Mapping "character vs. corrigibility": six positions and ten cruxes

The author examines the debate between character and corrigibility as alignment targets for artificial intelligence, mapping out six distinct accounts ranging from tool AI to cooperative alignment. The post details leading arguments from both camps, areas of broad agreement, and ten key cruxes that divide proponents of each approach. Ultimately, the author highlights how major labs like Anthropic and OpenAI reflect different philosophical balances between obedience and virtue. Open article →