AI Isn't Outthinking Mathematicians. It's Out-Remembering Them

When AI systems demonstrate remarkable proficiency in solving difficult mathematical problems, the common assumption points to advanced intelligence or intuition. However, this article presents a compelling alternative: AI's success is largely attributable to its access to an enormous symbolic working memory, or context window, that dwarfs human cognitive capacity.

• Human working memory is severely limited, impacting our ability to hold and manipulate multiple unfamiliar elements simultaneously, particularly in complex mathematical tasks.
• AI's context window functions like a massive external notebook, allowing it to retain entire problem statements, hundreds of intermediate equations, previous attempts, and definitions, overcoming the human 'bottleneck'.
• This externalization of reasoning is particularly effective in mathematics, where elements can be translated into explicit, stable, and unambiguous symbols.
• The article differentiates this passive, externalized AI memory from active human internal working memory, noting that AI's apparent 'thinking out loud' becomes an advantage in symbolic domains.
• The hypothesis predicts AI's greatest advantage in problems requiring extensive bookkeeping, long calculations, and many interacting constraints, rather than simple conceptual leaps.
• Ultimately, the author suggests a fairer comparison is AI with its tools against a human equipped with equally powerful external memory and verification systems.

The article concludes by comparing AI's current capabilities to a 'machine-amplified' John von Neumann—possessing immense speed and breadth of memory—rather than an Albert Einstein, who could fundamentally reconceptualize problems. It posits that AI excels at executing and exploring consequences within existing frameworks, driven by its superior memory, and highlights that the next true leap will be its ability to invent new ways of understanding reality.