Collaboration requires agents to coordinate their behavior on the fly, sometimes cooperating to solve a single task together and other times dividing it up into sub-tasks to work on in parallel. Underlying the human ability to collaborate is theory-of-mind (ToM), the ability to infer the hidden mental states that drive others to act. Here, we develop Bayesian Delegation, a decentralized multiagent learning mechanism with these abilities. Bayesian Delegation enables agents to rapidly infer the hidden intentions of others by inverse planning. We test Bayesian Delegation in a suite of multiagent Markov decision processes inspired by cooking problems. On these tasks, agents with Bayesian Delegation coordinate both their high-level plans (e.g., what sub-task they should work on) and their low-level actions (e.g., avoiding getting in each other's way). When matched with partners that act using the same algorithm, Bayesian Delegation outperforms alternatives. Bayesian Delegation is also a capable ad hoc collaborator and successfully coordinates with other agent types even in the absence of prior experience. Finally, in a behavioral experiment, we show that Bayesian Delegation makes inferences similar to human observers about the intent of others. Together, these results argue for the centrality of ToM for successful decentralized multi-agent collaboration.

When asked to identify objects having unique shapes and colors among other objects, English speakers often produce redundant color modifiers (“the red circle”) while Spanish speakers produce them less often (“el circulo (rojo)”). This cross‐linguistic difference has been attributed to a difference in word order between the two languages, under the incremental efficiency hypothesis (Rubio‐Fernández, Mollica, & Jara‐Ettinger, 2020). However, previous studies leave open the possibility that broad language differences between English and Spanish may explain this cross‐linguistic difference such that English speakers may generally produce more modifiers than Spanish speakers, including redundant ones, irrespective of word order. Here, we test the incremental efficiency hypothesis in a language production task crossing language (English, Spanish) with modifier type (color, number). Critically, number words occur on the same side of the noun in both English and Spanish. If broad language differences are responsible for the higher rate of color word production in English compared to Spanish, then the same effect should hold for number words. In contrast, the incremental efficiency hypothesis predicts an interaction between language and modifier type, due to different ordering for color words but identical ordering for number words. Our pre‐registered analyses offer strong support for the incremental efficiency hypothesis, demonstrating how seemingly small differences in language can cause us to describe the world in surprisingly different ways.

Collaboration requires agents to coordinate their behavior on the fly, sometimes cooperating to solve a single task together and other times dividing it up into sub-tasks to work on in parallel. Here, we develop Bayesian Delegation, a decentralized multi-agent learning mechanism with these abilities. Bayesian Delegation enables agents to rapidly infer the hidden intentions of others by inverse planning. We test Bayesian Delegation in a suite of multi-agent Markov decision processes inspired by cooking problems. On these tasks, agents with Bayesian Delegation coordinate both their high-level plans (e.g. what sub-task they should work on) and their low-level actions (e.g. avoiding getting in each other’s way). In a self-play evaluation, Bayesian Delegation outperforms alternative algorithms. Bayesian Delegation is also a capable ad-hoc collaborator and successfully coordinates with other agent types even in the absence of prior experience.