Cambridge, MA, June 12, 2025 (Globe Newswire) – Do AI doctors trust a diagnosis of skin cancer, even if they are more accurate than human doctors?
New research at MIT Sloan Management School sheds light on an inexplicable paradox. Despite the increased accuracy and efficiency of AI, people still prefer human decisions, even when AI performance improves. However, in other contexts, such as when predicting inventory trends, people rely more easily on AI than human experts. What explains this puzzle?
Research paper titled “”AI dislike or gratitude? Competence – Personalization Framework and Meta-Analytical Reviewhas been published Psychological breaking news. The author is Associate Professor Mitt Sloan of Jackson G. Lou. Professor Xin Qin, Associate Professor Chen Chen, PhD students Hansen Zhou and Xiaowei Dong, and postdoctoral researcher at Limei Cao at Sun Yat-Sen University. Xiang Zhou, a postdoctoral researcher at Shenzhen University; and Prince Higashi, associate professor at the University of Fudan.
The researchers conducted a meta-analysis of 163 studies, including over 80,000 participants. They proposed a new theory, A framework of competence and personalizationsuggesting that individuals focus on two important dimensions in determining whether they rely on AI vs. human in a particular decision context.
- AI's perceptual ability:Is AI perceived as more capable than humans in the context of this decision?
- We recognize the need for personalization:Is personalization recognized as necessary in the context of this decision?
Results show that people are more likely to prefer AI if AI is perceived as being more capable than humans in the context of a particular decision and Personalization is considered unnecessary. However, if any of these conditions are not met, the aversion of AI will manifest.
“People just don't like or hate AI,” Lu said. “Their response depends on the utilitarian needs of AI to effectively complete their work and whether they need to recognize their psychological needs as unique individuals.”
For example, even if AI systems have proven more accurate in identifying skin cancer from medical images, patients often still prefer human physicians. I feel that medical decisions require understanding of their own circumstances.
The meta-analysis also revealed key moderators that influence AI preferences. In countries with low physically specific algorithms, attitudes (not behaviors), and unemployment rates, there was a higher likelihood that physically specific (such as restaurant service robots) would appreciate AI compared to physically specific algorithms. On the other hand, aversion of AI was more pronounced in countries with higher (lower) levels of education and internet use.
“Understanding how people think about AI is just as important as improving the technology itself,” Qin said. “For AI to be trusted and adopted more widely, developers must consider not only how competent it is, but how well it suits the psychological needs of the users.”
This study provides valuable guidance to developers and policymakers, and encourages them to go beyond technical optimization to consider how human psychology shapes people's attitudes towards AI.
“Maxing AI's potential means understanding when it's welcome and not,” says Lu. “Only by addressing both abilities and personalization can we move towards meaningful human collaboration.”
-
New research at MIT Sloan Management School sheds light on an inexplicable paradox. Despite the increased accuracy and efficiency of AI, people still prefer human decisions, even when AI performance improves.
