scientific literature is intractable. In this emerging applications paper, we introduce a system to automate non-cancer generic drug evidence extraction from PubMed abstracts. Our primary contribution is to define the natural language processing pipeline required to obtain such evidence, comprising the following modules: querying, filtering, cancer type entity extraction, therapeutic association classification, and study type classification. Using the subject matter expertise on our team, we create our own datasets for these specialized domain-specific tasks. We obtain promising performance in each of the modules by utilizing modern language processing techniques and plan to treat them as baseline approaches for future improvement of individual components.

Machine learning models are prone to biased decisions due to biases in the datasets they are trained on. In this paper, we introduce a novel data augmentation technique to create a fairer dataset for model training that could also lend itself to understanding the type of bias existing in the dataset i.e. if bias arises from a lack of representation for a particular group (sampling bias) or if it arises because of human bias reflected in the labels (prejudice based bias). Given a dataset involving a protected attribute with a privileged and unprivileged group, we create an "ideal world'' dataset: for every data sample, we create a new sample having the same features (except the protected attribute(s)) and label as the original sample but with the opposite protected attribute value. The synthetic data points are sorted in order of their proximity to the original training distribution and added successively to the real dataset to create intermediate datasets. We theoretically show that two different notions of fairness: statistical parity difference (independence) and average odds difference (separation) always change in the same direction using such an augmentation. We also show submodularity of the proposed fairness-aware augmentation approach that enables an efficient greedy algorithm. We empirically study the effect of training models on the intermediate datasets and show that this technique reduces the two bias measures while keeping the accuracy nearly constant for three datasets. We then discuss the implications of this study on the disambiguation of sample bias and prejudice based bias and discuss how pre-processing techniques should be evaluated in general. The proposed method can be used by policy makers who want to use unbiased datasets to train machine learning models for their applications to add a subset of synthetic points to an extent that they are comfortable with to mitigate unwanted bias.

Today, AI is increasingly being used in many high-stakes decision-making applications in which fairness is an important concern. Already, there are many examples of AI being biased and making questionable and unfair decisions. The AI research community has proposed many methods to measure and mitigate unwanted biases, but few of them involve inputs from human policy makers. We argue that because different fairness criteria sometimes cannot be simultaneously satisfied, and because achieving fairness often requires sacrificing other objectives such as model accuracy, it is key to acquire and adhere to human policy makers' preferences on how to make the tradeoff among these objectives. In this paper, we propose a framework and some exemplar methods for eliciting such preferences and for optimizing an AI model according to these preferences.

Overlap between treatment groups is required for non-parametric estimation of causal effects. If a subgroup of subjects always receives the same intervention, we cannot estimate the effect of intervention changes on that subgroup without further assumptions. When overlap does not hold globally, characterizing local regions of overlap can inform the relevance of causal conclusions for new subjects, and can help guide additional data collection. To have impact, these descriptions must be interpretable for downstream users who are not machine learning experts, such as policy makers. We formalize overlap estimation as a problem of finding minimum volume sets subject to coverage constraints and reduce this problem to binary classification with Boolean rule classifiers. We then generalize this method to estimate overlap in off-policy policy evaluation. In several real-world applications, we demonstrate that these rules have comparable accuracy to black-box estimators and provide intuitive and informative explanations that can inform policy making.

We investigate the effect of variational autoencoder (VAE) based data anonymization and its ability to preserve anomalous subgroup properties. We present a Utility Guaranteed Deep Privacy (UGDP) system which casts existing anomalous pattern detection methods as a new utility measure for data synthesis. UGDP's approach shows that properties of an anomalous subset of records, identified in the original data set, are preserved through the anonymization of a VAE. This is despite the newly generated records being completely synthetic. More specifically, the Bias-Scan algorithm identifies a subgroup of records that are consistently over- (or under-) risked by a black-box classifier as an area of 'poor fit'. This scanning process is applied on both pre- and post- VAE synthesized data. The areas of poor fit (i.e. anomalous records) persist in both settings. We evaluate our approach using publicly available datasets from the financial industry. Our evaluation confirmed that the approach is able to produce synthetic datasets that preserved a high level of subgroup differentiation as identified initially in the original dataset. Such a distinction was maintained while having distinctly different records between the synthetic and original dataset.

