Automata-conditioned reinforcement learning (RL) has given promising results for learning multi-task policies capable of performing temporally extended objectives given at runtime, done by pretraining and freezing automata embeddings prior to training the downstream policy. However, no theoretical guarantees were given. This work provides a theoretical framework for the automata-conditioned RL problem and shows that it is probably approximately correct learnable. We then present a technique for learning provably correct automata embeddings, guaranteeing optimal multi-task policy learning. Our experimental evaluation confirms these theoretical results.

The rapid development of advanced AI agents and the imminent deployment of many instances of these agents will give rise to multi-agent systems of unprecedented complexity. These systems pose novel and under-explored risks. In this report, we provide a structured taxonomy of these risks by identifying three key failure modes (miscoordination, conflict, and collusion) based on agents' incentives, as well as seven key risk factors (information asymmetries, network effects, selection pressures, destabilising dynamics, commitment problems, emergent agency, and multi-agent security) that can underpin them. We highlight several important instances of each risk, as well as promising directions to help mitigate them. By anchoring our analysis in a range of real-world examples and experimental evidence, we illustrate the distinct challenges posed by multi-agent systems and their implications for the safety, governance, and ethics of advanced AI.

One approach for improving sample efficiency in cooperative multi-agent learning is to decompose overall tasks into sub-tasks that can be assigned to individual agents. We study this problem in the context of reward machines: symbolic tasks that can be formally decomposed into sub-tasks. In order to handle settings without a priori knowledge of the environment, we introduce a framework that can learn the optimal decomposition from model-free interactions with the environment. Our method uses a task-conditioned architecture to simultaneously learn an optimal decomposition and the corresponding agents' policies for each sub-task. In doing so, we remove the need for a human to manually design the optimal decomposition while maintaining the sample-efficiency benefits of improved credit assignment. We provide experimental results in several deep reinforcement learning settings, demonstrating the efficacy of our approach. Our results indicate that our approach succeeds even in environments with codependent agent dynamics, enabling synchronous multi-agent learning not achievable in previous works.

Goal-conditioned reinforcement learning is a powerful way to control an AI agent's behavior at runtime. That said, popular goal representations, e.g., target states or natural language, are either limited to Markovian tasks or rely on ambiguous task semantics. We propose representing temporal goals using compositions of deterministic finite automata (cDFAs) and use cDFAs to guide RL agents. cDFAs balance the need for formal temporal semantics with ease of interpretation: if one can understand a flow chart, one can understand a cDFA. On the other hand, cDFAs form a countably infinite concept class with Boolean semantics, and subtle changes to the automaton can result in very different tasks, making them difficult to condition agent behavior on. To address this, we observe that all paths through a DFA correspond to a series of reach-avoid tasks and propose pre-training graph neural network embeddings on "reach-avoid derived" DFAs. Through empirical evaluation, we demonstrate that the proposed pre-training method enables zero-shot generalization to various cDFA task classes and accelerated policy specialization without the myopic suboptimality of hierarchical methods.

The growing capabilities and increasingly widespread deployment of AI systems necessitate robust benchmarks for measuring their cooperative capabilities. Unfortunately, most multi-agent benchmarks are either zero-sum or purely cooperative, providing limited opportunities for such measurements. We introduce a general-sum variant of the zero-sum board game Diplomacy -- called Welfare Diplomacy -- in which players must balance investing in military conquest and domestic welfare. We argue that Welfare Diplomacy facilitates both a clearer assessment of and stronger training incentives for cooperative capabilities. Our contributions are: (1) proposing the Welfare Diplomacy rules and implementing them via an open-source Diplomacy engine; (2) constructing baseline agents using zero-shot prompted language models; and (3) conducting experiments where we find that baselines using state-of-the-art models attain high social welfare but are exploitable. Our work aims to promote societal safety by aiding researchers in developing and assessing multi-agent AI systems. Code to evaluate Welfare Diplomacy and reproduce our experiments is available at https://github.com/mukobi/welfare-diplomacy.

If much of the information is irrelevant, it's easy to To optimally coordinate with others in cooperative imagine how this could lead to significant increases in efficiency games, it is often crucial to have information for finding optimal policies. For example, this could about one's collaborators: successful driving requires allow a focused effort on few-shot or zero-shot adaptation to understanding which side of the road to co-players (Zand et al., 2022; Albrecht & Stone, 2017; Stone drive on. However, not every feature of collaborators et al., 2010; Hu et al., 2020) or more efficient DecPOMDP is strategically relevant: the fine-grained planning algorithms (Szer & Charpillet, 2006; Seuken & acceleration of drivers may be ignored while maintaining Zilberstein, 2007). In order to leverage these benefits, we optimal coordination. We show that there build the theory, data structures, and algorithms required to is a well-defined dichotomy between strategically distinguish between relevant and irrelevant information.