Video106experiments in the meta-analysis
Not better than the model? Get out of the loop
TL;DR In decision tasks, human and AI together did worse whenever the AI was better. Only step in where you beat the model.
Human plus AI should beat either alone, right? A Nature Human Behaviour meta-analysis of 106 experiments says: often not. On decision tasks, human–AI teams did worse than the better of the two.
The twist is who's better. When the human was better than the AI, combining helped. When the AI was better, adding a human made it worse. For creative work like writing and design the effect was positive, though not significant.
Try this
Before you review AI output, ask yourself honestly: am I better than the model at this specific task?
Sources: Vaccaro, Almaatouq & Malone (2024), Nature Human Behaviour.
