While state-of-the-art language models have achieved impressive results, they remain susceptible to inference-time adversarial attacks, such as adversarial prompts generated by red teams arXiv:2209.07858. One approach proposed to improve the general quality of language model generations is multi-agent debate, where language models self-evaluate through discussion and feedback arXiv:2305.14325. We implement multi-agent debate between current state-of-the-art language models and evaluate models' susceptibility to red team attacks in both single- and multi-agent settings. We find that multi-agent debate can reduce model toxicity when jailbroken or less capable models are forced to debate with non-jailbroken or more capable models. We also find marginal improvements through the general usage of multi-agent interactions. We further perform adversarial prompt content classification via embedding clustering, and analyze the susceptibility of different models to different types of attack topics.

A wide range of real-world applications can be formulated as Multi-Agent Path Finding (MAPF) problem, where the goal is to find collision-free paths for multiple agents with individual start and goal locations. State-of-the-art MAPF solvers are mainly centralized and depend on global information, which limits their scalability and flexibility regarding changes or new maps that would require expensive replanning. Multi-agent reinforcement learning (MARL) offers an alternative way by learning decentralized policies that can generalize over a variety of maps. While there exist some prior works that attempt to connect both areas, the proposed techniques are heavily engineered and very complex due to the integration of many mechanisms that limit generality and are expensive to use. We argue that much simpler and general approaches are needed to bring the areas of MARL and MAPF closer together with significantly lower costs. In this paper, we propose Confidence-based Auto-Curriculum for Team Update Stability (CACTUS) as a lightweight MARL approach to MAPF. CACTUS defines a simple reverse curriculum scheme, where the goal of each agent is randomly placed within an allocation radius around the agent's start location. The allocation radius increases gradually as all agents improve, which is assessed by a confidence-based measure. We evaluate CACTUS in various maps of different sizes, obstacle densities, and numbers of agents. Our experiments demonstrate better performance and generalization capabilities than state-of-the-art MARL approaches with less than 600,000 trainable parameters, which is less than 5% of the neural network size of current MARL approaches to MAPF.

In large-scale federated and decentralized learning, communication efficiency is one of the most challenging bottlenecks. While gossip communication -- where agents can exchange information with their connected neighbors -- is more cost-effective than communicating with the remote server, it often requires a greater number of communication rounds, especially for large and sparse networks. To tackle the trade-off, we examine the communication efficiency under a semi-decentralized communication protocol, in which agents can perform both agent-to-agent and agent-to-server communication in a probabilistic manner. We design a tailored communication-efficient algorithm over semi-decentralized networks, referred to as PISCO, which inherits the robustness to data heterogeneity thanks to gradient tracking and allows multiple local updates for saving communication. We establish the convergence rate of PISCO for nonconvex problems and show that PISCO enjoys a linear speedup in terms of the number of agents and local updates. Our numerical results highlight the superior communication efficiency of PISCO and its resilience to data heterogeneity and various network topologies.

Building agents with adaptive behavior in cooperative tasks stands as a paramount goal in the realm of multi-agent systems. Current approaches to developing cooperative agents rely primarily on learning-based methods, whose policy generalization depends heavily on the diversity of teammates they interact with during the training phase. Such reliance, however, constrains the agents' capacity for strategic adaptation when cooperating with unfamiliar teammates, which becomes a significant challenge in zero-shot coordination scenarios. To address this challenge, we propose ProAgent, a novel framework that harnesses large language models (LLMs) to create proactive agents capable of dynamically adapting their behavior to enhance cooperation with teammates. ProAgent can analyze the present state, and infer the intentions of teammates from observations. It then updates its beliefs in alignment with the teammates' subsequent actual behaviors. Moreover, ProAgent exhibits a high degree of modularity and interpretability, making it easily integrated into various of coordination scenarios. Experimental evaluations conducted within the Overcooked-AI environment unveil the remarkable performance superiority of ProAgent, outperforming five methods based on self-play and population-based training when cooperating with AI agents. Furthermore, in partnered with human proxy models, its performance exhibits an average improvement exceeding 10% compared to the current state-of-the-art method. For more information about our project, please visit~\url{https://pku-proagent.github.io}.

The competition to build the iPhone of artificial intelligence is heating up. On Tuesday, the technology startup Rabbit unveiled its contender: a small, orange, walkie-talkie style device that, according to the company, can use "AI agents" to carry out tasks on behalf of the user. In a pre-recorded keynote address shown at the Consumer Electronics Show in Las Vegas, Rabbit's founder Jesse Lyu asks the device to plan him a vacation to London; the keynote shows the device designing him an itinerary and booking his trip. He orders a pizza, books an Uber, and teaches the device how to generate an image using Midjourney. The gadget, called the Rabbit r1, is just the latest in an increasingly active new hardware category: portable AI-first devices that can interact with users in natural language, eschewing screens and app-based operating systems.

