Substantial research on deep learning-based emergent communication uses the referential game framework, specifically the Lewis signaling game, however we argue that successful communication in this game typically only need one or two symbols for target image classification because of a sampling pitfall in the training data. To address this issue, we provide a theoretical analysis and introduce a combinatorial algorithm SolveMinSym (SMS) to solve the symbolic complexity for classification, which is the minimum number of symbols in the message for successful communication. We use the SMS algorithm to create datasets with different symbolic complexity to empirically show that data with higher symbolic complexity increases the number of effective symbols in the emergent language.

Recent advancements in Web agents have introduced novel architectures and benchmarks showcasing progress in autonomous web navigation and interaction. However, most existing benchmarks prioritize effectiveness and accuracy, overlooking factors like safety and trustworthiness which are essential for deploying web agents in enterprise settings. We present STWebAgentBench, a benchmark designed to evaluate web agents safety and trustworthiness across six critical dimensions, essential for reliability in enterprise applications. This benchmark is grounded in a detailed framework that defines safe and trustworthy (ST) agent behavior. Our work extends WebArena with safety templates and evaluation functions to assess safety policy compliance rigorously. We introduce the Completion Under Policy to measure task success while adhering to policies, alongside the Risk Ratio, which quantifies policy violations across dimensions, providing actionable insights to address safety gaps. Our evaluation reveals that current SOTA agents struggle with policy adherence and cannot yet be relied upon for critical business applications. We open-source this benchmark and invite the community to contribute, with the goal of fostering a new generation of safer, more trustworthy AI agents. All code, data, environment reproduction resources, and video demonstrations are available at https://sites.google.com/view/st-webagentbench/home.

The AI safety literature is full of examples of powerful AI agents that, in blindly pursuing a specific and usually narrow objective, ends up with unacceptable and even catastrophic collateral damage to others. In this paper, we consider the problem of social harms that can result from actions taken by learning and utility-maximising agents in a multi-agent environment. The problem of measuring social harms or impacts in such multi-agent settings, especially when the agents are artificial generally intelligent (AGI) agents, was listed as an open problem in Everitt et al, 2018. We attempt a partial answer to that open problem in the form of market-based mechanisms to quantify and control the cost of such social harms. The proposed setup captures many well-studied special cases and is more general than existing formulations of multi-agent reinforcement learning with mechanism design in two ways: (i) the underlying environment is a history-based general reinforcement learning environment like in AIXI; (ii) the reinforcement-learning agents participating in the environment can have different learning strategies and planning horizons. To demonstrate the practicality of the proposed setup, we survey some key classes of learning algorithms and present a few applications, including a discussion of the Paperclips problem and pollution control with a cap-and-trade system.

Enabling effective collaboration among LLMs is a crucial step toward developing autonomous systems capable of solving complex problems. While LLMs are typically used as single-model generators, where humans critique and refine their outputs, the potential for jointly-trained collaborative models remains largely unexplored. Despite promising results in multi-agent communication and debate settings, little progress has been made in training models to work together on tasks. In this paper, we present a first step toward "Multi-agent LLM training" (MALT) on reasoning problems. Our approach employs a sequential multi-agent setup with heterogeneous LLMs assigned specialized roles: a generator, verifier, and refinement model iteratively solving problems. We propose a trajectory-expansion-based synthetic data generation process and a credit assignment strategy driven by joint outcome based rewards. This enables our post-training setup to utilize both positive and negative trajectories to autonomously improve each model's specialized capabilities as part of a joint sequential system. We evaluate our approach across MATH, GSM8k, and CQA, where MALT on Llama 3.1 8B models achieves relative improvements of 14.14%, 7.12%, and 9.40% respectively over the same baseline model. This demonstrates an early advance in multi-agent cooperative capabilities for performance on mathematical and common sense reasoning questions. More generally, our work provides a concrete direction for research around multi-agent LLM training approaches.

In many security applications of cyber-physical systems, a system designer must guarantee that critical missions are satisfied against attacks in the sensors and actuators of the CPS. Traditional security design of CPSs often assume that attackers have complete knowledge of the system. In this article, we introduce a class of deception techniques and study how to leverage asymmetric information created by deception to strengthen CPS security. Consider an adversarial interaction between a CPS defender and an attacker, who can perform sensor jamming attacks. To mitigate such attacks, the defender introduces asymmetrical information by deploying a "hidden sensor," whose presence is initially undisclosed but can be revealed if queried. We introduce hypergames on graphs to model this game with asymmetric information. Building on the solution concept called subjective rationalizable strategies in hypergames, we identify two stages in the game: An initial game stage where the defender commits to a strategy perceived rationalizable by the attacker until he deviates from the equilibrium in the attacker's perceptual game; Upon the deviation, a delay-attack game stage starts where the defender plays against the attacker, who has a bounded delay in attacking the sensor being revealed. Based on backward induction, we develop an algorithm that determines, for any given state, if the defender can benefit from hiding a sensor and revealing it later. If the answer is affirmative, the algorithm outputs a sensor revealing strategy to determine when to reveal the sensor during dynamic interactions. We demonstrate the effectiveness of our deceptive strategies through two case studies related to CPS security applications.

