LessWrong maps character vs corrigibility debate

The author examines the ongoing debate between character-based and corrigibility-based approaches to AI alignment, mapping six distinct accounts and ten core cruxes. The survey outlines competing arguments regarding instrumental convergence, safety risks, and the feasibility of defining a single alignment target. Ultimately, the analysis highlights the philosophical and practical divergences among frontier AI labs. Open article →