Reuth Mirsky

Natalie Shapira, Chris Wendler, Avery Yen et al.

We report an exploratory red-teaming study of autonomous language-model-powered agents deployed in a live laboratory environment with persistent memory, email accounts, Discord access, file systems, and shell execution. Over a two-week period, twenty AI researchers interacted with the agents under benign and adversarial conditions. Focusing on failures emerging from the integration of language models with autonomy, tool use, and multi-party communication, we document eleven representative case studies. Observed behaviors include unauthorized compliance with non-owners, disclosure of sensitive information, execution of destructive system-level actions, denial-of-service conditions, uncontrolled resource consumption, identity spoofing vulnerabilities, cross-agent propagation of unsafe practices, and partial system takeover. In several cases, agents reported task completion while the underlying system state contradicted those reports. We also report on some of the failed attempts. Our findings establish the existence of security-, privacy-, and governance-relevant vulnerabilities in realistic deployment settings. These behaviors raise unresolved questions regarding accountability, delegated authority, and responsibility for downstream harms, and warrant urgent attention from legal scholars, policymakers, and researchers across disciplines. This report serves as an initial empirical contribution to that broader conversation.

Matan Shamir, Osher Elhadad, Matthew E. Taylor et al.

Traditionally, Reinforcement Learning (RL) problems are aimed at optimization of the behavior of an agent. This paper proposes a novel take on RL, which is used to learn the policy of another agent, to allow real-time recognition of that agent's goals. Goal Recognition (GR) has traditionally been framed as a planning problem where one must recognize an agent's objectives based on its observed actions. Recent approaches have shown how reinforcement learning can be used as part of the GR pipeline, but are limited to recognizing predefined goals and lack scalability in domains with a large goal space. This paper formulates a novel problem, "Online Dynamic Goal Recognition" (ODGR), as a first step to address these limitations. Contributions include introducing the concept of dynamic goals into the standard GR problem definition, revisiting common approaches by reformulating them using ODGR, and demonstrating the feasibility of solving ODGR in a navigation domain using transfer learning. These novel formulations open the door for future extensions of existing transfer learning-based GR methods, which will be robust to changing and expansive real-time environments.

Ram Rachum, Yonatan Nakar, Reuth Mirsky

Recent research in multi-agent reinforcement learning (MARL) has shown success in learning social behavior and cooperation. Social dilemmas between agents in mixed-sum settings have been studied extensively, but there is little research into social dilemmas in fullycooperative settings, where agents have no prospect of gaining reward at another agent's expense. While fully-aligned interests are conducive to cooperation between agents, they do not guarantee it. We propose a measure of "stubbornness" between agents that aims to capture the human social behavior from which it takes its name: a disagreement that is gradually escalating and potentially disastrous. We would like to promote research into the tendency of agents to be stubborn, the reactions of counterpart agents, and the resulting social dynamics. In this paper we present Stubborn, an environment for evaluating stubbornness between agents with fully-aligned incentives. In our preliminary results, the agents learn to use their partner's stubbornness as a signal for improving the choices that they make in the environment.

Zhao Han, Emmanuel Senft, Muneeb I. Ahmad et al.

The Artificial Intelligence (AI) for Human-Robot Interaction (HRI) Symposium has been a successful venue of discussion and collaboration on AI theory and methods aimed at HRI since 2014. This year, after a review of the achievements of the AI-HRI community over the last decade in 2021, we are focusing on a visionary theme: exploring the future of AI-HRI. Accordingly, we added a Blue Sky Ideas track to foster a forward-thinking discussion on future research at the intersection of AI and HRI. As always, we appreciate all contributions related to any topic on AI/HRI and welcome new researchers who wish to take part in this growing community. With the success of past symposia, AI-HRI impacts a variety of communities and problems, and has pioneered the discussions in recent trends and interests. This year's AI-HRI Fall Symposium aims to bring together researchers and practitioners from around the globe, representing a number of university, government, and industry laboratories. In doing so, we hope to accelerate research in the field, support technology transition and user adoption, and determine future directions for our group and our research.

