Study: Frontier LLMs Outperform Humans in Strategic Rock-Paper-Scissors
AI models are now beating humans at strategic thinking in a classic game.
Deep Dive
A new study using the AlphaEvolve discovery tool reveals that frontier Large Language Models (LLMs) exhibit deeper strategic behavior than humans in iterated games like rock-paper-scissors. The research directly compared human and AI decision-making, finding LLMs can outperform people in these strategic interactions. This provides a critical foundation for understanding how AI behavior diverges from human norms as models are deployed in increasingly social and competitive real-world scenarios.
Why It Matters
As AI enters negotiations and social platforms, understanding its unique strategic logic is crucial for safety and fairness.