Explainable reinforcement learning (XRL) methods aim to help elucidate agent policies and decision-making processes. The majority of XRL approaches focus on local explanations, seeking to shed light on the reasons an agent acts the way it does at a specific world state. While such explanations are both useful and necessary, they typically do not portray the outcomes of the agent's selected choice of action. In this work, we propose ``COViz'', a new local explanation method that visually compares the outcome of an agent's chosen action to a counterfactual one. In contrast to most local explanations that provide state-limited observations of the agent's motivation, our method depicts alternative trajectories the agent could have taken from the given state and their outcomes. We evaluated the usefulness of COViz in supporting people's understanding of agents' preferences and compare it with reward decomposition, a local explanation method that describes an agent's expected utility for different actions by decomposing it into meaningful reward types. Furthermore, we examine the complementary benefits of integrating both methods. Our results show that such integration significantly improved participants' performance.

Truth discovery is a general name for a broad range of statistical methods aimed to extract the correct answers to questions, based on multiple answers coming from noisy sources. For example, workers in a crowdsourcing platform. In this paper, we consider an extremely simple heuristic for estimating workers' competence using average proximity to other workers. We prove that this estimates well the actual competence level and enables separating high and low quality workers in a wide spectrum of domains and statistical models. Under Gaussian noise, this simple estimate is the unique solution to the MLE with a constant regularization factor. Finally, weighing workers according to their average proximity in a crowdsourcing setting, results in substantial improvement over unweighted aggregation and other truth discovery algorithms in practice.

With Artificial Intelligence on the rise, human interaction with autonomous agents becomes more frequent. Effective human-agent collaboration requires users to understand the agent's behavior, as failing to do so may cause reduced productivity, misuse or frustration. Agent strategy summarization methods are used to describe the strategy of an agent to users through demonstrations. A summary's objective is to maximize the user's understanding of the agent's aptitude by showcasing its behaviour in a selected set of world states. While shown to be useful, we show that current methods are limited when tasked with comparing between agents, as each summary is independently generated for a specific agent. In this paper, we propose a novel method for generating dependent and contrastive summaries that emphasize the differences between agent policies by identifying states in which the agents disagree on the best course of action. We conducted user studies to assess the usefulness of disagreement-based summaries for identifying superior agents and conveying agent differences. Results show disagreement-based summaries lead to improved user performance compared to summaries generated using HIGHLIGHTS, a strategy summarization algorithm which generates summaries for each agent independently.

With advances in reinforcement learning (RL), agents are now being developed in high-stakes application domains such as healthcare and transportation. Explaining the behavior of these agents is challenging, as they act in large state spaces, and their decision-making can be affected by delayed rewards. In this paper, we explore a combination of explanations that attempt to convey the global behavior of the agent and local explanations which provide information regarding the agent's decision-making in a particular state. Specifically, we augment strategy summaries that demonstrate the agent's actions in a range of states with saliency maps highlighting the information it attends to. Our user study shows that intelligently choosing what states to include in the summary (global information) results in an improved analysis of the agents. We find mixed results with respect to augmenting summaries with saliency maps (local information).

Images are powerful. Visual information can attract attention, improve persuasion, trigger stronger emotions, and is easy to share and spread. We examine the characteristics of the popular images shared on Twitter as part of "Stop the Steal'', the widespread misinformation campaign during the 2020 U.S. election. We analyze the spread of the forty most popular images shared on Twitter as part of this campaign. Using a coding process, we categorize and label the images according to their type, content, origin, and role, and perform a mixed-method analysis of these images' spread on Twitter. Our results show that popular images include both photographs and text rendered as image. Only very few of these popular images included alleged photographic evidence of fraud; and none of the popular photographs had been manipulated. Most images reached a significant portion of their total spread within several hours from their first appearance, and both popular- and less-popular accounts were involved in various stages of their spread.

People have limited computational resources, yet they make complex strategic decisions over enormous spaces of possibilities. How do people efficiently search spaces with combinatorially branching paths? Here, we study players' search strategies for a winning move in a "k-in-a-row" game. We find that players use scoring strategies to prune the search space and augment this pruning by a "shutter" heuristic that focuses the search on the paths emanating from their previous move. This strong pruning has its costs-both computational simulations and behavioral data indicate that the shutter size is correlated with players' blindness to their opponent's winning moves. However, simulations of the search while varying the shutter size, complexity levels, noise levels, branching factor, and computational limitations indicate that despite its costs, a narrow shutter strategy is the dominant strategy for most of the parameter space. Finally, we show that in the presence of computational limitations, the shutter heuristic enhances the performance of deep learning networks in these end-game scenarios. Together, our findings suggest a novel adaptive heuristic that benefits search in a vast space of possibilities of a strategic game.