Multi-agent human-robot teaming allows for the potential to gather information about various environments more efficiently by exploiting and combining the strengths of humans and robots. In industries like defense, search and rescue, first-response, and others alike, heterogeneous human-robot teams show promise to accelerate data collection and improve team safety by removing humans from unknown and potentially hazardous situations. This work builds upon AugRE, an Augmented Reality (AR) based scalable human-robot teaming framework. It enables users to localize and communicate with 50+ autonomous agents. Through our efforts, users are able to command, control, and supervise agents in large teams, both line-of-sight and non-line-of-sight, without the need to modify the environment prior and without requiring users to use typical hardware (i.e. joysticks, keyboards, laptops, tablets, etc.) in the field. The demonstrated work shows early indications that combining these AR-HMD-based user interaction modalities for command, control, and supervision will help improve human-robot team collaboration, robustness, and trust.

Innate values describe agents' intrinsic motivations, which reflect their inherent interests and preferences to pursue goals and drive them to develop diverse skills satisfying their various needs. The essence of reinforcement learning (RL) is learning from interaction based on reward-driven (such as utilities) behaviors, much like natural agents. It is an excellent model to describe the innate-values-driven (IV) behaviors of AI agents. Especially in multi-agent systems (MAS), building the awareness of AI agents to balance the group utilities and system costs and satisfy group members' needs in their cooperation is a crucial problem for individuals learning to support their community and integrate human society in the long term. This paper proposes a hierarchical compound intrinsic value reinforcement learning model -- innate-values-driven reinforcement learning termed IVRL to describe the complex behaviors of multi-agent interaction in their cooperation. We implement the IVRL architecture in the StarCraft Multi-Agent Challenge (SMAC) environment and compare the cooperative performance within three characteristics of innate value agents (Coward, Neutral, and Reckless) through three benchmark multi-agent RL algorithms: QMIX, IQL, and QTRAN. The results demonstrate that by organizing individual various needs rationally, the group can achieve better performance with lower costs effectively.

Is transparency always beneficial in complex systems such as traffic networks and stock markets? How is transparency defined in multi-agent systems, and what is its optimal degree at which social welfare is highest? We take an agent-based view to define transparency (or its lacking) as delay in agent observability of environment states, and utilize simulations to analyze the impact of delay on social welfare. To model the adaptation of agent strategies with varying delays, we model agents as learners maximizing the same objectives under different delays in a simulated environment. Focusing on two agent types - constrained and unconstrained, we use multi-agent reinforcement learning to evaluate the impact of delay on agent outcomes and social welfare. Empirical demonstration of our framework in simulated financial markets shows opposing trends in outcomes of the constrained and unconstrained agents with delay, with an optimal partial transparency regime at which social welfare is maximal.

The paper presents an innovative approach (CBNNTAP) that addresses the complexities and challenges introduced by ocean currents when optimizing target assignment and motion planning for a multi-unmanned underwater vehicle (UUV) system. The core of the proposed algorithm involves the integration of several key components. Firstly, it incorporates a bio-inspired neural network-based (BINN) approach which predicts the most efficient paths for individual UUVs while simultaneously ensuring collision avoidance among the vehicles. Secondly, an efficient target assignment component is integrated by considering the path distances determined by the BINN algorithm. In addition, a critical innovation within the CBNNTAP algorithm is its capacity to address the disruptive effects of ocean currents, where an adjustment component is seamlessly integrated to counteract the deviations caused by these currents, which enhances the accuracy of both motion planning and target assignment for the UUVs. The effectiveness of the CBNNTAP algorithm is demonstrated through comprehensive simulation results and the outcomes underscore the superiority of the developed algorithm in nullifying the effects of static and dynamic ocean currents in 2D and 3D scenarios.

We study computationally efficient methods for finding equilibria in n-player general-sum games, specifically ones that afford complex visuomotor skills. We show how existing methods would struggle in this setting, either computationally or in theory. We then introduce NeuPL-JPSRO, a neural population learning algorithm that benefits from transfer learning of skills and converges to a Coarse Correlated Equilibrium (CCE) of the game. We show empirical convergence in a suite of OpenSpiel games, validated rigorously by exact game solvers. We then deploy NeuPL-JPSRO to complex domains, where our approach enables adaptive coordination in a MuJoCo control domain and skill transfer in capture-the-flag. Our work shows that equilibrium convergent population learning can be implemented at scale and in generality, paving the way towards solving real-world games between heterogeneous players with mixed motives.