We explore the potential for productive team-based collaboration between humans and Artificial Intelligence (AI) by presenting and conducting initial tests with a general framework that enables multiple human and AI agents to work together as peers. ChatCollab's novel architecture allows agents - human or AI - to join collaborations in any role, autonomously engage in tasks and communication within Slack, and remain agnostic to whether their collaborators are human or AI. Using software engineering as a case study, we find that our AI agents successfully identify their roles and responsibilities, coordinate with other agents, and await requested inputs or deliverables before proceeding. In relation to three prior multi-agent AI systems for software development, we find ChatCollab AI agents produce comparable or better software in an interactive game development task. We also propose an automated method for analyzing collaboration dynamics that effectively identifies behavioral characteristics of agents with distinct roles, allowing us to quantitatively compare collaboration dynamics in a range of experimental conditions. For example, in comparing ChatCollab AI agents, we find that an AI CEO agent generally provides suggestions 2-4 times more often than an AI product manager or AI developer, suggesting agents within ChatCollab can meaningfully adopt differentiated collaborative roles. Our code and data can be found at: https://github.com/ChatCollab.

Multi Agent Path Finding (MAPF) is critical for coordinating multiple robots in shared environments, yet robust execution of generated plans remains challenging due to operational uncertainties. The Action Dependency Graph (ADG) framework offers a way to ensure correct action execution by establishing precedence-based dependencies between wait and move actions retrieved from a MAPF planning result. The original construction algorithm is not only inefficient, with a quadratic worst-case time complexity it also results in a network with many redundant dependencies between actions. This paper introduces two key improvements to the ADG framework. First, we prove that wait actions are generally redundant and show that removing them can lead to faster overall plan execution on real robot systems. Second, we propose an optimized ADG construction algorithm, termed Sparse Candidate Partitioning (SCP), which skips unnecessary dependencies and lowers the time complexity to quasi-linear, thereby significantly improving construction speed.

In many settings, machine learning models may be used to inform decisions that impact individuals or entities who interact with the model. Such entities, or agents, may game model decisions by manipulating their inputs to the model to obtain better outcomes and maximize some utility. We consider a multi-agent setting where the goal is to identify the "worst offenders:" agents that are gaming most aggressively. However, identifying such agents is difficult without knowledge of their utility function. Thus, we introduce a framework in which each agent's tendency to game is parameterized via a scalar. We show that this gaming parameter is only partially identifiable. By recasting the problem as a causal effect estimation problem where different agents represent different "treatments," we prove that a ranking of all agents by their gaming parameters is identifiable. We present empirical results in a synthetic data study validating the usage of causal effect estimation for gaming detection and show in a case study of diagnosis coding behavior in the U.S. that our approach highlights features associated with gaming.

Higher-dimensional sliding puzzles are constructed on the vertices of a $d$-dimensional hypercube, where $2^d-l$ vertices are distinctly coloured. Rings with the same colours are initially set randomly on the vertices of the hypercube. The goal of the puzzle is to move each of the $2^d-l$ rings to pre-defined target vertices on the cube. In this setting, the $k$-rule constraint represents a generalisation of edge collision for the movement of colours between vertices, allowing movement only when a hypercube face of dimension $k$ containing a ring is completely free of other rings. Starting from an initial configuration, what is the minimum number of moves needed to make ring colours match the vertex colours? An algorithm that provides us with such a number is called God's algorithm. When such an algorithm exists, it does not have a polynomial time complexity, at least in the case of the 15-puzzle corresponding to $k=1$ in the cubical puzzle. This paper presents a comprehensive computational study of different scenarios of the higher-dimensional puzzle. A benchmark of three computational techniques, an exact algorithm (the A* search) and two approximately optimal search techniques (an evolutionary algorithm (EA) and reinforcement learning (RL)) is presented in this work. The experiments show that all three methods can successfully solve the puzzle of dimension three for different face dimensions and across various difficulty levels. When the dimension increases, the A* search fails, and RL and EA methods can still provide a generally acceptable solution, i.e. a distribution of a number of moves with a median value of less than $30$. Overall, the EA method consistently requires less computational time, while failing in most cases to minimise the number of moves for the puzzle dimensions $d=4$ and $d=5$.

This paper investigates dynamic task allocation for multi-agent systems (MASs) under resource constraints, with a focus on maximizing the global utility of agents while ensuring a conflict-free allocation of targets. We present a more adaptable submodular maximization framework for the MAS task allocation under resource constraints. Our proposed distributed greedy bundles algorithm (DGBA) is specifically designed to address communication limitations in MASs and provides rigorous approximation guarantees for submodular maximization under $q$-independent systems, with low computational complexity. Specifically, DGBA can generate a feasible task allocation policy within polynomial time complexity, significantly reducing space complexity compared to existing methods. To demonstrate practical viability of our approach, we apply DGBA to the scenario of active observation information acquisition within a micro-satellite constellation, transforming the NP-hard task allocation problem into a tractable submodular maximization problem under a $q$-independent system constraint. Our method not only provides a specific performance bound but also surpasses benchmark algorithms in metrics such as utility, cost, communication time, and running time.