AI decodes Gödel's notebooks, finds hundreds of theorems
Original: AI unlocks Gödel's notebooks, said to contain hundreds of proved theorems
Why This Matters
Shows AI can unlock archival scientific material inaccessible due to language and format barriers.
Claude AI processed 1,009 pages of Kurt Gödel's handwritten 1940s notebooks in days, revealing hundreds of previously unknown proved theorems, including anticipations of results not published until the 1970s–80s.
Finnish logician Jan von Plato had been leading a Helsinki University project since 2017 to decode 16 notebooks Gödel wrote in the early 1940s. The obstacle: 1,009 pages of dense mathematical text written largely in Gabelsberger, a German shorthand already obsolete in Gödel's lifetime. Progress crawled at roughly one page per week. In September, von Plato brought in Anthropic's Claude. Within two days, the model had produced a comprehensive digest of all the material. The findings stunned researchers. Gödel had proved hundreds of theorems that were entirely unknown to the mathematical community. Among them: at least 14 original logical themes never subsequently pursued by anyone, and dozens of results that pre-empt discoveries not made publicly until the 1970s and 80s. "I was flabbergasted," von Plato said. "Gödel was far ahead of his time." Gödel's incompleteness theorems, published in 1931, demonstrated fundamental limits of formal mathematical systems and directly inspired Alan Turing's foundational work in computation. The new notebook findings suggest his unpublished output may be as significant as his published record.