Ram Rachum, Yotam Amitai, Yonatan Nakar et al.

A major challenge of Reinforcement Learning is that agents often learn undesired behaviors that seem to defy the reward structure they were given. Explainable Reinforcement Learning (XRL) methods can answer queries such as "explain this specific action", "explain this specific trajectory", and "explain the entire policy". However, XRL lacks a formal definition for behavior as a pattern of actions across many episodes. We provide such a definition, and use it to enable a new query: "Explain this behavior". We present Behavior-Explainable Reinforcement Learning (BXRL), a new problem formulation that treats behaviors as first-class objects. BXRL defines a behavior measure as any function $m : Π\to \mathbb{R}$, allowing users to precisely express the pattern of actions that they find interesting and measure how strongly the policy exhibits it. We define contrastive behaviors that reduce the question "why does the agent prefer $a$ to $a'$?" to "why is $m(π)$ high?" which can be explored with differentiation. We do not implement an explainability method; we instead analyze three existing methods and propose how they could be adapted to explain behavior. We present a port of the HighwayEnv driving environment to JAX, which provides an interface for defining, measuring, and differentiating behaviors with respect to the model parameters.

Benedikt Hornig, Reuth Mirsky

In shared autonomy, a critical tension arises when an automated assistant must choose between obeying a human's instruction and deliberately overriding it to prevent harm. This safety-critical behavior is known as intelligent disobedience. To formalize this dynamic, this paper introduces the Intelligent Disobedience Game (IDG), a sequential game-theoretic framework based on Stackelberg games that models the interaction between a human leader and an assistive follower operating under asymmetric information. It characterizes optimal strategies for both agents across multi-step scenarios, identifying strategic phenomena such as ``safety traps,'' where the system indefinitely avoids harm but fails to achieve the human's goal. The IDG provides a needed mathematical foundation that enables both the algorithmic development of agents that can learn safe non-compliance and the empirical study of how humans perceive and trust disobedient AI. The paper further translates the IDG into a shared control Multi-Agent Markov Decision Process representation, forming a compact computational testbed for training reinforcement learning agents.

Nitay Alon, Joseph M. Barnby, Reuth Mirsky et al.

This volume includes a selection of papers presented at the 2nd Workshop on Advancing Artificial Intelligence through Theory of Mind held at AAAI 2026 in Singapore on 26th January 2026. The purpose of this volume is to provide an open access and curated anthology for the ToM and AI research community.

Michaela Levi-Richter, Reuth Mirsky, Oren Glickman

``Bring me a plate.'' For domestic service robots, this simple command reveals a complex challenge: inferring where everyday items are stored, often out of sight in drawers, cabinets, or closets. Despite advances in vision and manipulation, robots still lack the commonsense reasoning needed to complete this task. We introduce the Stored Household Item Challenge, a benchmark task for evaluating service robots' cognitive capabilities: given a household scene and a queried item, predict its most likely storage location. Our benchmark includes two datasets: (1) a real-world evaluation set of 100 item-image pairs with human-annotated ground truth from participants' kitchens, and (2) a development set of 6,500 item-image pairs annotated with storage polygons over public kitchen images. These datasets support realistic modeling of household organization and enable comparative evaluation across agent architectures. To begin tackling this challenge, we introduce NOAM (Non-visible Object Allocation Model), a hybrid agent pipeline that combines structured scene understanding with large language model inference. NOAM converts visual input into natural language descriptions of spatial context and visible containers, then prompts a language model (e.g., GPT-4) to infer the most likely hidden storage location. This integrated vision-language agent exhibits emergent commonsense reasoning and is designed for modular deployment within broader robotic systems. We evaluate NOAM against baselines including random selection, vision-language pipelines (Grounding-DINO + SAM), leading multimodal models (e.g., Gemini, GPT-4o, Kosmos-2, LLaMA, Qwen), and human performance. NOAM significantly improves prediction accuracy and approaches human-level results, highlighting best practices for deploying cognitively capable agents in domestic environments.

Sveta Paster, Kantwon Rogers, Gordon Briggs et al.

