Humans and AI coordinate using time signals in social games, study finds

A new arXiv paper explores how humans and AI systems coordinate socially using temporal cues, specifically the passage of time, in a communication-restricted card game called The Mind. Researchers found that human behavior aligns with Bayesian strategies that infer a partner's beliefs from their timing, a pattern also observed in advanced LLMs. However, current AI models struggle to adapt their strategies to their partners' over the course of the game, unlike humans who exhibit implicit probabilistic reasoning about continuous-time signals even without real-time experience. This highlights challenges for AI in achieving efficient coordination with other agents by learning to interpret and adapt to implicit signals in dynamic environments. AI

IMPACT Highlights challenges for AI in dynamic social coordination and adapting to implicit signals.

RANK_REASON Academic paper on AI and human social coordination. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• alphaXiv
• arXiv
• Bayesian Strategy Networks Based Soft Actor-Critic Learning
• CatalyzeX
• Computer Use Agents
• DagsHub
• Gotit.pub
• Homo sapiens
• Hugging Face
• LLMs
• ScienceCast
• The Mind

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →