Who checks the answers AI gives in math
An explainer that reads the debate over AI-produced math results against the history of computer proofs. The four color theorem in 1976 was proved with computer calculations no person could follow to the end, which caused a debate, and the Kepler conjecture left 12 referees only 99% sure after 4 years until a formal proof, checked line by line by computer, finished it in 2014. This release also drew both high praise and refusals, and reports followed that some results were withdrawn or corrected soon after publication. Two days after the release, a mathematician slightly improved one result and moved it into a formal proof language, showing a division of labor in which machines give answers, machines and people share the checking, and understanding is still left to people. The conclusion is to ask not only what was solved but also who checked it and who understood it.
Key points
- The same question runs through the four color theorem and the Kepler conjecture
- Reports of some withdrawals and corrections soon after release
- Two days later a person improved a result and formalized it
- Record who checked and who understood, too
This is a note: my own write-up after reading several sources. It has no source link; events checked against a source are kept as records.