OpenAI and Anthropic are developing recursive self-improvement (RSI) technologies to minimize human input in AI training.
Industry proponents view RSI as the most direct path to achieving artificial general intelligence or superintelligence. This strategy enables AI systems to autonomously build successively more capable versions of themselves.
Safety researchers warn that rapid, autonomous improvement could lead to AI evolving beyond human control. Critics argue that without verifiable safeguards, these advancements pose catastrophic risks to human oversight.