People increasingly rely on Artificial Intelligence (AI) based systems to aid decision-making in various domains and often face a choice between alternative systems. We explored the effects of users' perception of AI systems' warmth (perceived intent) and competence (perceived ability) on their choices. In a series of studies, we manipulated AI systems' warmth and competence levels. We show that, similar to the judgments of other people, there is often primacy for warmth over competence. Specifically, when faced with a choice between a high-competence system and a high-warmth system, more participants preferred the high-warmth system. Moreover, the precedence of warmth persisted even when the high-warmth system was overtly deficient in its competence compared to an alternative high competence-low warmth system. The current research proposes that it may be vital for AI systems designers to consider and communicate the system's warmth characteristics to its potential users.

Matching is a task at the heart of any data integration process, aimed at identifying correspondences among data elements. Matching problems were traditionally solved in a semi-automatic manner, with correspondences being generated by matching algorithms and outcomes subsequently validated by human experts. Human-in-the-loop data integration has been recently challenged by the introduction of big data and recent studies have analyzed obstacles to effective human matching and validation. In this work we characterize human matching experts, those humans whose proposed correspondences can mostly be trusted to be valid. We provide a novel framework for characterizing matching experts that, accompanied with a novel set of features, can be used to identify reliable and valuable human experts. We demonstrate the usefulness of our approach using an extensive empirical evaluation. In particular, we show that our approach can improve matching results by filtering out inexpert matchers.

With advances in reinforcement learning (RL), agents are now being developed in high-stakes application domains such as healthcare and transportation. Explaining the behavior of these agents is challenging, as the environments in which they act have large state spaces, and their decision-making can be affected by delayed rewards, making it difficult to analyze their behavior. To address this problem, several approaches have been developed. Some approaches attempt to convey the global behavior of the agent, describing the actions it takes in different states. Other approaches devised local explanations which provide information regarding the agent's decision-making in a particular state. In this paper, we combine global and local explanation methods, and evaluate their joint and separate contributions, providing (to the best of our knowledge) the first user study of combined local and global explanations for RL agents. Specifically, we augment strategy summaries that extract important trajectories of states from simulations of the agent with saliency maps which show what information the agent attends to. Our results show that the choice of what states to include in the summary (global information) strongly affects people's understanding of agents: participants shown summaries that included important states significantly outperformed participants who were presented with agent behavior in a set of world-states that are likely to appear during gameplay. We find mixed results with respect to augmenting demonstrations with saliency maps (local information), as the addition of saliency maps, in the form of raw heat maps, did not significantly improve performance in most cases. However, we do find some evidence that saliency maps can help users better understand what information the agent relies on during its decision-making, suggesting avenues for future work that can further improve explanations of RL agents.

AI agents support high stakes decision-making processes from driving cars to prescribing drugs, making it increasingly important for human users to understand their behavior. Policy summarization methods aim to convey strengths and weaknesses of such agents by demonstrating their behavior in a subset of informative states. Some policy summarization methods extract a summary that optimizes the ability to reconstruct the agent's policy under the assumption that users will deploy inverse reinforcement learning. In this paper, we explore the use of different models for extracting summaries. We introduce an imitation learning-based approach to policy summarization; we demonstrate through computational simulations that a mismatch between the model used to extract a summary and the model used to reconstruct the policy results in worse reconstruction quality; and we demonstrate through a human-subject study that people use different models to reconstruct policies in different contexts, and that matching the summary extraction model to these can improve performance. Together, our results suggest that it is important to carefully consider user models in policy summarization.

People collaborate in carrying out such complex activities as treating patients, co-authoring documents and developing software. While technologies such as Dropbox and Github enable groups to work in a distributed manner, coordinating team members' individual activities poses significant challenges. In this paper, we formalize the problem of "information sharing in loosely-coupled extended-duration teamwork." We develop a new representation, Mutual Influence Potential Networks (MIP-Nets), to model collaboration patterns and dependencies among activities, and an algorithm, MIP-DOI, that uses this representation to reason about information sharing.

This paper proposes a mapping between multi-agent pathfinding (MAPF) and combinatorial auctions (CAs). In MAPF, agents need to reach their goal destinations without colliding. Algorithms for solving MAPF aim at assigning agents non-conflicting paths that minimize agents' travel costs. In CA problems, agents bid over bundles of items they desire. Auction mechanisms aim at finding an allocation of bundles that maximizes social welfare. In the proposed mapping of MAPF to CAs, agents bid on paths to their goals and the auction allocates non-colliding paths to the agents. Using this formulation, auction mechanisms can be naturally used to solve a range of MAPF problem variants. In particular, auction mechanisms can be applied to non-cooperative settings with self-interested agents while providing optimality guarantees and robustness to manipulations by agents. The paper further shows how to efficiently implement an auction mechanism for MAPF, utilizing methods and representations from both the MAPF and CA literatures.

