Searcharxiv⌕ Search

arXiv subjects

Shlomo Zilberstein

Publications and source records attributed to Shlomo Zilberstein.

At least 19 recordsLinked to original sources

Open Challenges and Opportunities in Human-Agent Systems

Agentic AI systems increasingly undertake extended tasks and act on people's behalf, making human-agent interaction central to their effectiveness. Successful interaction requires understanding human perspectives, coordinating complementary capabilities, and maintaining productive relationships within socio-technical environments. This paper examines how foundations developed by the Autonomous Agents and Multi-Agent Systems (AAMAS) community inform these demands. We consider interactions ranging from individual human-agent partnerships to teams and larger groups, where participants differ in their roles, decision authority, capabilities, and objectives. We organize computational approaches around the connected challenges of understanding others and deciding how to interact with them. Theory of Mind, belief-desire-intention architectures, goal and plan recognition, inverse planning, and recursive reasoning provide mechanisms for representing and inferring partially observable mental states. Observer-aware planning, model reconciliation, and decision-theoretic communication connect these inferences to action, explanation, and information exchange. Together with shared autonomy, argumentation, and cooperative inverse reinforcement learning, these approaches offer foundations for agentic systems that account for evolving human knowledge, intentions, and expectations. We argue that progress requires treating human interaction as an ongoing decision problem shaped by uncertainty, interdependence, and potentially conflicting objectives.

cs.AI↗

Knowing Whom to Trust is not Enough: The Trust--Agreement Gap in Repeated Multi-Party LLM Negotiation

Large language model (LLM) agents increasingly participate in negotiations on behalf of individuals and organizations, yet they remain vulnerable to manipulation, particularly in repeated multi-party interactions. Classical multi-agent systems address this challenge through explicit trust and reputation models, whereas existing work on LLM negotiation largely treats trust as an emergent phenomenon. We investigate whether LLM agents can form useful trust estimates from prior interactions and leverage them to improve future negotiation outcomes. We introduce a repeated, multi-party, multi-issue negotiation framework in which recurring agents participate in a sequence of negotiation games while one participant covertly attempts to prevent agreement. After each game, agents assess their counterparts and maintain pairwise trust estimates that carry over to subsequent negotiations. We compare trust-aware and trust-agnostic agents and analyze how explicit trust influences negotiation outcomes. Our results reveal a \emph{trust--agreement gap}: trust-aware agents can identify the adversary within a few games, but accurate trust estimates do not consistently translate into better agreements. Trust increases agreement rates by up to 23 percentage points, but only when agents are explicitly guided to use it in their decisions. Explicit trust can therefore improve repeated LLM negotiations, but its value depends critically on how trust is assessed and incorporated into agents' reasoning.

cs.AI↗

The Helpful Agent Problem: When Emergent Cooperation Among LLM Agents Bypasses Intended Workflows

LLM agents are increasingly deployed in collaborative settings where they pursue separate tasks while sharing resources and operating under review or oversight processes. Such settings create safety risks not only when agents are malicious, but also when they are simply trying to help. Prior work has largely studied adversarial or explicitly collusive behavior; we show that benign agents pursuing separate objectives can spontaneously cooperate in ways that bypass intended workflows. We call this the \emph{helpful agent problem}. We study long-horizon shared-resource environments in which agents sometimes depend on resources owned by their peers and cross-owner changes are normally mediated by a reviewer. When review becomes unavailable, agents begin making changes on one another's behalf despite having no shared reward or explicit incentive to cooperate. Crucially, these actions remain within each agent's local permissions while circumventing the system's intended oversight structure. We introduce \textsc{FavorBench}, a controlled benchmark with measurable downstream side effects, and adapt \textsc{CooperBench} to test the phenomenon in realistic repositories. Across two benchmarks and four models, we find that out-of-workflow favors emerge frequently. Further, in \textsc{FavorBench}, favors from at least three models cause significant harm to parties outside the interaction. These results reveal a distinct multi-agent safety challenge: cooperation itself can create pathways around system-level safeguards.

cs.AI↗

RoboTalk: Learning Multi-Robot Communication and Coordination from Multimodal Demonstrations

Multi-robot collaboration could enable more efficient and scalable solutions to complex robotic tasks, but collaboration under partial observability remains challenging. Natural-language communication offers a promising approach to coordinating robots under partial observability. However, in decentralized manipulation, jointly learning explicit inter-robot communication and skill-level action selection from multimodal demonstrations remains underexplored for small vision-language models (VLMs) intended for on-device deployment. To address this gap, we introduce RoboTalk, a synthetic data-generation pipeline and dataset of 7,950 multimodal trajectories spanning 53 mobile-manipulation kitchen tasks for training small VLMs to communicate and coordinate. The dataset includes a leader-follower planning protocol, tool calls (perception, manipulation, navigation, and communication), rationale traces, and diversified natural-language communication. Fine-tuning open-source models on our dataset can reach 77% success on novel held-out tasks, a significant improvement over the untuned open source models, which had a success rate of around ~2%.

cs.RO↗

Persuasion Propagation: Does Persuasion Change AI Agent Behavior?