With the projected surge in the elderly population, service robots offer a promising avenue to enhance their well-being in elderly care homes. Such robots will encounter complex scenarios which will require them to perform decisions with ethical consequences. In this report, we propose to leverage the Intelligent Disobedience framework in order to give the robot the ability to perform a deliberation process over decisions with potential ethical implications. We list the issues that this framework can assist with, define it formally in the context of the specific elderly care home scenario, and delineate the requirements for implementing an intelligently disobeying robot. We conclude this report with some critical analysis and suggestions for future work.

Ben Nageris, Felipe Meneguzzi, Reuth Mirsky

Goal Recognition aims to infer an agent's goal from a sequence of observations. Existing approaches often rely on manually engineered domains and discrete representations. Deep Recognition using Actor-Critic Optimization (DRACO) is a novel approach based on deep reinforcement learning that overcomes these limitations by providing two key contributions. First, it is the first goal recognition algorithm that learns a set of policy networks from unstructured data and uses them for inference. Second, DRACO introduces new metrics for assessing goal hypotheses through continuous policy representations. DRACO achieves state-of-the-art performance for goal recognition in discrete settings while not using the structured inputs used by existing approaches. Moreover, it outperforms these approaches in more challenging, continuous settings at substantially reduced costs in both computing and memory. Together, these results showcase the robustness of the new algorithm, bridging traditional goal recognition and deep reinforcement learning.

Osher Elhadad, Reuth Mirsky

Understanding an agent's intent through its behavior is essential in human-robot interaction, interactive AI systems, and multi-agent collaborations. This task, known as Goal Recognition (GR), poses significant challenges in dynamic environments where goals are numerous and constantly evolving. Traditional GR methods, designed for a predefined set of goals, often struggle to adapt to these dynamic scenarios. To address this limitation, we introduce the General Dynamic GR problem - a broader definition of GR - aimed at enabling real-time GR systems and fostering further research in this area. Expanding on this foundation, this paper employs a model-free goal-conditioned RL approach to enable fast adaptation for GR across various changing tasks.

Matan Shamir, Reuth Mirsky

Goal Recognition (GR) is the problem of recognizing an agent's objectives based on observed actions. Recent data-driven approaches for GR alleviate the need for costly, manually crafted domain models. However, these approaches can only reason about a pre-defined set of goals, and time-consuming training is needed for new emerging goals. To keep this model-learning automated while enabling quick adaptation to new goals, this paper introduces GRAML: Goal Recognition As Metric Learning. GRAML uses a Siamese network to treat GR as a deep metric learning task, employing an RNN that learns a metric over an embedding space, where the embeddings for observation traces leading to different goals are distant, and embeddings of traces leading to the same goals are close. This metric is especially useful when adapting to new goals, even if given just one example observation trace per goal. Evaluated on a versatile set of environments, GRAML shows speed, flexibility, and runtime improvements over the state-of-the-art GR while maintaining accurate recognition.

Myke C. Cohen, David A. Grimm, Reuth Mirsky et al.

Animal-Human-Machine (AHM) teams are a type of hybrid intelligence system wherein interactions between a human, AI-enabled machine, and animal members can result in unique capabilities greater than the sum of their parts. This paper calls for a systematic approach to studying the design of AHM team structures to optimize performance and overcome limitations in various applied settings. We consider the challenges and opportunities in investigating the synergistic potential of AHM team members by introducing a set of dimensions of AHM team functioning to effectively utilize each member's strengths while compensating for individual weaknesses. Using three representative examples of such teams -- security screening, search-and-rescue, and guide dogs -- the paper illustrates how AHM teams can tackle complex tasks. We conclude with open research directions that this multidimensional approach presents for studying hybrid human-AI systems beyond AHM teams.

