講演情報
[3O2-IS-3-04]The uncanny valley electroencephalogram measurement methods using image discrimination tasks for robots and humans
〇Mao Takagi1, Yukako Ito1, Motoyuki Sanada2, Yasushi Naruse2, Ikuko Eguchi Yairi1 (1. Sophia University, 2. National Institute of Information and Communications Technology)
regular
キーワード：
Uncanny valley、Robot、ERP、EEG、Classification Task
The uncanny valley hypothesis proposes that as artificial agents become more human-like, affinity initially increases but then sharply decreases past a certain point before rising again with further human likeness. This study investigates an objective method for measuring the uncanny valley using EEG. In the experiment, 100 robot images with Human-Likeness (HL) scores from 0 to 100 in increments, plus 10 human images, served as stimuli. Participants viewed each image 12 times while performing a robot/human discrimination task, during which EEG data were collected. The data were processed, and event-related potentials (ERPs) were analyzed for each HL level. Results showed that the P2 component's amplitude, a positive wave occurring 150–200 ms after stimulus onset at electrode Fz, varied across HL conditions. It was low at HL 0–10, increased up to HL 60–70, dropped sharply at HL 70–90, and recovered at HL 90–100, matching the uncanny valley pattern. As P2 is linked to cognitive processes like stimulus change and feature detection, it could serve as an indicator for the uncanny valley. These findings suggest P2 amplitude during robot/human image discrimination may provide an objective measure of the uncanny valley.