AI AGENTS ASSIST MODEL DEVELOPMENT, HUMANS RETAIN CONTROL

Researchers examined their own AI model development process and found a nuanced partnership between human and machine work. AI agents contributed significant labor—proposing methods and attempting tasks—but decision-making remained firmly in human hands. Key findings include: - Method proposals: AI agents supplied up to 55% of proposed approaches - Final decisions: Humans made over 85% of all choices - Task initiation: One-third of attempted tasks wouldn't have been started without AI suggestions The research challenges the narrative that increased AI activity automatically translates to greater autonomy. Instead, the data suggests AI systems function as capable assistants that augment human judgment rather than replace it. The authors emphasize that volume of AI contributions doesn't indicate decision-making authority. As AI tools become more capable, maintaining clear human oversight in critical development stages remains essential.