Reproduce “Understanding Human Limits in Pattern Recognition: A Computational Model of Sequential Reasoning in Rock, Paper, Scissors” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Logan Cross, Erik Brockbank, Tobias Gerstenberg, Judith E. Fan, Daniel L. K. Yamins, Nick Haber
Cross et al. (2025) introduce Hypothetical Minds (HM), an LLM-based agent capable of modeling human sequential reasoning in Rock, Paper, Scissors, accurately reproducing human successes and failures in pattern recognition. The research demonstrates that hypothesis generation is a primary bottleneck in recognizing complex opponent strategies, and shows that verbal scaffolding can significantly improve the model's performance on these challenging tasks.
Key insights
- Large Language Models, when integrated into structured cognitive architectures like HM, can serve as accurate computational models of human cognitive processes, replicating nuanced patterns of success and failure.
- The primary bottleneck in recognizing moderately complex opponent strategies is hypothesis generation—the ability to discover the correct strategy—rather than the capacity to evaluate or execute a known strategy.
- Verbal scaffolding, particularly attention-directing prompts, can effectively redirect an agent's hypothesis search toward relevant patterns, offering a method to overcome hypothesis generation limitations.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.