The standard risk minimization paradigm of machine learning is brittle when operating in environments whose test distributions are different from the training distribution due to spurious correlations. Training on data from many environments and finding invariant predictors reduces the effect of spurious features by concentrating models on features that have a causal relationship with the outcome. In this work, we pose such invariant risk minimization as finding the Nash equilibrium of an ensemble game among several environments. By doing so, we develop a simple training algorithm that uses best response dynamics and, in our experiments, yields similar or better empirical accuracy with much lower variance than the challenging bi-level optimization problem of Arjovsky et al. (2019). One key theoretical contribution is showing that the set of Nash equilibria for the proposed game are equivalent to the set of invariant predictors for any finite number of environments, even with nonlinear classifiers and transformations. As a result, our method also retains the generalization guarantees to a large set of environments shown in Arjovsky et al. (2019). The proposed algorithm adds to the collection of successful game-theoretic machine learning algorithms such as generative adversarial networks.

A trade-off between accuracy and fairness is almost taken as a given in the existing literature on fairness in machine learning. Yet, it is not preordained that accuracy should decrease with increased fairness. Novel to this work, we examine fair classification through the lens of mismatched hypothesis testing: trying to find a classifier that distinguishes between two ideal distributions when given two mismatched distributions that are biased. Using Chernoff information, a tool in information theory, we theoretically demonstrate that, contrary to popular belief, there always exist ideal distributions such that optimal fairness and accuracy (with respect to the ideal distributions) are achieved simultaneously: there is no trade-off. Moreover, the same classifier yields the lack of a trade-off with respect to ideal distributions while yielding a trade-off when accuracy is measured with respect to the given (possibly biased) dataset. To complement our main result, we formulate an optimization to find ideal distributions and derive fundamental limits to explain why a trade-off exists on the given biased dataset. We also derive conditions under which active data collection can alleviate the fairness-accuracy trade-off in the real world. Our results lead us to contend that it is problematic to measure accuracy with respect to data that reflects bias, and instead, we should be considering accuracy with respect to ideal, unbiased data.

Improving maternal, newborn, and child health (MNCH) outcomes is a critical target for global sustainable development. Our research is centered on building predictive models, evaluating their interpretability, and generating actionable insights about the markers (features) and triggers (events) associated with vulnerability in MNCH. In this work, we demonstrate how a tool for inspecting "black box" machine learning models can be used to generate actionable insights from models trained on demographic health survey data to predict neonatal mortality.

In recent years, the rapid advances in Artificial Intelligence (AI) techniques along with an ever-increasing availability of healthcare data have made many novel analyses possible. Significant successes have been observed in a wide range of tasks such as next diagnosis prediction, AKI prediction, adverse event predictions including mortality and unexpected hospital re-admissions. However, there has been limited adoption and use in the clinical practice of these methods due to their black-box nature. A significant amount of research is currently focused on making such methods more interpretable or to make post-hoc explanations more accessible. However, most of this work is done at a very low level and as a result, may not have a direct impact at the point-of-care. This tutorial will provide an overview of the landscape of different approaches that have been developed for explainability in healthcare. Specifically, we will present the problem of explainability as it pertains to various personas involved in healthcare viz. data scientists, clinical researchers, and clinicians. We will chart out the requirements for such personas and present an overview of the different approaches that can address such needs. We will also walk-through several use-cases for such approaches. In this process, we will provide a brief introduction to explainability, charting its different dimensions as well as covering some relevant interpretability methods spanning such dimensions. We will touch upon some practical guides for explainability and provide a brief survey of open source tools such as the IBM AI Explainability 360 Open Source Toolkit.

