In scientific communication, realistic avatars may promote more credibility than cartoon-like ones

Tiktok has exceptional “testimonies” like Nikola Tesla and Marie Curie, delivering a short science-related message that has attracted millions of opinions. This is just one of many examples of AI-generated avatars used to convey science. This is a strategy that may also have its drawbacks.

Image and animation generation via artificial intelligence is a rapidly growing field, with quality constantly improving. However, many avatars, although realistic, are still easily picked up by human observers, such as small flaws (glitches, delays, inconsistent facial expressions and lip sinks).

Jasmine Beake, a researcher at the Centre for Advanced Internet Research (CAIS) in Bouchum, Germany, and other authors of the Journal of Science Communication’s work, have noticed that these avatars can cause what is known in cognitive science as the “mysterious valley.”

The Spooky Valley describes the human reaction to humanoid butter (digital or robotic). Hyperreal but not perfect can cause a strong discomfort, but more stylized or cartoonish humanoid figures.

The eerie valley can cause total rejection for viewers, and Baake and his colleagues wondered how the human-like traits of AI-Avatars, representing science communicators, influence the credibility that the viewers have caused them.

“We wanted to do research into the perception of these avatars and in particular how their degree of realism and gender influence recipient perceptions of reliability,” explains Baake.

The study (conducted in German, German) included a series of videos featuring AI-generated avatars depicting both male and female science communicators. There were four experimental conditions, which differed depending on the realism of the avatar (very high vs. cartoon style) and gender (male or female).

Nearly 500 participants were recruited through a representative online sample from Germany, selected to reflect population in terms of age, gender and education.

“Because of the degree of realism, I assumed that more stylistic avatars would be perceived as more reliable, with the eerie valley hypothesis in mind,” Baake says.

“And based on existing literature on gendered perceptions of science communicators, we show that male scientists are often associated with greater abilities – based on concerns that AI-generated avatars could reflect and reinforce such stereotypes due to biased training data, we hypothesized that male avatars are more reliable than female avatars.

But Baake and his colleagues were surprised. In their experiments, realistic avatars were more positively evaluated than cartoon style avatars. In particular, the survey given to participants after watching the video was rated as assessing the perceived avatar’s ability, integrity and mercy.

The more realistic avatar scored slightly higher in all three dimensions. Regarding gender, the effect was partial. Male avatars were perceived as more competent, but no significant differences were found in terms of honesty or compassion.

“Our findings did not find a descent into a creepy valley with a higher degree of realism, at least in our conditions,” Baake commented. Furthermore, individual factors, such as viewers’ previous AI knowledge and trust in science, were found to moderate perceptions of reliability.

The findings of this study show that more realistic, human-like avatars seem to be suitable for the transmission of scientific content.

However, Baake found no eerie valley effect here, but future research should test a broader level of realism to investigate whether a intermediate eerie valley effect appears between the two conditions tested so far, and whether people perceive different avatar realism depending on the observer.