Children with complex health conditions require care from a large, diverse team of caregivers that includes multiple types of medical professionals, parents and community support organizations. Coordination of their outpatient care, essential for good outcomes, presents major challenges. Extensive healthcare research has shown that the use of integrated, team-based care plans improves care coordination, but such plans are rarely deployed in practice. This paper reports on a study of care teams treating children with complex conditions at a major university tertiary care center. This study investigated barriers to plan implementation and resultant care coordination problems. It revealed the complex nature of teamwork in complex care, which poses challenges to team coordination that extend beyond those identified in prior work and handled by existing coordination systems. The paper builds on a computational teamwork theory to identify opportunities for technology to support increased plan-based complex-care coordination and to propose design approaches for systems that enable and enhance such coordination.

Teamwork and care coordination are of increasing importance to health care delivery and patient safety and health. My research aims at developing agents that are able to make intelligent information sharing decisions to support a diverse, evolving team of care providers in constructing and maintaining a shared plan that operates in uncertain environments.

This paper defines the "Single Agent in a Team Decision" (SATD) problem. SATD differs from prior multi-agent communication problems in the assumptions it makes about teammates' knowledge of each other's plans and possible observations. The paper proposes a novel integrated logical-decision-theoretic approach to solving SATD problems, called MDP-PRT. Evaluation of MDP-PRT shows that it outperforms a previously proposed communication mechanism that did not consider the timing of communication and compares favorably with a coordinated Dec-POMDP solution that uses knowledge about all possible observations.

An ability to predict the mileage at failure of components in a complicated system, particularly in automobiles, is a challenging task. In the current work, a methodology for estimating the distribution of failure and survival rate of automobile components affected by multiple factors is presented. A novel adaptation of an ensemble recursive partitioning and tree-based learning method, mutually exclusive forest, is introduced. The proposed method is capable of handling a high dimensional dataset and maximizing the extracted information to estimate the distribution of mileage at failure of automobile components. Each tree in the proposed mutually exclusive forest uses a mutually exclusive set of factors in each of its constituent decision trees to classify the failure data. Information across the trees is combined to obtain the failure rate distribution of an automobile component with respect to mileage. A case study, based on real-world field failure data and censored data of automobile components, is presented to evaluate the proposed algorithm. Results show similar results to the C-Forest approach in terms of prediction quality, while generating models with significantly lower space that are easier to interpret.

A popular use of crowdsourcing is to collect and aggregate individual worker responses to problems to reach a correct answer. This paper studies the relationship between the computation complexity class of problems, and the ability of a group to agree on a correct solution. We hypothesized that for NP-Complete (NPC) problems, groups would be able to reach a majority-based correct solution once it was suggested by a group member and presented to the other members, due to the "easy to verify" (i.e., verification in polynomial time) characteristic of this complexity class. In contrast, when posed with PSPACE-Complete (PSC) "hard to verify" problems (i.e., verification in exponential time), groups will not necessarily be able to choose a correct solution even if such a solution has been presented. Consequently, increasing the size of the group is expected to facilitate the ability of the group to converge on a correct solution when solving NPC problems, but not when solving PSC problems. To test this hypothesis we conducted preliminary experiments in which we evaluated people's ability to solve an analytical problem and their ability to recognize a correct solution. In our experiments, participants were significantly more likely to recognize correct and incorrect solutions for NPC problems than for PSC problems, even for problems of similar difficulties (as measured by the percentage of participants who solved the problem). This is a first step towards formalizing a relationship between the computationally complexity of a problem and the crowd's ability to converge to a correct solution to the problem.

Modern pedagogical software is open-ended and flexible, allowing students to solve problems through exploration and trial-and-error. Such exploratory settings provide for a rich educational environment for students, but they challenge teachers to keep track of students’ progress and to assess their performance. This article presents techniques for recognizing students’ activities in such pedagogical software and visualizing these activities to teachers. It describes a new plan recognition algorithm that uses a recursive grammar that takes into account repetition and interleaving of activities. This algorithm was evaluated empirically using an exploratory environment for teaching chemistry used by thousands of students in several countries. It was always able to correctly infer students’ plans when the appropriate grammar was available. We designed two methods for visualizing students’ activities for teachers: one that visualizes students’ inferred plans, and one that visualizes students’ interactions over a timeline. Both of these visualization methods were preferred to and found more helpful than a baseline method which showed a movie of students’ interactions. These results demonstrate the benefit of combining novel AI techniques and visualization methods for the purpose of designing collaborative systems that support students in their problem solving and teachers in their understanding of students’ performance.