As artificial intelligence algorithms make further inroads in high-stakes societal applications, there are increasing calls from multiple stakeholders for these algorithms to explain their outputs. To make matters more challenging, different personas of consumers of explanations have different requirements for explanations. Toward addressing these needs, we introduce AI Explainability 360, an open-source Python toolkit featuring ten diverse and state-of-the-art explainability methods and two evaluation metrics. Equally important, we provide a taxonomy to help entities requiring explanations to navigate the space of interpretation and explanation methods, not only those in the toolkit but also in the broader literature on explainability. For data scientists and other users of the toolkit, we have implemented an extensible software architecture that organizes methods according to their place in the AI modeling pipeline. The toolkit is not only the software, but also guidance material, tutorials, and an interactive web demo to introduce AI explainability to different audiences. Together, our toolkit and taxonomy can help identify gaps where more explainability methods are needed and provide a platform to incorporate them as they are developed.

Risk assessment is a growing use for machine learning models. When used in high-stakes applications, especially ones regulated by anti-discrimination laws or governed by societal norms for fairness, it is important to ensure that learned models do not propagate and scale any biases that may exist in training data. In this paper, we add on an additional challenge beyond fairness: unsupervised domain adaptation to covariate shift between a source and target distribution. Motivated by the real-world problem of risk assessment in new markets for health insurance in the United States and mobile money-based loans in East Africa, we provide a precise formulation of the machine learning with covariate shift and score parity problem. Our formulation focuses on situations in which protected attributes are not available in either the source or target domain. We propose two new weighting methods: prevalence-constrained covariate shift (PCCS) which does not require protected attributes in the target domain and target-fair covariate shift (TFCS) which does not require protected attributes in the source domain. We empirically demonstrate their efficacy in two applications.

Artificial intelligence systems are being increasingly deployed due to their potential to increase the efficiency, scale, consistency, fairness, and accuracy of decisions. However, as many of these systems are opaque in their operation, there is a growing demand for such systems to provide explanations for their decisions. Conventional approaches to this problem attempt to expose or discover the inner workings of a machine learning model with the hope that the resulting explanations will be meaningful to the consumer. In contrast, this paper suggests a new approach to this problem. It introduces a simple, practical framework, called Teaching Explanations for Decisions (TED), that provides meaningful explanations that match the mental model of the consumer. We illustrate the generality and effectiveness of this approach with two different examples, resulting in highly accurate explanations with no loss of prediction accuracy for these two examples.

Whereas previous post-processing approaches for increasing the fairness of predictions of biased classifiers address only group fairness, we propose a method for increasing both individual and group fairness. Our novel framework includes an individual bias detector used to prioritize data samples in a bias mitigation algorithm aiming to improve the group fairness measure of disparate impact. We show superior performance to previous work in the combination of classification accuracy, individual fairness and group fairness on several real-world datasets in applications such as credit, employment, and criminal justice.

In previous work, we proposed a scalable multi-party verification scheme for expensive iterative computations on a Blockchain substrate by appropriate storage and endorsement of frames of iterates. In this work, we extend the framework to verify sets of complete computations with different unordered hyperparameters and develop frame ordering and compression algorithms to enable scalability in the system. We illustrate the efficacy of the proposed approach by verifying the OpenMalaria epidemiological simulation.

We propose the labeled Cech complex, the plain labeled Vietoris-Rips complex, and the locally scaled labeled Vietoris-Rips complex to perform persistent homology inference of decision boundaries in classification tasks. We provide theoretical conditions and analysis for recovering the homology of a decision boundary from samples. Our main objective is quantification of deep neural network complexity to enable matching of datasets to pre-trained models to facilitate the functioning of AI marketplaces; we report results for experiments using MNIST, FashionMNIST, and CIFAR10.