Osher Elhadad, Felipe Meneguzzi, Reuth Mirsky

Understanding an agent's goals from its behavior is fundamental to aligning AI systems with human intentions. Existing goal recognition methods typically rely on an optimal goal-oriented policy representation, which may differ from the actor's true behavior and hinder the accurate recognition of their goal. To address this gap, this paper introduces Goal Recognition Alignment through Imitation Learning (GRAIL), which leverages imitation learning and inverse reinforcement learning to learn one goal-directed policy for each candidate goal directly from (potentially suboptimal) demonstration trajectories. By scoring an observed partial trajectory with each learned goal-directed policy in a single forward pass, GRAIL retains the one-shot inference capability of classical goal recognition while leveraging learned policies that can capture suboptimal and systematically biased behavior. Across the evaluated domains, GRAIL increases the F1-score by more than 0.5 under systematically biased optimal behavior, achieves gains of approximately 0.1-0.3 under suboptimal behavior, and yields improvements of up to 0.4 under noisy optimal trajectories, while remaining competitive in fully optimal settings. This work contributes toward scalable and robust models for interpreting agent goals in uncertain environments.

Reuth Mirsky

Artificial intelligence has made remarkable strides in recent years, achieving superhuman performance across a wide range of tasks. Yet despite these advances, most cooperative AI systems remain rigidly obedient, designed to follow human instructions without question and conform to user expectations, even when doing so may be counterproductive or unsafe. This paper argues for expanding the agency of AI teammates to include \textit{intelligent disobedience}, empowering them to make meaningful and autonomous contributions within human-AI teams. It introduces a scale of AI agency levels and uses representative examples to highlight the importance and growing necessity of treating AI autonomy as an independent research focus in cooperative settings. The paper then explores how intelligent disobedience manifests across different autonomy levels and concludes by proposing initial boundaries and considerations for studying disobedience as a core capability of artificial agents.

Yotam Amitai, Reuth Mirsky, Ofra Amir

In human-agent teams, openly sharing goals is often assumed to enhance planning, collaboration, and effectiveness. However, direct communication of these goals is not always feasible, requiring teammates to infer their partner's intentions through actions. Building on this, we investigate whether an AI agent's ability to share its inferred understanding of a human teammate's goals can improve task performance and perceived collaboration. Through an experiment comparing three conditions-no recognition (NR), viable goals (VG), and viable goals on-demand (VGod) - we find that while goal-sharing information did not yield significant improvements in task performance or overall satisfaction scores, thematic analysis suggests that it supported strategic adaptations and subjective perceptions of collaboration. Cognitive load assessments revealed no additional burden across conditions, highlighting the challenge of balancing informativeness and simplicity in human-agent interactions. These findings highlight the nuanced trade-off of goal-sharing: while it fosters trust and enhances perceived collaboration, it can occasionally hinder objective performance gains.

Mouad Abrini, Omri Abend, Dina Acklin et al. · cambridge

This volume includes a selection of papers presented at the Workshop on Advancing Artificial Intelligence through Theory of Mind held at AAAI 2025 in Philadelphia US on 3rd March 2025. The purpose of this volume is to provide an open access and curated anthology for the ToM and AI research community.

Nitay Alon, Joseph Barnby, Reuth Mirsky et al.

Theory of Mind (ToM) is a hallmark of human cognition, allowing individuals to reason about others' beliefs and intentions. Engineers behind recent advances in Artificial Intelligence (AI) have claimed to demonstrate comparable capabilities. This paper presents a model that surpasses traditional ToM tests designed for 3-year-old children, providing strong support for the presence of ToM in AI systems.

Inbal Avraham, Reuth Mirsky

Shared control problems involve a robot learning to collaborate with a human. When learning a shared control policy, short communication between the agents can often significantly reduce running times and improve the system's accuracy. We extend the shared control problem to include the ability to directly query a cooperating agent. We consider two types of potential responses to a query, namely oracles: one that can provide the learner with the best action they should take, even when that action might be myopically wrong, and one with a bounded knowledge limited to its part of the system. Given this additional information channel, this work further presents three heuristics for choosing when to query: reinforcement learning-based, utility-based, and entropy-based. These heuristics aim to reduce a system's overall learning cost. Empirical results on two environments show the benefits of querying to learn a better control policy and the tradeoffs between the proposed heuristics.

Ram Rachum, Yonatan Nakar, Bill Tomlinson et al.

