Fig 1.
The panellists were in a private room away from the main conference, each wearing a head-mounted display through which they could see and interact with virtual representations of one another.
(A) During rehearsals with others present (B) During the actual panel.
Fig 2.
The virtual room where the discussion took place.
(A) An overview of the room showing all the panellists. (B) The Moderator is speaking, and note also the view through the window. (C) The spatial relationship between the Moderator and AT. (D) The AT avatar speaking.
Table 1.
The post-panel questionnaire. The first column gives the variable names used for analysis, and the second column the question.
Fig 3.
Top: an audience view showing also RO operating and recording the video. Bottom: By the end of the virtual panel all the panellists were on stage while AT continued to be displayed.
Fig 4.
Overview of the pipeline for AT’s intervention. XP’s intervention was transcribed using Whisper STT.
If the keyword was detected, a prompt was generated for ChatGPT that included the summary of the conversation. The answer received became AT’s intervention, which was converted to audio using ElevenLabs TTS and played through AT’s avatar.
Fig 5.
Box plots of the VR and Choice questions.
Fig 6.
Box plots for the turing questions.
Table 2.
Topics identified.
Table 3.
Representative sentences.
Table 4.
Topics identified.
Table 5.
Topics identified.