The Darwin Gödel Machine (DGM) is a self-improving AI system introduced in 2025 by researchers at the Japanese AI company Sakana AI and the University of British Columbia in Canada. A coding agent edits its own source code, and each modified version is evaluated on coding benchmarks to verify whether it performs better. The system keeps an archive of the agents it has produced and selects from it to create further variants, exploring many lineages in parallel.
The name combines the Gödel machine, a theoretical self-improving AI proposed by Jürgen Schmidhuber, with Darwinian evolution. Whereas a Gödel machine changes its code only after mathematically proving that the change is beneficial, the Darwin Gödel Machine replaces proof with empirical validation.