A new study published on arXiv investigates outcome skews in chess positions that are assessed as equal by strong engines like Stockfish 18. Researchers analyzed millions of games played on Lichess and found that even when engine evaluations are within 10 centipawns of zero and stable, human players exhibit consistent biases. These skews, favoring either White or Black, were found to be reproducible across different sets of players, time periods, and rating bands, suggesting they are inherent properties of the positions themselves rather than random chance. The study indicates that engine evaluations, while precise, are not sufficient to predict human outcomes in chess, as the disfavored side tends to spend more time contemplating moves in these positions. AI
IMPACT Highlights limitations of AI evaluation in complex human decision-making, suggesting AI is not yet a sufficient statistic for human outcomes.
RANK_REASON The cluster contains an academic paper detailing a novel research finding. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
