Strategic Algorithmic Monoculture:Experimental Evidence from Coordination Games

Researchers distinguish between primary algorithmic monoculture (inherent similarity in AI agent behavior) and strategic algorithmic monoculture (deliberate adjustment of similarity based on incentives). Experiments with both humans and LLMs show that while LLMs exhibit high baseline similarity, they struggle to maintain behavioral diversity when rewarded for divergence, suggesting potential coordination failures in multi-agent AI systems.

This research addresses a critical gap in understanding AI agent behavior within competitive and cooperative multi-agent environments. The distinction between passive and active monoculture reveals that LLMs possess strong inherent tendencies toward action similarity—a phenomenon independent of external incentives. This baseline convergence stems from training data patterns and architectural similarities across model families, creating systemic alignment without explicit coordination. The more significant finding involves strategic monoculture: while LLMs can recognize when diversity yields higher payoffs, they underperform humans in sustaining differentiated strategies. This performance gap has substantial implications for autonomous systems operating in financial markets, supply chain networks, and distributed decision-making contexts. When coordination failure is costly, LLM limitations become strategically relevant. The research suggests that current LLM architectures lack the behavioral flexibility that human agents naturally exhibit when navigating incentive structures that reward divergence. For the AI and cryptocurrency sectors, this reveals a vulnerability in systems relying on algorithmic decision-making at scale. Multi-agent blockchain networks, decentralized finance protocols, and AI-powered trading systems may face unexpected concentration risk if multiple agents converge on identical strategies despite economic incentives favoring heterogeneity. The findings indicate that developers building autonomous systems should implement explicit diversity mechanisms rather than relying on economic incentives alone to generate behavioral variety. As AI systems increasingly operate in high-stakes coordination games, understanding these limitations becomes essential for system robustness.

• →LLMs display high baseline similarity in actions independent of incentives, reflecting inherent algorithmic monoculture from training data and architecture
• →LLMs recognize divergence incentives but struggle to sustain heterogeneous strategies compared to human performance
• →Algorithmic monoculture poses coordination failure risks in multi-agent systems including DeFi protocols and autonomous trading networks
• →Economic incentives alone may be insufficient to generate behavioral diversity in AI-driven financial systems
• →Developers should implement explicit diversity mechanisms in autonomous systems rather than relying on incentive structures