Why AI and humans play games so differently

When researchers substitute LLMs for human subjects in game-theory experiments, the results don't match—humans and AI make systematically different strategic moves. The paper frames this gap as bounded rationality: humans are constrained by cognitive limits, while LLMs retrieve answers from training data, bypassing those constraints entirely. The difference matters for anyone using AI agents in behavioral experiments or strategic simulations—you're not getting human-like reasoning, just pattern matching that mimics it.