We describe a study based on a panel discussion that took place in social virtual reality (VR) at the conference IEEE VR 2025. Each panellist was embodied in a virtual body that looked like themselves. The VR scene was projected onto a screen in front of the audience in the main conference theatre. During the course of the panel two of the panellists swapped avatars, but tried to act as if they were the other person. Additionally, a large language model-controlled Alan Turing (AT) avatar participated. The study aimed to assess the audience’s ability to detect the human identity swap and their perception of the AT panellist. We found that about 40 of 100 attendees who answered a post panel survey did not notice the body swap, highlighting a form of change blindness towards social identity in VR. While AT was seen as less realistic and somewhat distracting, its inclusion demonstrated the increasing capabilities of AI in natural language processing and interaction. The paper emphasises the critical need for ethical considerations, such as identity verification and guidelines for representing historical figures, as virtual reality platforms that can represent historical figures in combination with LLMs become more widespread.
Oliva et al. (Thu,) studied this question.