News Release

Uncanny valley for interactive social agents: an experimental study

Peer-Reviewed Publication

Beijing Zhongke Journal Publising Co. Ltd.

The four different robots used in the experiment.

image: The figure shows these human-like agents/entities. All robots share the same architecture. view more 

Credit: Beijing Zhongke Journal Publising Co. Ltd.

In this study, we evaluate the uncanny valley theory through live human interactions with four human-like entities:

1. Maya: simple voice assistant (only a human voice).

2. : child-sized programmable humanoid robot with articulated limbs but without skin or hair.

3. Nicole: virtual human with a complete virtual human-like embodiment.

4. Nadine: complete life-sized humanoid robot with skin, articulated hands, and other human-like features.

Our study intends to answer the following research questions.

· Uncanny valley for Interactive Humanoid Robots: Exploring this theory to provide an in-depth look into how people's emotions and perceptions vary for different types of human-like interactive robots. Examining how the Uncanny valley affects the current human generation, which is more accustomed to advanced technologies and may be more open to human-like entities.

·Using AI for Uncanny valley quantification: Quantifying the emotional responses of participants using surveys and multimodal emotion and sentiment analyses (using visual, audio, and text).

The results show that all four robots were perceived in various manners and that the emotions expressed varied in their regard. Nadinewas chosen as the “favorite” robot. The only characteristic that she failed to top was likeability, where she tied with Nao. Furthermore, Maya was seen as the least anthropomorphic and animate, which is predictable given that she is only a voice assistant. Nao and Nicole scored the same on anthropomorphism and animacy. However, Nao was significantly more likeable, which indicated that a physical body may elicit a higher degree of likeability. Nadine, the most humanoid robot, was seen as the most likable robot, with similar findings obtained through the sentiment and facial expression analyses. Nao, the humanoid-but-toyish robot, was also seen as likable and provoked the highest positive surprise. However, it incurred high disgust, as determined through the analyses of audio data, without evoking many emotions. However, while being the most likable, Nadine generated the most sadness and was the second most feared. This fear is possibly indicative of the “uncanny valley” effect. Nonetheless, with the “uncanny valley,” eeriness would be expected to increase with the increasing degree of anthropomorphism, which was not the case. The highest fear was expressed towards Maya, the voice assistant without visual characteristics. Furthermore, owing to the human-like appearance of Nadine, most participants were tentative and fearful as they wanted to impress her. This fear can be considered good, as it indicates that people want to connect with her. Extending our multimodal analysis to other cues such as pose estimation, body language cues could be considered for detecting the evidence of an “uncanny valley”. While Nao had a toy-like, childish appeal, Nadine had a human-resembling body, which was clearly sufficient for increasing her likability compared to a bodyless or virtual agent. However, all robots were wellliked. The lowest-ranked robot scored above 60/100, which indicated that no “uncanny valley” effect was determined in this study. Furthermore, the lack of a correlation between being more anthropomorphic and less likeable and provoking more negative emotions provides evidence against the hypothesis. Compared to previous research, the robots used in this investigation might not have provoked the uncanny valley effect, as they have been carefully and coherently designed and constructed.

The results showed that the robots incited different emotions, but the most anthropomorphic robots were the most liked. Furthermore, all robots were well-liked, and there was no correlation between the anthropomorphism of the robots and the negative emotions they provoked. Therefore, this study, like previous studies, did not observe or substantiate the uncanny valley hypothesis. This could be because of the characteristics of these specific robots or the specifics of today's world in which both humanoid and nonhumanoid robots are becoming increasingly prevalent and people are accustomed to them. Regardless, the future design of interactive robots should be open to creating anthropomorphic robots, while ensuring a coherent design.

Disclaimer: AAAS and EurekAlert! are not responsible for the accuracy of news releases posted to EurekAlert! by contributing institutions or for the use of any information through the EurekAlert system.