AI-Generated Faces Outrank Real People in Trustworthiness, Study Finds

Humans are increasingly unable to distinguish between real faces and those generated by artificial intelligence, and, surprisingly, often find AI-generated faces more trustworthy than authentic human ones, according to a study from Lancaster University. This finding raises concerns about the potential for AI-powered identity fraud and online scams.

Researchers found that participants in their study were only slightly better than chance, achieving 58.4% accuracy, when attempting to identify AI-generated faces from real ones. This difficulty is attributed to advancements in AI image generation, which have overcome previous tell-tale signs like physical anomalies such as extra fingers or misaligned teeth.

"The fact that people often perceive AI-generated faces as trustworthy makes them particularly powerful tools for online scams and disinformation," said lead author Alexis McGuire, a PhD student at Lancaster University. "For example, a text-based scam may become more convincing if it is accompanied by a face that people instinctively trust."

The study involved 169 participants who were shown a mix of real and AI-generated faces and asked to determine their origin. While accuracy varied slightly based on the AI model and ethnicity, the overall trend indicated a significant challenge in detection.

An additional test revealed that real human faces were rated the least trustworthy, scoring an average of 4.04 on a scale of one to seven. Faces generated by older AI models, known as Generative Adversarial Networks (GANs), scored 4.36, while those produced by newer 'diffusion models' achieved the highest trust score of 4.7. Curiously, participants rated diffusion model faces as more trustworthy even when they perceived them as less realistic, suggesting that trustworthiness and realism judgments may stem from different psychological mechanisms.

Researchers hypothesize that AI-generated faces might appear more trustworthy because they tend to average out facial features, aligning closely with a generalized human facial template that the brain finds familiar. Additionally, AI models often create idealized and attractive faces, which studies have long linked to perceptions of trustworthiness.

These findings highlight a growing vulnerability to online deception, as AI-generated faces can be seamlessly integrated into fraudulent schemes, making them more convincing to unsuspecting individuals. The research team has made an online survey available for the public to test their own ability to spot AI-generated faces.