This preregistered systematic review and meta-analysis of 106 experimental studies (370 effect sizes) investigated when human-AI combinations outperform humans or AI alone. The analysis revealed that, on average, human-AI combinations performed worse than the best of humans or AI alone. However, performance losses were more pronounced in decision-making tasks, while content creation tasks showed potential gains. The relative performance of humans and AI also significantly influenced the results; human-AI combinations outperformed both when humans initially outperformed AI, but underperformed when AI was initially superior. Limitations include potential publication bias and variations in study designs.

Michelle Vaccaro, Abdullah Almaatouq, Thomas Malone