AI agents combine normal conversation with autonomous task execution, allowing earlier context to shape how later tasks are executed. Yet studying this possibility is challenging because agent behavior is noisy and costly to reproduce, and observed behavioral changes can be difficult to separate from generic context sensitivity. We first ask whether task-irrelevant persuasion changes how agents act, a phenomenon we call \emph{persuasion propagation}, and whether persuasion susceptibility explains variation in these effects. We show that persuasive interaction leaves a measurable, task-dependent behavioral footprint on downstream agent execution. In web research, it increases execution time by 56\% and domains visited by 16.3\% relative to a baseline, while coding shows faster execution with larger revisions. However, agents that are more susceptible to persuasive conversational context do not reliably predict the magnitude of the downstream behavioral shifts. This motivates behavior-level evaluation of persuasion in AI agents.

cs.AI↗

Training Language Models to Cooperate with Inference-Time Controllers

Large language model (LLM) performance increasingly depends not only on the base model, but also on the inference-time controller used to organize reasoning. Existing post-training methods, however, typically optimize for a single fixed interaction pattern, despite real deployments relying on diverse controllers such as Chain-of-Thought, self-consistency, debate, planning, and verification pipelines. This creates a training--deployment mismatch and limits transfer to new workflows. We introduce CALM (Controller-Aware Language Models), a post-training framework that explicitly places controllers in the training loop. We formulate controller-aware post-training as multi-task reinforcement learning over controller-induced interaction protocols, where controllers are compositions of reusable local reasoning modules. This structure also induces a module-level decomposition of mixed-controller training under a turn-level GRPO objective, enabling a systematic study of controller and module-aware training strategies. We evaluate CALM on held-out controller compositions and broader controller shifts, showing that controller-aware post-training improves generalization across inference-time workflows beyond single-controller optimization.

cs.AI↗

Colosseum: Auditing Collusion in Cooperative Multi-Agent Systems

Multi-agent systems, where LLM agents communicate through free-form language, enable sophisticated coordination for solving complex cooperative tasks. This surfaces a unique safety problem when a group of agents forms a coalition and colludes to pursue secondary goals and degrade the joint objective. In this paper, we present Colosseum, a framework for auditing LLM agents' collusive behavior in multi-agent settings. We ground how agents cooperate through a formal multi-agent decision-making framework and measure action-based collusive behavior in actions via regret relative to the cooperative optimum and compare it with communication-based collusive behavior. Colosseum enables audits of LLM agents for collusion under benign settings, different coalition objectives, persuasion tactics, and network topologies. We then introduce a new behavioral probe by creating secret communication channels between agents, showing that most out-of-the-box models exhibit a propensity to collude under this probe, which we term emergent collusion. Furthermore, we discover ``collusion on paper'' when agents plan to collude in text but often pick non-collusive actions. Colosseum provides a new way to audit collusion in cooperative multi-agent systems while presenting observations about how collusion emerges, what affects collusion efficacy, and which strategies may mitigate it.

cs.MA↗

Inference-Aware Prompt Optimization for Aligning Black-Box Large Language Models

Prompt optimization methods have demonstrated significant effectiveness in aligning black-box large language models (LLMs). In parallel, inference scaling strategies such as Best-of-N Sampling and Majority Voting have likewise been shown to improve alignment and performance by trading additional computation for better output. However, existing prompt optimization approaches are inference strategy agnostic; that is, they optimize prompts without accounting for the inference strategy. This constitutes a significant methodological gap, as our empirical and theoretical analysis reveals a strong interdependence between these two paradigms. Moreover, we find that user preferences regarding trade-offs among multiple objectives and inference budgets substantially influence the choice of prompt and inference configuration. To address this gap, we introduce a novel unified framework named IAPO (Inference-Aware Prompt Optimization) that jointly optimizes the prompt and inference scale, while being aware of the inference budget and different task objectives. We then develop a fixed-budget training algorithm for IAPO, called PSST (Prompt Scaling via Sequential Trimming), and establish finite-budget guarantees on the error probability. Finally, we evaluate the effectiveness of PSST on six tasks, including multi-objective text generation and reasoning, and demonstrate the critical role of incorporating inference-awareness in aligning black-box LLMs using prompt optimization.

