2020/01/25 by Raphael Köster, Raphaël Koster, Köster, Raphael +6 · 1 voice
Computer Science · Social Sciences · #Artificial Intelligence (cs.AI) #Crime Patterns and Interventions #Crime, Illicit Activities, and Governance #Cybercrime and Law Enforcement Studies #FOS: Computer and information sciences #Multiagent Systems (cs.MA) #cs.AI #cs.MA
paper · pdf · doi:10.48550/arxiv.2001.09318
openalex publication_date 2020/01/25 · arxiv published 2020/01/25 · arxiv updated 2020/01/25 · openalex created_date 2020/07/02 · openalex updated_date 2026/07/28
How can societies learn to enforce and comply with social norms? Here we\ninvestigate the learning dynamics and emergence of compliance and enforcement\nof social norms in a foraging game, implemented in a multi-agent reinforcement\nlearning setting. In this spatiotemporally extended game, individuals are\nincentivized to implement complex berry-foraging policies and punish\ntransgressions against social taboos covering specific berry types. We show\nthat agents benefit when eating poisonous berries is taboo, meaning the\nbehavior is punished by other agents, as this helps overcome a\ncredit-assignment problem in discovering delayed health effects. Critically,\nhowever, we also show that introducing an additional taboo, which results in\npunishment for eating a harmless berry, improves the rate and stability with\nwhich agents learn to punish taboo violations and comply with taboos.\nCounterintuitively, our results show that an arbitrary taboo (a "silly rule")\ncan enhance social learning dynamics and achieve better outcomes in the middle\nstages of learning. We discuss the results in the context of studying\nnormativity as a group-level emergent phenomenon.\n