Behavioral scientists have classically documented aversion to algorithmic decision aids, from simple linear models to AI. Sentiment, however, is changing and possibly accelerating AI helper usage. AI assistance is, arguably, most valuable when humans must make complex choices. We argue that classic experimental methods used to study heuristics and biases are insufficient for studying complex choices made with AI helpers. We adapted an experimental paradigm designed for studying complex choices in such contexts. We show that framing and anchoring effects impact how people work with an AI helper and are predictive of choice outcomes. The evidence suggests that some participants, particularly those in a loss frame, put too much faith in the AI helper and experienced worse choice outcomes by doing so. The paradigm also generates computational modeling-friendly data allowing future studies of human-AI decision making.

Theory of Mind (ToM) is a psychological construct that captures the ability to ascribe mental states to others and then use those representations for explaining and predicting behavior. We review recent progress in endowing artificially intelligent robots with ToM. A broad array of modeling, experimental, and benchmarking approaches and methods are present in the extant literature. Unlike other domains of human cognition for which research has achieved super-human capabilities, ToM for robots lacks a unified construct and is not consistently benchmarked or validated—realities which possibly hinder progress in this domain. We argue that this is, at least in part, due to inconsistent defining of ToM, no presence of a unifying modeling construct, and the absence of a shared data resource. We believe these would improve the ability of the research community to compare the ToM abilities of different systems. We suggest that establishing a shared definition of ToM, creating a shared data resource that supports consistent benchmarking & validation, and developing a generalized modeling tool are critical steps towards giving robots ToM capabilities that lay observers will recognize as such.

Understanding the decisions of AI-driven systems and the rationale behind such decisions is key to the success of the human-robot team. However, the complexity and the "black-box" nature of many AI algorithms create a barrier for establishing such understanding within their human counterparts. Reinforcement Learning (RL), a machine-learning algorithm based on the simple idea of action-reward mappings, has a rich quantitative representation and a complex iterative reasoning process that present a significant obstacle to human understanding of, for example, how value functions are constructed, how the algorithms update the value functions, and how such updates impact the action/policy chosen by the robot. In this paper, we discuss our work to address this challenge by developing a decision-tree based explainable model for RL to make a robot’s decision-making process more transparent. Set in a human-robot virtual teaming testbed, we conducted a study to assess the impact of the explanations, generated using decision trees, on building transparency, calibrating trust, and improving the overall human-robot team’s performance. We discuss the design of the explainable model and the positive impact of the explanations on outcome measures.

Trust calibration is critical to the success of human-agent interaction (HAI). However, individual differences are ubiquitous in people's trust relationships with autonomous systems. To assist its heterogeneous human teammates calibrate their trust in it, an agent must first dynamically model them as individuals, rather than communicating with them all in the same manner. It can then generate expectations of its teammates' behavior and optimize its own communication based on the current state of the trust relationship it has with them. In this work, we examine how an agent can generate accurate expectations given observations of only the teammate's trust-related behaviors (e.g., did the person follow or ignore its advice?). In addition to this limited input, we also seek a specific output: accurately predicting its human teammate's future trust behavior (e.g., will the person follow or ignore my next suggestion?). In this investigation, we construct a model capable of generating such expectations using data gathered in a human-subject study of behavior in a simulated human-robot interaction (HRI) scenario. We first analyze the ability of measures from a pre-survey on trust-related traits to accurately predict subsequent trust behaviors. However, as the interaction progresses, this effect is dwarfed by the direct experience. We therefore analyze the ability of sequences of prior behavior by the teammate to accurately predict subsequent trust behaviors. Such behavioral sequences have shown to be indicative of the subjective beliefs of other teammates, and we show here that they have a predictive power as well.

Trust is a critical factor for achieving the full potential of human-robot teams. Researchers have theorized that people will more accurately trust an autonomous system, such as a robot, if they have a more accurate understanding of its decision-making process. Studies have shown that hand-crafted explanations can help maintain trust when the system is less than 100% reliable. In this work, we leverage existing agent algorithms to provide a domain-independent mechanism for robots to automatically generate such explanations. To measure the explanation mechanism's impact on trust, we collected self-reported survey data and behavioral data in an agent-based online testbed that simulates a human-robot team task. The results demonstrate that the added explanation capability led to improvement in transparency, trust, and team performance. Furthermore, by observing the different outcomes due to variations in the robot's explanation content, we gain valuable insight that can help lead to refinement of explanation algorithms to further improve human-robot trust calibration.

The relative influence of perception and situation in emotional judgments has been extensively debated in psychology. A main issue in this debate concerns how these sources of information are integrated. This work proposes a method able to make probabilistic predictions of appraisals of other agents, using mental models of those agents. From these appraisal predictions, predictions about another agent's expressions are made, integrated with observations of the other agent's ambiguous emotional expressions using Bayesian techniques, resulting in updates to the agent's mental models. Our method is inspired by psychological work on human interpretation of emotional expressions. We demonstrate how these appraisals of others' emotions and observations of their expressions can be an integral part of an agent capable of Theory of Mind reasoning.

The prevalence of negotiation in social interaction has motivated researchers to develop virtual agents that can understand, facilitate, teach and even carry out negotiations. While much of this research has analyzed how to maximize the objective outcome, there is a growing body of work demonstrating that subjective perceptions of the outcome also play a critical role in human negotiation behavior. People derive subjective value from not only the outcome, but also from the process by which they achieve that outcome, from their relationship with their negotiation partner, etc. The affective responses evoked by these subjective valuations can be very different from what would be evoked by the objective outcome alone. We investigate such subjective valuations within human-agent negotiation in four variations of a wartime negotiation scenario taken from the political science literature. We observe that the objective outcomes of these negotiations are not strongly related to the human negotiators' subjective perceptions, as measured by the Subjective Value Index. We examine the game dynamics and agent behaviors to identify features that induce different subjective values in the participants, even when separated from the effect of the objective outcomes. We thus are able to identify characteristics of the interaction between negotiation and battlefield processes that most impact people's subjective valuations.

The prevalence of negotiation in social interaction has motivated researchers to develop virtual agents that can understand, facilitate, teach and even carry out negotiations. While much of this research has analyzed how to maximize the objective outcome, there is a growing body of work demonstrating that subjective perceptions of the outcome also play a critical role in human negotiation behavior. People derive subjective value from not only the outcome, but also from the process by which they achieve that outcome, from their relationship with their negotiation partner, etc. The affective responses evoked by these subjective valuations can be very different from what would be evoked by the objective outcome alone. We investigate such subjective valuations within human-agent negotiation in four variations of a wartime negotiation game. We observe that the objective outcomes of these negotiations are not strongly correlated with the human negotiators' subjective perceptions, as measured by the Subjective Value Index. We examine the game dynamics and agent behaviors to identify features that induce different subjective values in the participants. We thus are able to identify characteristics of the negotiation process and the agents' behavior that most impact people's subjective valuations in our wartime negotiation games.

Recent years are seeing an increasing need for on-line monitoring of teams of cooperating agents, e.g., for visualization, or performance tracking. However, in monitoring deployed teams, we often cannot rely on the agents to always communicate their state to the monitoring system. This paper presents a non-intrusive approach to monitoring by 'overhearing', where the monitored team's state is inferred (via plan-recognition) from team-members' *routine* communications, exchanged as part of their coordinated task execution, and observed (overheard) by the monitoring system. Key challenges in this approach include the demanding run-time requirements of monitoring, the scarceness of observations (increasing monitoring uncertainty), and the need to scale-up monitoring to address potentially large teams. To address these, we present a set of complementary novel techniques, exploiting knowledge of the social structures and procedures in the monitored team: (i) an efficient probabilistic plan-recognition algorithm, well-suited for processing communications as observations; (ii) an approach to exploiting knowledge of the team's social behavior to predict future observations during execution (reducing monitoring uncertainty); and (iii) monitoring algorithms that trade expressivity for scalability, representing only certain useful monitoring hypotheses, but allowing for any number of agents and their different activities to be represented in a single coherent entity. We present an empirical evaluation of these techniques, in combination and apart, in monitoring a deployed team of agents, running on machines physically distributed across the country, and engaged in complex, dynamic task execution. We also compare the performance of these techniques to human expert and novice monitors, and show that the techniques presented are capable of monitoring at human-expert levels, despite the difficulty of the task.

