Behavioral and neural evidence for the underestimated attractiveness of faces synthesized using an artificial neural network

Abstract

Recent advancements in artificial intelligence (AI) have not eased human anxiety about AI. If such anxiety diminishes human preference for AI-synthesized visual information, the preference should be reduced solely by the belief that the information is synthesized by AI, independently of its appearance. This study tested this hypothesis by asking experimental participants to rate the attractiveness of faces synthesized by an artificial neural network, under the false instruction that some faces were real and others were synthetic. This experimental design isolated the impact of belief on attractiveness ratings from the actual facial appearance. Brain responses were also recorded with fMRI to examine the neural basis of this belief effect. The results showed that participants rated faces significantly lower when they believed them to be synthetic, and this belief altered the responsiveness of fMRI signals to facial attractiveness in the right fusiform cortex. These findings support the notion that human preference for visual information is reduced solely due to the belief that the information is synthesized by AI, suggesting that AI and robot design should focus not only on enhancing appearance but also on alleviating human anxiety about them.