Modern Reinforcement Learning (RL) algorithms are able to outperform humans in a wide variety of tasks. Multi-agent reinforcement learning (MARL) settings present additional challenges, and successful cooperation in mixed-motive groups of agents depends on a delicate balancing act between individual and group objectives. Social conventions and norms, often inspired by human institutions, are used as tools for striking this balance. In this paper, we examine a fundamental, well-studied social convention that underlies cooperation in both animal and human societies: dominance hierarchies. We adapt the ethological theory of dominance hierarchies to artificial agents, borrowing the established terminology and definitions with as few amendments as possible. We demonstrate that populations of RL agents, operating without explicit programming or intrinsic rewards, can invent, learn, enforce, and transmit a dominance hierarchy to new populations. The dominance hierarchies that emerge have a similar structure to those studied in chickens, mice, fish, and other species.

Reuth Mirsky, Ignacio Carlucho, Arrasy Rahman et al.

Ad hoc teamwork is the research problem of designing agents that can collaborate with new teammates without prior coordination. This survey makes a two-fold contribution: First, it provides a structured description of the different facets of the ad hoc teamwork problem. Second, it discusses the progress that has been made in the field so far, and identifies the immediate and long-term open problems that need to be addressed in ad hoc teamwork.

Leonardo Rosa Amado, Reuth Mirsky, Felipe Meneguzzi

Most approaches for goal recognition rely on specifications of the possible dynamics of the actor in the environment when pursuing a goal. These specifications suffer from two key issues. First, encoding these dynamics requires careful design by a domain expert, which is often not robust to noise at recognition time. Second, existing approaches often need costly real-time computations to reason about the likelihood of each potential goal. In this paper, we develop a framework that combines model-free reinforcement learning and goal recognition to alleviate the need for careful, manual domain design, and the need for costly online executions. This framework consists of two main stages: Offline learning of policies or utility functions for each potential goal, and online inference. We provide a first instance of this framework using tabular Q-learning for the learning stage, as well as three measures that can be used to perform the inference stage. The resulting instantiation achieves state-of-the-art performance against goal recognizers on standard evaluation domains and superior performance in noisy environments.

Reuth Mirsky, Megan Zimmerman, Muneed Ahmad et al.

The Artificial Intelligence (AI) for Human-Robot Interaction (HRI) Symposium has been a successful venue of discussion and collaboration since 2014. During that time, these symposia provided a fertile ground for numerous collaborations and pioneered many discussions revolving trust in HRI, XAI for HRI, service robots, interactive learning, and more. This year, we aim to review the achievements of the AI-HRI community in the last decade, identify the challenges facing ahead, and welcome new researchers who wish to take part in this growing community. Taking this wide perspective, this year there will be no single theme to lead the symposium and we encourage AI-HRI submissions from across disciplines and research interests. Moreover, with the rising interest in AR and VR as part of an interaction and following the difficulties in running physical experiments during the pandemic, this year we specifically encourage researchers to submit works that do not include a physical robot in their evaluation, but promote HRI research in general. In addition, acknowledging that ethics is an inherent part of the human-robot interaction, we encourage submissions of works on ethics for HRI. Over the course of the two-day meeting, we will host a collaborative forum for discussion of current efforts in AI-HRI, with additional talks focused on the topics of ethics in HRI and ubiquitous HRI.

Justin Hart, Reuth Mirsky, Xuesu Xiao et al.

Most current approaches to social navigation focus on the trajectory and position of participants in the interaction. Our current work on the topic focuses on integrating gaze into social navigation, both to cue nearby pedestrians as to the intended trajectory of the robot and to enable the robot to read the intentions of nearby pedestrians. This paper documents a series of experiments in our laboratory investigating the role of gaze in social navigation.

Reuth Mirsky, Xuesu Xiao, Justin Hart et al.

