If you find value in content like this, sign up for my Talent Edge Weekly newsletter.
This article summarizes findings from a recent Nature Human Behaviour paper on AI-human collaboration. Researchers analyzed 106 experimental studies on human-AI collaboration conducted between January 2020 and July 2023. A key finding: on average, AI-human combinations do not outperform the best AI-only or human-only systems—contrary to expectations. For example, one finding shows that AI alone had the highest accuracy in detecting fake hotel reviews (73%), compared to 69% for AI-human teams and 55% for humans. AI by itself also outperforms in demand forecasting and medical diagnosis. AI and human worker combinations are more likely to provide an advantage on tasks in which humans already outperform AI on their own, tasks involving creating content, and creation tasks involving generative AI. The detailed research paper and summary article provide additional examples. These insights highlight the need for strategic decisions on where AI adds the most value—and where it does not. With this in mind, I’m resharing a Deloitte AI Institute report that offers a framework for determining which tasks are best suited for AI, AI-assisted humans, or humans alone.
