{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/prosocial-learning-agents-solve-generalized","title":"Prosocial learning agents solve generalized Stag Hunts better than selfish ones","arxiv_id":"1709.02865","date":"2017-09-08","proceeding":null,"authors":["Alexander Peysakhovich","Adam Lerer"],"abstract":"Deep reinforcement learning has become an important paradigm for constructing\nagents that can enter complex multi-agent situations and improve their policies\nthrough experience. One commonly used technique is reactive training - applying\nstandard RL methods while treating other agents as a part of the learner's\nenvironment. It is known that in general-sum games reactive training can lead\ngroups of agents to converge to inefficient outcomes. We focus on one such\nclass of environments: Stag Hunt games. Here agents either choose a risky\ncooperative policy (which leads to high payoffs if both choose it but low\npayoffs to an agent who attempts it alone) or a safe one (which leads to a safe\npayoff no matter what). We ask how we can change the learning rule of a single\nagent to improve its outcomes in Stag Hunts that include other reactive\nlearners. We extend existing work on reward-shaping in multi-agent\nreinforcement learning and show that that making a single agent prosocial, that\nis, making them care about the rewards of their partners can increase the\nprobability that groups converge to good outcomes. Thus, even if we control a\nsingle agent in a group making that agent prosocial can increase our agent's\nlong-run payoff. We show experimentally that this result carries over to a\nvariety of more complex environments with Stag Hunt-like dynamics including\nones where agents must learn from raw input pixels.","url_abs":"http://arxiv.org/abs/1709.02865v2","url_pdf":"http://arxiv.org/pdf/1709.02865v2.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"prosocial-learning-agents-solve-generalized","repo_url":"https://github.com/sharedcare/MARL-Cooperative","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"unanswered"}}],"tasks":[{"task_slug":"deep-reinforcement-learning","task_name":"Deep Reinforcement Learning"},{"task_slug":"multi-agent-reinforcement-learning","task_name":"Multi-agent Reinforcement Learning"},{"task_slug":"reinforcement-learning","task_name":"Reinforcement Learning"},{"task_slug":"reinforcement-learning-1","task_name":"Reinforcement Learning (RL)"},{"task_slug":"reinforcement-learning-2","task_name":"reinforcement-learning"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1709.02865","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}