Autonomous cyber-physical agents play an increasingly large role in our lives. To ensure that they behave in ways aligned with the values of society, we must develop techniques that allow these agents to not only maximize their reward in an environment, but also to learn and follow the implicit constraints of society. We detail a novel approach that uses inverse reinforcement learning to learn a set of unspecified constraints from demonstrations and reinforcement learning to learn to maximize environmental rewards. A contextual bandit-based orchestrator then picks between the two policies: constraint-based and environment reward-based. The contextual bandit orchestrator allows the agent to mix policies in novel ways, taking the best actions from either a reward-maximizing or constrained policy. In addition, the orchestrator is transparent on which policy is being employed at each time step. We test our algorithms using Pac-Man and show that the agent is able to learn to act optimally, act within the demonstrated constraints, and mix these two functions in complex ways.

In recent years there has been an increasing trend in which data scientists and domain experts work together to tackle complex scientific questions. However, such collaborations often face challenges. In this paper, we aim to decipher this collaboration complexity through a semi-structured interview study with 22 interviewees from teams of bio-medical scientists collaborating with data scientists. In the analysis, we adopt the Olsons' four-dimensions framework proposed in Distance Matters to code interview transcripts. Our findings suggest that besides the glitches in the collaboration readiness, technology readiness, and coupling of work dimensions, the tensions that exist in the common ground building process influence the collaboration outcomes, and then persist in the actual collaboration process. In contrast to prior works' general account of building a high level of common ground, the breakdowns of content common ground together with the strengthen of process common ground in this process is more beneficial for scientific discovery. We discuss why that is and what the design suggestions are, and conclude the paper with future directions and limitations.

The United Nations Development Programme (UNDP) helps countries implement the United Nations (UN) Sustainable Development Goals (SDGs), an agenda for tackling major societal issues such as poverty, hunger, and environmental degradation by the year 2030. A key service provided by UNDP to countries that seek it is a review of national development plans and sector strategies by policy experts to assess alignment of national targets with one or more of the 169 targets of the 17 SDGs. Known as the Rapid Integrated Assessment (RIA), this process involves manual review of hundreds, if not thousands, of pages of documents and takes weeks to complete. In this work, we develop a natural language processing-based methodology to accelerate the workflow of policy experts. Specifically we use paragraph embedding techniques to find paragraphs in the documents that match the semantic concepts of each of the SDG targets. One novel technical contribution of our work is in our use of historical RIAs from other countries as a form of neighborhood-based supervision for matches in the country under study. We have successfully piloted the algorithm to perform the RIA for Papua New Guinea’s national plan, with the UNDP estimating it will help reduce their completion time from an estimated 3-4 weeks to 3 days.

Your computer is continuously executing programs, but does it really understand them? Not in any meaningful sense. That burden falls upon human knowledge workers, who are increasingly asked to write and understand code. They would benefit greatly from intelligent tools that reveal the connections between their code and its subject matter. Towards this prospect, we present an AI system that forms semantic representations of computer programs, using techniques from knowledge representation and program analysis. These representations are created through a novel algorithm for the semantic enrichment of dataflow graphs. We illustrate its workings with examples from the field of data science. The algorithm is undergirded by a new ontology language for modeling computer programs and a new ontology about data science, written in this language.

Non-discrimination is a recognized objective in algorithmic decision making. In this paper, we introduce a novel probabilistic formulation of data pre-processing for reducing discrimination. We propose a convex optimization for learning a data transformation with three goals: controlling discrimination, limiting distortion in individual data samples, and preserving utility. We characterize the impact of limited sample size in accomplishing this objective. Two instances of the proposed optimization are applied to datasets, including one on real-world criminal recidivism. Results show that discrimination can be greatly reduced at a small cost in classification accuracy.

Recommendation for e-commerce with a mix of durable and nondurable goods has characteristics that distinguish it from the well-studied media recommendation problem. The demand for items is a combined effect of form utility and time utility, i.e., a product must both be intrinsically appealing to a consumer and the time must be right for purchase. In particular for durable goods, time utility is a function of inter-purchase duration within product category because consumers are unlikely to purchase two items in the same category in close temporal succession. Moreover, purchase data, in contrast to rating data, is implicit with non-purchases not necessarily indicating dislike. Together, these issues give rise to the positive-unlabeled demand-aware recommendation problem that we pose via joint low-rank tensor completion and product category inter-purchase duration vector estimation. We further relax this problem and propose a highly scalable alternating minimization approach with which we can solve problems with millions of users and millions of items in a single thread. We also show superior prediction accuracies on multiple real-world datasets.

