Images of faces created by Artificial Intelligence (AI) are seen as more trustworthy than images of genuine faces say researchers, who warn of the risks of online fraud and other harms.

This is the first ever study to examine the trustworthiness of AI faces created by the latest diffusion technology and was led by Alexis McGuire with Paul Taylor and Sophie Nightingale from Lancaster University, Maty Bohacek from Stanford University and Hany Farid from the University of California, Berkeley.

Psychology PhD student Alexis McGuire said: “Our research shows that people are at risk of being fooled by AI-generated images. These AI models have democratised the online space and they are accessible for anyone without technical skills to create fake faces that can be used for a variety of different harms. It is important to inform the public about the ease of creating such images and the potential misuses, and ways in which they might fall victim, for example, through the spread of misinformation, identify fraud, and catfishing.”

Humans are experts at processing real faces, automatically assessing a face in as little as 100 milliseconds. However, AI-generated faces are highly realistic and are becoming more trustworthy with newer, more sophisticated technology creating fake images that can fool people into thinking they are genuine around a third of the time.

When 169 participants were asked to look at a collection of 96 faces (diverse across race, gender, and age) presented at random and indicate whether each face was real or AI-synthesised, their average accuracy was 58.4% - only slightly better than random guessing (similar to flipping a coin at 50%). Surprisingly, faces generated by the newer AI diffusion model (DM) were rated as less realistic than faces produced by an earlier AI model (GAN).

However, in a follow up experiment, a new set of participants were asked to rate the trustworthiness of 96 faces presented at random on a scale of one (very untrustworthy) to seven (very trustworthy).

Real faces were rated as the least trustworthy with an average trust rating of 4.03. Both types of AI-synthesised faces were rated as more trustworthy than real faces while faces produced by the diffusion model (DM) were more trustworthy than both the real and GAN faces. GAN faces received an average trust rating of 4.36, and diffusion-synthesized DM faces were the most trustworthy with an average rating of 4.70.

Researchers say it is puzzling that AI synthesised faces generated by the newer AI diffusion model (DM) were rated as less realistic than faces produced by an earlier model (GAN) - but DM faces were still rated the most trustworthy.

Alexis McGuire said: “This finding presents a paradox and thus highlights the possibility that realism and trustworthiness judgements are driven by two different psychological mechanisms.”

She warned of how AI faces generated using the latest DM technology could contribute to an overall erosion of trust in society.

“As AI-generated images become more sophisticated and more accessible, as a society, we are increasingly exposed to artificially-generated faces—often in nefarious and exploitative scenarios, such as political disinformation, financial and identity fraud, and catfishing. It is critical to understand the threat this democratisation of generative AI brings as well as developing strategies to mitigate potential harms to individuals, organisations, and democracies.”

The research in the Journal of Vision entitled “AI-Generated Faces are Becoming More Trustworthy” was funded by The Centre for Research and Evidence on Security Threats (CREST) and Security Lancaster.

Anyone interested is encouraged to take part in an anonymous online survey called ‘Examining Individual Differences in the Detection of Real and AI-generated Faces’.

Participants will see an array of faces on at a time and be asked to rate if they are real or AI, along with a few other questions for example to rate their confidence. There will be a score at the end.