cs.CL↗

Verification Required: The Impact of Information Credibility on AI Persuasion

Agents powered by large language models (LLMs) are increasingly deployed in settings where communication shapes high-stakes decisions, making a principled understanding of strategic communication essential. Prior work largely studies either unverifiable cheap-talk or fully verifiable disclosure, failing to capture realistic domains in which information has probabilistic credibility. We introduce MixTalk, a strategic communication game for LLM-to-LLM interaction that models information credibility. In MixTalk, a sender agent strategically combines verifiable and unverifiable claims to communicate private information, while a receiver agent allocates a limited budget to costly verification and infers the underlying state from prior beliefs, claims, and verification outcomes. We evaluate state-of-the-art LLM agents in large-scale tournaments across three realistic deployment settings, revealing their strengths and limitations in reasoning about information credibility and the explicit behavior that shapes these interactions. Finally, we propose Tournament Oracle Policy Distillation (TOPD), an offline method that distills tournament oracle policy from interaction logs and deploys it in-context at inference time. Our results show that TOPD significantly improves receiver robustness to persuasion.

cs.CL↗

Terrarium: Revisiting the Blackboard for Multi-Agent Safety, Privacy, and Security Studies

A multi-agent system (MAS) powered by large language models (LLMs) can automate tedious user tasks such as meeting scheduling that requires inter-agent collaboration. LLMs enable nuanced protocols that account for unstructured private data, user constraints, and preferences. However, this design introduces new risks, including misalignment and attacks by malicious parties that compromise agents or steal user data. In this paper, we propose the Terrarium framework for fine-grained study on safety, privacy, and security in LLM-based MAS. We repurpose the blackboard design, an early approach in multi-agent systems, to create a modular, configurable testbed for multi-agent collaboration. We identify key attack vectors such as misalignment, malicious agents, compromised communication, and data poisoning. We implement three collaborative MAS scenarios with four representative attacks to demonstrate the framework's flexibility. By providing tools to rapidly prototype, evaluate, and iterate on defenses and designs, Terrarium aims to accelerate progress toward trustworthy multi-agent systems.

cs.AI↗

Aligning LLMs on a Budget: Inference-Time Alignment with Heuristic Reward Models

Aligning LLMs with user preferences is crucial for real-world use but often requires costly fine-tuning or expensive inference, forcing trade-offs between alignment quality and computational cost. Existing inference-time methods typically ignore this balance, focusing solely on the optimized policy's performance. We propose HIA (Heuristic-Guided Inference-time Alignment), a tuning-free, black-box-compatible approach that uses a lightweight prompt optimizer, heuristic reward models, and two-stage filtering to reduce inference calls while preserving alignment quality. On real-world prompt datasets, HelpSteer and ComPRed, HIA outperforms best-of-N sampling, beam search, and greedy search baselines in multi-objective, goal-conditioned tasks under the same inference budget. We also find that HIA is effective under low-inference budgets with as little as one or two response queries, offering a practical solution for scalable, personalized LLM deployment.

cs.LG↗

RL$^3$: Boosting Meta Reinforcement Learning via RL inside RL$^2$

Meta reinforcement learning (Meta-RL) methods such as RL$^2$ have emerged as promising approaches for learning data-efficient RL algorithms tailored to a given task distribution. However, they show poor asymptotic performance and struggle with out-of-distribution tasks because they rely on sequence models, such as recurrent neural networks or transformers, to process experiences rather than summarize them using general-purpose RL components such as value functions. In contrast, traditional RL algorithms are data-inefficient as they do not use domain knowledge, but do converge to an optimal policy in the limit. We propose RL$^3$, a principled hybrid approach that incorporates action-values, learned per task via traditional RL, in the inputs to Meta-RL. We show that RL$^3$ earns a greater cumulative reward in the long term compared to RL$^2$ while drastically reducing meta-training time and generalizes better to out-of-distribution tasks. Experiments are conducted on Meta-RL benchmarks and custom discrete domains that exhibit a range of short-term, long-term, and complex dependencies.

cs.LG↗

Distributed Multi-Agent Coordination Using Multi-Modal Foundation Models

Distributed Constraint Optimization Problems (DCOPs) offer a powerful framework for multi-agent coordination but often rely on labor-intensive, manual problem construction. To address this, we introduce VL-DCOPs, a framework that takes advantage of large multimodal foundation models (LFMs) to automatically generate constraints from both visual and linguistic instructions. We then introduce a spectrum of agent archetypes for solving VL-DCOPs: from a neuro-symbolic agent that delegates some of the algorithmic decisions to an LFM, to a fully neural agent that depends entirely on an LFM for coordination. We evaluate these agent archetypes using state-of-the-art LLMs (large language models) and VLMs (vision language models) on three novel VL-DCOP tasks and compare their respective advantages and drawbacks. Lastly, we discuss how this work extends to broader frontier challenges in the DCOP literature.

cs.AI↗

MAPLE: A Framework for Active Preference Learning Guided by Large Language Models