Racial discrimination in decision-making scenarios such as police arrests appears to be a violation of expected utility theory. Drawing on results from the science of information, we discuss an information-based model of signal detection over a population that generates such behavior as an alternative explanation to taste-based discrimination by the decision maker or differences among the racial populations. This model uses the decision rule that maximizes expected utility-the likelihood ratio test-but constrains the precision of the threshold to a small discrete set. The precision constraint follows from both bounded rationality in human recollection and finite training data for estimating priors. When combined with social aspects of human decision making and precautionary cost settings, the model predicts the own-race bias that has been observed in several econometric studies.

Estimating the skills, talents, and expertise of employees is essential for human capital management in knowledge-based organizations across industries and sectors. In this paper, we describe an approach to infer the expertise of employees from their enterprise data and digital footprints. Using a novel big data workflow with components of information retrieval and search, data fusion, matrix completion, and ordinal regression clustering, we are able to automatically find evidence of expertise and determine appropriate evidence weights for different queries and data sources that we merge and present in a manner consumable by businesspeople. We illustrate the system on sample data from the IBM Corporation where it has been deployed.

In our previous work we have presented an approach to learn interpretable classification rules using a Boolean compressed sensing formulation. Our approach uses a linear programming (LP) relaxation and allows us to find interpretable (sparse) classification rules that achieve good generalization accuracy. However, the resulting LP representation for problems with either a large number of samples or large number of continuous features tends to become challenging for off-the-shelf LP solvers. We have explored a screening approach which allows us to dramatically reduce the number of active features without sacrificing optimality. In this work we explore reducing the number of samples in a sequential setting where we can certify reaching a near-optimal solution while only solving the LP on a small fraction of the available data points. In a batch setting this approach can dramatically reduce the computational complexity of the rule-learning LP formulation. In an online setting we derive stochastic upper and lower bounds on the the LP objective for unseen samples. This allows early stopping when we detect that the classifier will not change significantly with additional samples. The upper bounds are related to the learning curve literature in machine learning, and our lower bounds appear not to have been explored. Finally, we discuss a quick approach to compute the complete regularization path balancing rule interpretability versus accuracy.

Topological signal processing, especially persistent homology, is a growing field of study for analyzing sets of data points that has been heretofore applied to unlabeled data. In this work, we consider the case of labeled data and examine the topology of the decision boundary separating different labeled classes. Specifically, we propose a novel approach to construct simplicial complexes of decision boundaries, which can be used to understand their topology. Furthermore, we illustrate one use case for this line of theoretical work in kernel selection for supervised classification problems.

In hypothesis testing, the phenomenon of label noise, in which hypothesis labels are switched at random, contaminates the likelihood functions. In this paper, we develop a new method to determine the decision rule when we do not have knowledge of the uncontaminated likelihoods and contamination probabilities, but only have knowledge of the contaminated likelihoods. In particular we pose a minimax optimization problem that finds a decision rule robust against this lack of knowledge. The method simplifies by application of linear programming theory. Motivation for this investigation is provided by problems encountered in workforce analytics.

Some of the largest worldwide employers today are knowledge-based enterprises whose most important asset is human capital. Knowledge workers are unique, each having individualized skills, competencies and expertise, which constantly evolve and expand. Managing and planning for such a workforce critically depends on the ability to construct complete, accurate, and real-time representation and inventory of the expertise of employees in a form that integrates with business processes. In this session Saška will describe how enterprise expertise assessment process can be posed as predictive modeling and recommendation problem, and will present results and lessons learned from an actual deployment of IBM Expertise, a corporate-wide expertise recommendation and management system.

