In scientific communication, realistic avatars can promote more reliability than cartoons

On Tiktok, there are exceptional “testimonies” like Nikola Tesla or Marie Curie delivering short messages related to science that have collected millions of views. This is only one of the many examples where the avatars generated by AI are used to communicate science – a strategy that could also have its drawbacks.

The generation of images and animations through artificial intelligence is a rapidly growing area, constantly improving quality. However, many avatars, although realistic, always have minor defects – glithes, delays, incoherent facial expressions or lip synchronization – sometimes barely perceptible, but always easily picked up by a human observer.

Jasmin Baake, researcher at the Center for Advanced Internet Studies (CAI), Bochum, Germany and the other authors of a study in the Journal of Science Communication realized that these avatars could trigger a phenomenon known in cognitive sciences as the “strange valley”.

The strange valley describes a human reaction to humanoid avatars (digital or robotic): when they seem super realistic but not quite perfect, they can cause strong discomfort, while the more stylized humanoid figures or caricatu one.

The strange valley can cause outright rejection among viewers, and Baake and its colleagues have wondered to what extent the human characteristics of the AI-AVATARS representing scientific communicators influence the reliability attributed to them by the spectator.

“We wanted to do research on the perception of these avatars and especially on how their degree of realism and their sex could have an impact on the recipient’s reliability perception,” explains Baake.

The study (conducted in Germany, in German) involved a series of videos starring avatars generated by AI depicting scientific communicators – both male and female. The experimental conditions were four, varying according to the realism of avatar (very high vs caricatural style) and sex (male or woman).

The nearly 500 participants were recruited by an online sample representative in Germany, selected to reflect the population in terms of age, sex and education.

“For the degree of realism, we assumed that with the hypothesis of the strange valley in mind, the most stylistic avatars, so those who seemed a little more caricatured, would be perceived as more reliable,” explains Baake.

“And, on the basis of the existing literature on the gender perceptions of scientific communicators – which show that male scientists are often associated with greater competence – composed of concerns that the avatars generated by AI can reflect and strengthen such stereotypes due to biased training data, we have hypothetical that male male avatars would be perceived.”

However, Baake and his colleagues were surprised: in their experiences, realistic avatars were evaluated more positively than those of cartoon style. In particular, the questionnaires given to the participants after watching the videos evaluated the competence, the perceived integrity and benevolence of the avatars, which reflect the perceived reliability together.

The most realistic avatars have obtained a slightly higher score on the three dimensions. As for the genre, the effect was partial: male avatars were perceived as more competent, but no significant difference was found in terms of integrity or benevolence.

“With our results, we could not find a descent into the strange valley with a higher degree of realism, at least in our conditions,” comments Baake. In addition, individual factors – such as previous knowledge of AI and the confidence of viewers in science – have been found to moderate reliability perceptions.

According to the results of the study, more realistic human -type avatars seem adapted to the communication of scientific content.

However, Baake underlines that even if no effect of the strange valley has been found here, future studies should test a wider range of realism levels to determine whether an intermediate effect of the strange valley could emerge between the two conditions tested so far and if people perceive the realism of the avatars according to the observer.