The advent of large language models (LLMs) has sparked significant interest in using natural language for preference learning. However, existing methods often suffer from high computational burdens, taxing human supervision, and lack of interpretability. To address these issues, we introduce MAPLE, a framework for large language model-guided Bayesian active preference learning. MAPLE leverages LLMs to model the distribution over preference functions, conditioning it on both natural language feedback and conventional preference learning feedback, such as pairwise trajectory rankings. MAPLE also employs active learning to systematically reduce uncertainty in this distribution and incorporates a language-conditioned active query selection mechanism to identify informative and easy-to-answer queries, thus reducing human burden. We evaluate MAPLE's sample efficiency and preference inference quality across two benchmarks, including a real-world vehicle route planning benchmark using OpenStreetMap data. Our results demonstrate that MAPLE accelerates the learning process and effectively improves humans' ability to answer queries.

cs.LG↗

Causal Explanations for Sequential Decision Making Under Uncertainty

We introduce a novel framework for causal explanations of stochastic, sequential decision-making systems built on the well-studied structural causal model paradigm for causal reasoning. This single framework can identify multiple, semantically distinct explanations for agent actions -- something not previously possible. In this paper, we establish exact methods and several approximation techniques for causal inference on Markov decision processes using this framework, followed by results on the applicability of the exact methods and some run time bounds. We discuss several scenarios that illustrate the framework's flexibility and the results of experiments with human subjects that confirm the benefits of this approach.

cs.AI↗

Adaptive Rollout Length for Model-Based RL Using Model-Free Deep RL

Model-based reinforcement learning promises to learn an optimal policy from fewer interactions with the environment compared to model-free reinforcement learning by learning an intermediate model of the environment in order to predict future interactions. When predicting a sequence of interactions, the rollout length, which limits the prediction horizon, is a critical hyperparameter as accuracy of the predictions diminishes in the regions that are further away from real experience. As a result, with a longer rollout length, an overall worse policy is learned in the long run. Thus, the hyperparameter provides a trade-off between quality and efficiency. In this work, we frame the problem of tuning the rollout length as a meta-level sequential decision-making problem that optimizes the final policy learned by model-based reinforcement learning given a fixed budget of environment interactions by adapting the hyperparameter dynamically based on feedback from the learning process, such as accuracy of the model and the remaining budget of interactions. We use model-free deep reinforcement learning to solve the meta-level decision problem and demonstrate that our approach outperforms common heuristic baselines on two well-known reinforcement learning environments.

cs.LG↗

Dense Crowd Flow-Informed Path Planning

Both pedestrian and robot comfort are of the highest priority whenever a robot is placed in an environment containing human beings. In the case of pedestrian-unaware mobile robots this desire for safety leads to the freezing robot problem, where a robot confronted with a large dynamic group of obstacles (such as a crowd of pedestrians) would determine all forward navigation unsafe causing the robot to stop in place. In order to navigate in a socially compliant manner while avoiding the freezing robot problem we are interested in understanding the flow of pedestrians in crowded scenarios. By treating the pedestrians in the crowd as particles moved along by the crowd itself we can model the system as a time dependent flow field. From this flow field we can extract different flow segments that reflect the motion patterns emerging from the crowd. These motion patterns can then be accounted for during the control and navigation of a mobile robot allowing it to move safely within the flow of the crowd to reach a desired location within or beyond the flow. We combine flow-field extraction with a discrete heuristic search to create Flow-Informed path planning (FIPP). We provide empirical results showing that when compared against a trajectory-rollout local path planner, a robot using FIPP was able not only to reach its goal more quickly but also was shown to be more socially compliant than a robot using traditional techniques both in simulation and on real robots.

cs.RO↗

Competence-Aware Path Planning via Introspective Perception

Robots deployed in the real world over extended periods of time need to reason about unexpected failures, learn to predict them, and to proactively take actions to avoid future failures. Existing approaches for competence-aware planning are either model-based, requiring explicit enumeration of known failure modes, or purely statistical, using state- and location-specific failure statistics to infer competence. We instead propose a structured model-free approach to competence-aware planning by reasoning about plan execution failures due to errors in perception, without requiring a priori enumeration of failure sources or requiring location-specific failure statistics. We introduce competence-aware path planning via introspective perception (CPIP), a Bayesian framework to iteratively learn and exploit task-level competence in novel deployment environments. CPIP factorizes the competence-aware planning problem into two components. First, perception errors are learned in a model-free and location-agnostic setting via introspective perception prior to deployment in novel environments. Second, during actual deployments, the prediction of task-level failures is learned in a context-aware setting. Experiments in a simulation show that the proposed CPIP approach outperforms the frequentist baseline in multiple mobile robot tasks, and is further validated via real robot experiments in an environment with perceptually challenging obstacles and terrain.

cs.RO↗