Taken at Face Value: Do Robots Trigger Face-Typical Processing?

Social robots are increasingly integrated into everyday environments, yet effective and sociallyengaging interactions remain a challenge – particularly due to difficulties in interpreting therobots’ internal states and behaviors. A key obstacle is that robots often fail to engage the socialcognitive mechanisms typically elicited by human interaction partners. Human faces, inparticular, are processed holistically through specialized neural networks that support rapidinferences about identity, emotion, and intention. In contrast, robot faces vary widely in theirresemblance to the human face configuration, both in terms of the number and the arrangementof features (e.g., eyes-over-nose-over-mouth), raising the question of whether they elicit human-levelface processing. Across three experiments, we used the face inversion paradigm to examinewhether and under which conditions robot faces are processed like human faces. Experiment 1showed that robot faces generally disrupt face-typical holistic processing. However, additionalexperiments demonstrated improved holistic processing when robot faces incorporated a greaternumber of human facial features (Experiment 2) or featured distinctly human as opposed to nonhuman features (Experiment 3). These findings suggest that synthetic faces with a high numberof human-like features can engage holistic face processing comparable to that seen with realhuman faces. Given the link between holistic face processing and the attribution of mental statesand higher-order social cognition, these results may indicate that meticulous facial design couldmake robots appear more social, which in turn could promote social cognition in human-robotinteractions.