Health insurance companies prefer to enter new markets in which individuals likely to enroll in their plans have a low annual cost. When deciding which new markets to enter, health cost data for the new markets is unavailable to them, but health cost data for their own enrolled members is available. To address the problem of assessing risk in new markets, i.e., estimating the cost of likely enrollees, we pose a regression problem with demographic data as predictors combined with a novel three-population covariate shift. Since this application deals with health data that is protected by privacy laws, we cannot use the raw data of the insurance company's members directly for training the regression and covariate shift. Therefore, to construct a full solution, we also develop a novel method to achieve k-anonymity with the workload-driven quality of data distribution preservation achieved through dithered quantization and Rosenblatt's transformation. We illustrate the efficacy of the solution using real-world, publicly available data.

Convex relaxations for sparse representation problems, which aim to find sparse solutions to systems of equations, have enabled a variety of exciting applications in high-dimensional settings. Yet, with dimensions large enough, even these convex formulations become prohibitively expensive. Screening methods attempt to use duality theory to dramatically reduce the size of the optimization problem through easily computable certificates that many of the variables must be zero in the optimal solution. In this paper we consider learning sparse classification rules via Boolean compressed sensing and develop screening procedures that can significantly reduce the size of the resulting linear program. Boolean compressed sensing deals with systems of Boolean equations (instead of linear equations in traditional compressed sensing); we develop screening methods specifically for this setting. We demonstrate the effectiveness of our screening rules on several real-world classification data sets.

Unconditional cash transfers to the extreme poor via mobile telephony represent a radical, new approach to giving. GiveDirectly is a non-governmental organization (NGO) at the vanguard of delivering this proven and effective approach to reducing poverty. In this work, we streamline an important step in the operations of the NGO by developing and deploying a data-driven system for locating villages with extreme poverty in Kenya and Uganda. Using the type of roof of a home, thatched or metal, as a proxy for poverty, we develop a new remote sensing approach for selecting extremely poor villages to target for cash transfers. We develop an analytics algorithm that estimates housing quality and density in patches of publicly-available satellite imagery by learning a predictive model with sieves of template matching results combined with color histograms as features. We develop and deploy a crowdsourcing interface to obtain labeled training data. We deploy the predictive model to construct a fine-scale heat map of poverty and integrate this discovered knowledge into the processes of GiveDirectly's operations. Aggregating estimates at the village level, we produce a ranked list from which top villages are included in GiveDirectly's planned distribution of cash transfers. The automated approach increases village selection efficiency significantly.

Strategic planning and talent management in large enterprises composed of knowledge workers requires complete, accurate, and up-to-date representation of the expertise of employees in a form that integrates with business processes. Like other similar organizations operating in dynamic environments, the IBM Corporation strives to maintain such current and correct information, specifically assessments of employees against job roles and skill sets from its expertise taxonomy. In this work, we deploy an analytics-driven solution that infers the expertise of employees through the mining of enterprise and social data that is not specifically generated and collected for expertise inference. We consider job role and specialty prediction and pose them as supervised classification problems. We evaluate a large number of feature sets, predictive models and postprocessing algorithms, and choose a combination for deployment. This expertise analytics system has been deployed for key employee population segments, yielding large reductions in manual effort and the ability to continually and consistently serve up-to-date and accurate data for several business functions. This expertise management system is in the process of being deployed throughout the corporation.

Bayesian hypothesis testing and minimax hypothesis testing represent extreme instances of detection in which the prior probabilities of the hypotheses are either completely and precisely known, or are completely unknown. Group minimax, also known as Gamma -minimax, is a robust intermediary between Bayesian and minimax hypothesis testing that allows for coarse or partial advance knowledge of the hypothesis priors by using information on sets in which the prior lies. Existing work on group minimax, however, does not consider the question of how to define the sets or groups of priors; it is assumed that the groups are given. In this paper, we propose a novel intermediate detection scheme formulated through the quantization of the space of prior probabilities that optimally determines groups and also representative priors within the groups. We show that when viewed from a quantization perspective, group minimax amounts to determining centroids with a minimax Bayes risk error divergence distortion criterion: the appropriate Bregman divergence for this task. In addition, the optimal partitioning of the space of prior probabilities is a Bregman Voronoi diagram. Together, the optimal grouping and representation points are an epsilon -net with respect to Bayes risk error divergence, and permit a rate-distortion type asymptotic analysis of detection performance with the number of groups. Examples of detecting signals corrupted by additive white Gaussian noise and of distinguishing exponentially-distributed signals are presented.

Bounded confidence opinion dynamic models have received much recent interest as models of information propagation in social networks and localized distributed averaging. However in the existing literature, opinions are only viewed as abstract quantities rather than as part of a decision-making system. In this work, opinion dynamics are examined when agents are Bayesian decision makers that perform hypothesis testing or signal detection. Bounded confidence is defined on prior probabilities of hypotheses through Bayes risk error divergence, the appropriate measure between priors in hypothesis testing. This definition contrasts with the measure used between opinions in the standard model: absolute error. It is shown that the rapid convergence of prior probabilities to a small number of limiting values is similar to that seen in the standard model. The most interesting finding in this work is that the number of these limiting values changes with the signal-to-noise ratio in the hypothesis testing task. The number of final values or clusters is maximal at intermediate signal-to-noise ratios, suggesting that the most contentious issues lead to the largest number of factions.

We propose an interpretable rule-based classification system based on ideas from Boolean compressed sensing. We represent the problem of learning individual conjunctive clauses or individual disjunctive clauses as a Boolean group testing problem, and apply a novel linear programming relaxation to find solutions. We derive results for exact rule recovery which parallel the conditions for exact recovery of sparse signals in the compressed sensing literature: although the general rule recovery problem is NP-hard, under some conditions on the Boolean ‘sensing’ matrix, the rule can be recovered exactly. This is an exciting development in rule learning where most prior work focused on heuristic solutions. Furthermore we construct rule sets from these learned clauses using set covering and boosting. We show competitive classification accuracy using the proposed approach.

Wireless sensor networks are composed of distributed sensors that can be used for signal detection or classification. The likelihood functions of the hypotheses are often not known in advance, and decision rules have to be learned via supervised learning. A specific learning algorithm is Fisher discriminant analysis (FDA), the classification accuracy of which has been previously studied in the context of wireless sensor networks. Previous work, however, does not take into account the communication protocol or battery lifetime; in this paper we extend existing studies by proposing a model that captures the relationship between battery lifetime and classification accuracy. To do so, we combine the FDA with a model that captures the dynamics of the carrier-sense multiple-access (CSMA) algorithm, the random-access algorithm used to regulate communications in sensor networks. This allows us to study the interaction between the classification accuracy, battery lifetime and effort put towards learning, as well as the impact of the back-off rates of CSMA on the accuracy. We characterize the tradeoff between the length of the training stage and accuracy, and show that accuracy is non-monotone in the back-off rate due to changes in the training sample size and overfitting.

Classification algorithms used to support the decisions of human analysts are often used in settings in which zero-one loss is not the appropriate indication of performance. The zero-one loss corresponds to the operating point with equal costs for false alarms and missed detections, and no option for the classifier to leave uncertain test samples unlabeled. A generalization bound for ensemble classification at the standard operating point has been developed based on two interpretable properties of the ensemble: strength and correlation, using the Chebyshev inequality. Such generalization bounds for other operating points have not been developed previously and are developed in this paper. Significantly, the bounds are empirically shown to have much practical utility in determining optimal parameters for classification with a reject option, classification for ultralow probability of false alarm, and classification for ultralow probability of missed detection. Counter to the usual guideline of large strength and small correlation in the ensemble, different guidelines are recommended by the derived bounds in the ultralow false alarm and missed detection probability regimes.

Matrix factorization from a small number of observed entries has recently garnered much attention as the key ingredient of successful recommendation systems. One unresolved problem in this area is how to adapt current methods to handle changing user preferences over time. Recent proposals to address this issue are heuristic in nature and do not fully exploit the time-dependent structure of the problem. As a principled and general temporal formulation, we propose a dynamical state space model of matrix factorization. Our proposal builds upon probabilistic matrix factorization, a Bayesian model with Gaussian priors. We utilize results in state tracking, i.e. the Kalman filter, to provide accurate recommendations in the presence of both process and measurement noise. We show how system parameters can be learned via expectation-maximization and provide comparisons to current published techniques.

