Electrophysiological Correlates of the Uncanny Valley

To understand how the brain responds to robots with varying levels of human-likeness, we investigated the electrophysiological neural correlates of the uncanny valley effect while human observers viewed pictures of various robot faces. We examined three event-related potentials (N400, N170, and Late Positive Potential, or LPP) and two oscillatory components (frontal alpha asymmetry and frontal-midline theta) and found characteristic non-linear trajectories of the N400 amplitude when plotted against the self-rated human-likeness of the robot faces, similar to the trajectory found for the self-rated likability score. The frontal alpha asymmetry also showed a similar trajectory but not was statistically significant, while the N170, LPP, and frontal-midline theta showed parabolic trajectories against the human-likeness scores. The N170 appeared to show larger amplitudes for more mechanical robot faces with little change across more human-like faces. The LPP and frontal-midline theta produced the highest amplitudes for the middle ranges of the human-likeness scores, with lower amplitudes for more mechanical and near-human-like faces. These results corroborated the proposal that the uncanny valley effect stems from expectation violation for human faces, but they surprisingly provided little evidence for affective reactions to uncanny robot faces. The LPP and frontal-midline theta could reflect increased cognitive control for ambiguous faces rather than their uncanniness, whereas the N170 appeared to reflect visual processing of facial configurations irrespective of their emotional contents. These findings suggest that the uncanny valley effect is primarily a cognitive phenomenon, and its emotional impact might be less prominent than commonly believed.