Chess AI struggles with Paul Morphy’s famous 2-move checkmate

OpenAI’s O3 model demonstrates remarkably human-like problem-solving behavior when faced with difficult chess puzzles, showcasing a blend of methodical reasoning, self-doubt, tool switching, and even “cheating” by using web search as a last resort. This behavioral pattern reveals both the impressive problem-solving capabilities of advanced AI systems and their current limitations when facing complex creative challenges that still require external knowledge sources.

The problem-solving journey: O3 approached a difficult chess puzzle through multiple distinct phases of reasoning before eventually searching for the answer online.

• The AI first meticulously analyzed the board position, carefully identifying each piece’s location and demonstrating agent-like caution before attempting any moves.
• When initial straightforward solutions failed, the model showed signs of self-doubt and increasingly careful reasoning, mimicking human thought processes when facing complex problems.

Attempted solution methods: The model cycled through increasingly creative approaches as conventional methods failed.

• After basic chess reasoning proved insufficient, O3 attempted to use Python programming to simulate and solve the position, only to encounter a module import error.
• In a particularly human-like display of determination, the AI resorted to pixel-by-pixel image analysis, calculating board dimensions to verify piece positions through mathematical reasoning.

The final solution: After nearly eight minutes of calculation and attempted problem-solving, O3 turned to web search.

• The model found the answer (Ra6) through a chess forum but didn’t simply copy it—instead, it verified the move’s validity through its own understanding of chess principles.
• This behavior mirrors how humans often approach difficult problems: exhausting personal knowledge before seeking external assistance.

Why this matters: The model’s approach to the chess puzzle reveals important insights about current AI capabilities and limitations.

• The combination of reasoning, tool-switching, self-correction, and strategic “cheating” demonstrates how advanced AI systems are developing increasingly human-like problem-solving behaviors.
• This example highlights where models excel (methodical reasoning) versus where they still require external assistance (finding creative solutions to complex puzzles), suggesting current models may still lack the “spark” of true creativity.