In this article, we propose a random walk-based model to predict legislators' votes on a set of bills. In particular, we first convert roll call data, i.e. the recorded votes and the corresponding deliberative bodies, to a heterogeneous graph, where both the legislators and bills are treated as vertices. Three types of weighted edges are then computed accordingly, representing legislators' social and political relations, bills' semantic similarity, and legislator-bill vote relations. Through performing two-stage random walks over this heterogeneous graph, we can estimate legislative votes on past and future bills. We apply this proposed method on real legislative roll call data of the United States Congress and compare to state-of-the-art approaches. The experimental results demonstrate the superior performance and unique prediction power of the proposed model.

Signals in response to time-localized events of a common phenomenon tend to exhibit a common shape, but with variable time scale, amplitude, and delay across trials in many domains. We develop a new formulation to learn the common shape and variables from noisy signal samples with a Bayesian signal model and a Markov chain Monte Carlo inference scheme involving Gibbs sampling and independent Metropolis-Hastings. Our experiments with generated and real-world data show that the algorithm is robust to missing data, outperforms the existing approaches and produces easily interpretable outputs.

A study of generalization error in signal detection by multiple spatially-distributed and -correlated sensors is provided when the detection rule is learned from a finite number of training samples via the classical linear discriminant analysis formulation. Spatial correlation among sensors is modeled by a Gauss-Markov random field defined on a nearest neighbor graph according to inter-sensor spatial distance, where sensors are placed randomly on a growing bounded region of the plane. A fairly simple approximate expression for generalization error is derived involving few parameters. It is shown that generalization error is minimized not when there are an infinite number of sensors, but a number of sensors equal to half the number of samples in the training set. The minimum generalization error is related to a single parameter of the sensor spatial location distribution, derived based on weak laws of large numbers in geometric probability. The finite number of training samples acts like a budgeting variable, similar to a total communication power constraint.

The generalization error, or probability of misclassification, of ensemble classifiers has been shown to be bounded above by a function of the mean correlation between the constituent (i.e., base) classifiers and their average strength. This bound suggests that increasing the strength and/or decreasing the correlation of an ensemble's base classifiers may yield improved performance under the assumption of equal error costs. However, this and other existing bounds do not directly address application spaces in which error costs are inherently unequal. For applications involving binary classification, Receiver Operating Characteristic (ROC) curves, performance curves that explicitly trade off false alarms and missed detections, are often utilized to support decision making. To address performance optimization in this context, we have developed a lower bound for the entire ROC curve that can be expressed in terms of the class-specific strength and correlation of the base classifiers.

Arthroplasty, the implantation of prostheses into joints, is a surgical procedure that is affecting a larger and larger number of patients over time. As a result, it is increasingly important to develop imaging techniques to noninvasively examine joints with prostheses after surgery, both statically and dynamically in 3-D. The static problem is considered here, with the aim to create a 3-D shape model of the bone as well as the prosthesis using a set of 2-D X-rays from various viewpoints. The most important challenge to be addressed is the lack of texture, the most common feature to recover shape from multiple views. In order to overcome this limitation, we reformulate the problem using a novel multiview segmentation approach where an active contours 3-D surface evolution with level-set implementation is used to recover the shape of bones and prostheses in postoperative joints. The recovered shape may then be used to track 3-D motions in dynamic X-ray sequences to obtain kinematic information.

Bayesian hypothesis testing is investigated when the prior probabilities of the hypotheses, taken as a random vector, must be quantized. Nearest neighbor and centroid conditions for quantizer optimality are derived using mean Bayes risk error as a distortion measure. An example of optimal quantization for hypothesis testing is provided. Human decision making is briefly studied assuming quantized prior Bayesian hypothesis testing; this model explains several experimental findings.