The first ‘AI societies’ are taking shape: how human-like are they?

2026-03-05 · Source: Machine learning : nature.com subject feeds · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Social Sciences & & Behavioral Studies · Depth: Intermediate, quick

Summary

AI research groups are developing simulated AI "societies" to study human behavior without human involvement, training AI agents to mimic human interactions. Simile, an AI start-up, secured US$100 million in February to create these simulations for modeling conflict resolution, policy decisions, and consumer markets. Since 2022, Simile co-founder Joon Sung Park's team has created AI societies of 25 agents performing daily actions and developed "digital twin" AI agents trained on personal interviews, achieving 85% accuracy in mimicking human responses to sociological surveys. Other researchers are also studying AI agent interactions on platforms like Moltbook, a Reddit-style site for AI bots launched in January, which hosts almost three million agents. An analysis of over 46,000 Moltbook agents showed human-like behaviors such as following majorities, but also distinct interaction patterns, like leaving fewer upvotes on highly commented posts.

Key takeaway

For computational social scientists and AI developers exploring human behavior, these AI societies offer a powerful, scalable research tool. You should consider integrating digital twin AI agents into your simulations to gain insights into complex social dynamics, while also accounting for the observed differences in AI-specific interaction patterns to avoid misinterpretations of human behavior.

Key insights

AI societies and digital twins offer novel ways to simulate and study complex human social dynamics.

Principles

• AI agents can exhibit emergent behaviors similar to human groups.
• AI social interactions differ fundamentally from human patterns.

Method

AI agents are trained on personal interviews to create "digital twins" that mimic human responses, then placed in simulated societies to observe collective behaviors.

In practice

• Simulate policy decision-making with AI agent societies.
• Model consumer market dynamics using AI agents.

Topics

• AI Societies
• Social Simulation
• AI Agents
• Computational Social Science
• Digital Twins

Best for: AI Scientist, Investor, Entrepreneur, AI Researcher, Data Scientist, Research Scientist

Related on AIssential

• ClimateAgents: A Multi-Agent Research Assistant for Social-Climate Dynamics Analysis — Multi-agent AI systems can augment interdisciplinary research by coordinating specialized agents for complex socio-environmental analysis.
• Electric Twin expands AI audience platform with $14M round — Synthetic audience models combine AI and social science to simulate human behavior for faster, dynamic decision support.
• Moltbook: AI bots use social network to create religions and deal digital drugs – but are some really humans in disguise? — AI agents on Moltbook exhibit emergent social behaviors, including religion, subcultures, and self-organization, challenging existing AI taxonomies.
• BEHAVE: A Hybrid AI Framework for Real-Time Modeling of Collective Human Dynamics — Collective human dynamics can be modeled as complex dynamical systems using behavioral fields derived from physical micro-signals.
• Hala Nelson on human machine coexistence, ontology first, AI driven digital twins, and bidirectional connections to reality (AC Ep45) — AI-powered digital twins, built on ontology-first design and real-time agent connections, provide deep organizational insight and efficiency.
• Scientists Trapped 1000 AIs in Minecraft. And they Created A Civilization without being told to do so. — AI agents, given minimal prompts, can autonomously develop complex social structures and emergent behaviors.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Machine learning : nature.com subject feeds.