Despite the significant progress in multiagent teamwork, existing research does not address the optimality of its prescriptions nor the complexity of the teamwork problem. Without a characterization of the optimality-complexity tradeoffs, it is impossible to determine whether the assumptions and approximations made by a particular theory gain enough efficiency to justify the losses in overall performance. To provide a tool for use by multiagent researchers in evaluating this tradeoff, we present a unified framework, the COMmunicative Multiagent Team Decision Problem (COM-MTDP). The COM-MTDP model combines and extends existing multiagent theories, such as decentralized partially observable Markov decision processes and economic team theory. In addition to their generality of representation, COM-MTDPs also support the analysis of both the optimality of team performance and the computational complexity of the agents' decision problem. In analyzing complexity, we present a breakdown of the computational complexity of constructing optimal teams under various classes of problem domains, along the dimensions of observability and communication cost. In analyzing optimality, we exploit the COM-MTDP's ability to encode existing teamwork theories and models to encode two instantiations of joint intentions theory taken from the literature. Furthermore, the COM-MTDP model provides a basis for the development of novel team coordination algorithms. We derive a domain-independent criterion for optimal communication and provide a comparative analysis of the two joint intentions instantiations with respect to this optimal policy. We have implemented a reusable, domain-independent software package based on COM-MTDPs to analyze teamwork coordination strategies, and we demonstrate its use by encoding and evaluating the two joint intentions strategies within an example domain.

Adjustable autonomy refers to entities dynamically varying their own autonomy, transferring decision-making control to other entities (typically agents transferring control to human users) in key situations. Determining whether and when such transfers-of-control should occur is arguably the fundamental research problem in adjustable autonomy. Previous work has investigated various approaches to addressing this problem but has often focused on individual agent-human interactions. Unfortunately, domains requiring collaboration between teams of agents and humans reveal two key shortcomings of these previous approaches. First, these approaches use rigid one-shot transfers of control that can result in unacceptable coordination failures in multiagent settings. Second, they ignore costs (e.g., in terms of time delays or effects on actions) to an agent's team due to such transfers-of-control. To remedy these problems, this article presents a novel approach to adjustable autonomy, based on the notion of a transfer-of-control strategy. A transfer-of-control strategy consists of a conditional sequence of two types of actions: (i) actions to transfer decision-making control (e.g., from an agent to a user or vice versa) and (ii) actions to change an agent's pre-specified coordination constraints with team members, aimed at minimizing miscoordination costs. The goal is for high-quality individual decisions to be made with minimal disruption to the coordination of the team. We present a mathematical model of transfer-of-control strategies. The model guides and informs the operationalization of the strategies using Markov Decision Processes, which select an optimal strategy, given an uncertain environment and costs to the individuals and teams. The approach has been carefully evaluated, including via its use in a real-world, deployed multi-agent system that assists a research group in its daily activities.

Probabilistic context-free grammars (PCFGs) provide a simple way to represent a particular class of distributions over sentences in a context-free language. Efficient parsing algorithms for answering particular queries about a PCFG (i.e., calculating the probability of a given sentence, or finding the most likely parse) have been developed and applied to a variety of pattern-recognition problems. We extend the class of queries that can be answered in several ways: (1) allowing missing tokens in a sentence or sentence fragment, (2) supporting queries about intermediate structure, such as the presence of particular nonterminals, and (3) flexible conditioning on a variety of types of evidence. Our method works by constructing a Bayesian network to represent the distribution of parse trees induced by a given PCFG. The network structure mirrors that of the chart in a standard parser, and is generated using a similar dynamic programming approach. We present an algorithm for constructing Bayesian networks from PCFGs, and show how queries or patterns of queries on the network correspond to interesting queries on PCFGs. The network formalism also supports extensions to encode various context sensitivities within the probabilistic dependency structure.