2022/01/29 by Mingwei Ma, Jizhou Liu, Ma, Mingwei +7 · 2 citations
Computer Science · Psychology · #Action (physics) #Artificial Intelligence (cs.AI) #Artificial intelligence #Cognitive science #Computer science #Computer security #Dependency (UML) #Explainable Artificial Intelligence (XAI) #Exploit #FOS: Computer and information sciences #Human–computer interaction #Machine Learning (cs.LG) #Machine learning #Multiagent Systems (cs.MA) #Object (grammar) #Point (geometry) #Process (computing) #Psychology #Reinforcement Learning in Robotics #Representation (politics) #Task (project management) #Topic Modeling
paper · pdf · doi:10.48550/arxiv.2201.12658
openalex publication_date 2022/01/29 · openalex created_date 2025/10/10 · openalex updated_date 2026/08/05
An unaddressed challenge in multi-agent coordination is to enable AI agents to exploit the semantic relationships between the features of actions and the features of observations. Humans take advantage of these relationships in highly intuitive ways. For instance, in the absence of a shared language, we might point to the object we desire or hold up our fingers to indicate how many objects we want. To address this challenge, we investigate the effect of network architecture on the propensity of learning algorithms to exploit these semantic relationships. Across a procedurally generated coordination task, we find that attention-based architectures that jointly process a featurized representation of observations and actions have a better inductive bias for learning intuitive policies. Through fine-grained evaluation and scenario analysis, we show that the resulting policies are human-interpretable. Moreover, such agents coordinate with people without training on any human data.