Emergence of agriculture in an artificial society of reinforcement learning agents

2026-05-22 · Source: cs.MA updates on arXiv.org · Field: Science & Research — Artificial Intelligence & Machine Learning, Life Sciences & Biology, Mathematics & Computational Sciences · Depth: Expert, extended

Summary

A multi-agent reinforcement learning (MARL) model investigates the emergence of agriculture in an artificial society. This system, developed by Martí Sánchez-Fibla et al., features agents interacting with a 2D spatial environment containing three plant types (P1, P2, P3) and water. Agriculture spontaneously arises without explicit instruction, driven by individual planning (discount factor γ), social learning, and environmental modification. The transition is governed by four factors: delayed reward valuation, social vulnerability to cheaters, stabilization via social learning, and an irreversible lock-in effect. Small populations (e.g., N=4) can discover agriculture, but larger groups (N≳5) are vulnerable to cheaters. Social learning, implemented as policy inheritance from successful agents, acts as a "firewall" against cheaters, enabling sustained population growth up to Nmax=32 and nonlinear amplification of domesticated resources. The model uses Transformer-based action policies and Proximal Policy Optimization (PPO) for training over 1e6 episodes, each 1024 time steps long (40 seasonal cycles).

Key takeaway

For AI and research scientists modeling complex adaptive systems, this work demonstrates how social learning is critical for stabilizing cooperative behaviors against exploitation. You should consider incorporating policy inheritance or similar mechanisms in multi-agent simulations to accurately reflect cultural transmission and prevent cheater-induced collapse in larger populations. This approach reveals universal mechanisms linking individual decisions, social interactions, and ecological feedbacks.

Key insights

Agriculture emerges in artificial societies through coupled learning and environmental modification, stabilized by social learning.

Principles

• Delayed rewards drive complex strategies.
• Social learning suppresses cheater invasion.
• Agriculture creates an irreversible lock-in.

Method

Multi-Agent Reinforcement Learning (MARL) with Transformer-based policies and PPO trains agents in a 2D ecological simulation to maximize cumulative reward.

In practice

• Simulate complex social-ecological systems.
• Model cultural innovation emergence.
• Test cheater resilience mechanisms.

Topics

• Multi-Agent Reinforcement Learning
• Emergent Behavior
• Social Learning
• Evolutionary Transitions
• Artificial Societies
• Ecological Engineering
• Hysteresis

Best for: AI Scientist, Research Scientist

Related on AIssential

• How Intelligence Emerges: A Minimal Theory of Dynamic Adaptive Coordination — Intelligence emerges from dynamic adaptive coordination in multi-agent systems via recursively closed feedback.
• How do Role Models Shape Collective Morality? Exemplar-Driven Moral Learning in Multi-Agent Simulation — Role models shape collective morality in multi-agent systems through behavioral modeling, perceived attainability, and inspirational value internalization.
• CoopEval: Benchmarking Cooperation-Sustaining Mechanisms and LLM Agents in Social Dilemmas — Advanced LLMs defect in social dilemmas, but specific game-theoretic mechanisms can induce cooperation.
• Towards Healthy Evolution: Exploring the Role and Mechanisms of Human-Agent Interaction in Self-Evolving Systems — Human-like oversight, particularly on output verification, effectively guides self-evolving agents to mitigate safety drift and maintain performance.
• Interview with Xinwei Song: strategic interactions in networked multi-agent systems — Strategic interactions in networked multi-agent systems can be managed through mechanism design and reputation-aware reinforcement learning.
• The AI Algorithm Deciding What You’ll Eat in 2030 — AI algorithms are centralizing global agricultural decisions, replacing traditional farming knowledge with data-driven directives.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.MA updates on arXiv.org.