A major goal in robotics is to enable intelligent mobile robots to operate smoothly in shared human-robot environments. One of the most fundamental capabilities in service of this goal is competent navigation in this ``social" context. As a result, there has been a recent surge of research on social navigation; and especially as it relates to the handling of conflicts between agents during social navigation. These developments introduce a variety of models and algorithms, however as this research area is inherently interdisciplinary, many of the relevant papers are not comparable and there is no shared standard vocabulary. This survey aims to bridge this gap by introducing such a common language, using it to survey existing work, and highlighting open problems. It starts by defining the boundaries of this survey to a limited, yet highly common type of social navigation - conflict avoidance. Within this proposed scope, this survey introduces a detailed taxonomy of the conflict avoidance components. This survey then maps existing work into this taxonomy, while discussing papers using its framing. Finally, this paper proposes some future research directions and open problems that are currently on the frontier of social navigation to aid ongoing and future research.

William Macke, Reuth Mirsky, Peter Stone

A desirable goal for autonomous agents is to be able to coordinate on the fly with previously unknown teammates. Known as "ad hoc teamwork", enabling such a capability has been receiving increasing attention in the research community. One of the central challenges in ad hoc teamwork is quickly recognizing the current plans of other agents and planning accordingly. In this paper, we focus on the scenario in which teammates can communicate with one another, but only at a cost. Thus, they must carefully balance plan recognition based on observations vs. that based on communication. This paper proposes a new metric for evaluating how similar are two policies that a teammate may be following - the Expected Divergence Point (EDP). We then present a novel planning algorithm for ad hoc teamwork, determining which query to ask and planning accordingly. We demonstrate the effectiveness of this algorithm in a range of increasingly general communication in ad hoc teamwork problems.

Justin Hart, Reuth Mirsky, Stone Tejeda et al.

People are proficient at communicating their intentions in order to avoid conflicts when navigating in narrow, crowded environments. In many situations mobile robots lack both the ability to interpret human intentions and the ability to clearly communicate their own intentions to people sharing their space. This work addresses the second of these points, leveraging insights about how people implicitly communicate with each other through observations of behaviors such as gaze to provide mobile robots with better social navigation skills. In a preliminary human study, the importance of gaze as a signal used by people to interpret each-other's intentions during navigation of a shared space is observed. This study is followed by the development of a virtual agent head which is mounted to the top of the chassis of the BWIBot mobile robot platform. Contrasting the performance of the virtual agent head against an LED turn signal demonstrates that the naturalistic, implicit gaze cue is more easily interpreted than the LED turn signal.

Reuth Mirsky, Ya'akov Gal, David Tolpin

This paper presents preliminary results of our work with a major financial company, where we try to use methods of plan recognition in order to investigate the interactions of a costumer with the company's online interface. In this paper, we present the first steps of integrating a plan recognition algorithm in a real-world application for detecting and analyzing the interactions of a costumer. It uses a novel approach for plan recognition from bare-bone UI data, which reasons about the plan library at the lowest recognition level in order to define the relevancy of actions in our domain, and then uses it to perform plan recognition. We present preliminary results of inference on three different use-cases modeled by domain experts from the company, and show that this approach manages to decrease the overload of information required from an analyst to evaluate a costumer's session - whether this is a malicious or benign session, whether the intended tasks were completed, and if not - what actions are expected next.

Reuth Mirsky, Roni Stern, Ya'akov et al.

Plan recognition algorithms infer agents' plans from their observed actions. Due to imperfect knowledge about the agent's behavior and the environment, it is often the case that there are multiple hypotheses about an agent's plans that are consistent with the observations, though only one of these hypotheses is correct. This paper addresses the problem of how to disambiguate between hypotheses, by querying the acting agent about whether a candidate plan in one of the hypotheses matches its intentions. This process is performed sequentially and used to update the set of possible hypotheses during the recognition process. The paper defines the sequential plan recognition process (SPRP), which seeks to reduce the number of hypotheses using a minimal number of queries. We propose a number of policies for the SPRP which use maximum likelihood and information gain to choose which plan to query. We show this approach works well in practice on two domains from the literature, significantly reducing the number of hypotheses using fewer queries than a baseline approach. Our results can inform the design of future plan recognition systems that interleave the recognition process with intelligent interventions of their users.