A celebrity staring at you with a frozen expression can be somewhat titillating — until you realize she is actually just a wax model. While her physical resemblance to a living human is an impressive work of art, it may be slightly creepy to some.
This uncomfortable perception of ‘creepiness’ is marked by a phenomenon called the “uncanny valley” effect. The term stems from the idea that an agent’s likeability increases with the prominence of humanoid characteristics, but suddenly drops when the similarity becomes too extreme and disconcerting. It’s not that the concept of a mechanical agent must inherently irk us or that every agent with an artificial smile is guaranteed to induce chills. Rather, it’s simply that the agent appears human without actually being human. The “uncanny valley” effect may be symptomatic of the brain generating a prediction error as it tries to match incoming humanoid features against a database of unique human characteristics.
The effect, originally based on anecdotal evidence and then later investigated in the fields of robotics and computer graphics, is an increasingly important topic in neuroscientific research. Findings from a study published in this past April issue of Social Cognitive and Affective Neuroscience suggest the discomfort is due to an intuitive disconnect between the perceived agent’s physical appearance and bodily motion. The group of researchers conducted a functional MRI study on participants’ neural responses to video clips of agents performing common actions like waving, nodding, and picking up a piece of paper off a tabletop. The agents were divided into three conditions: android, human, and robot. The android, named Repliee Q2, was modeled after the actual human featured in the human video clip. As for the robot, its features were highlighted by stripping it down to its metal innards, such that it looked unmistakably inanimate.
Upon viewing the video clips, participants were immediately told which videos featured either a robot or human and were then put through the fMRI scanner. Once inside the scanner, participants were shown dozens of two-second videos of the agents spread 500ms apart.
The most significant change in brain response occurred when subjects were shown the android, specifically in the regions of the parietal and frontal cortex. Repetition suppression was strongest in these regions under the android condition because the processing of the android stimulus caused neural conflicts as the brain tried to match android features with human ones. In other words, the brain tended to freak out when it saw the android because it was trying to figure out what it meant.
In contrast, fMRI results showed that participants had no trouble processing the correspondence between human appearance and human movement or robotic appearance and robotic movement. Ayse Pinar Saygin, the corresponding author of the study, explains this smooth association as the brain “… looking for its expectations to be met — for appearance and motion to be congruent.”
This need for congruency, exhibited by the “uncanny valley” effect, extends to the likeability of anthropomorphic robots, as well as characters in video games and animated movies. Characters that share low to moderate human characteristics like dolls, cartoon animals, and R2D2 often elicit more positive responses compared to the animated characters in the film The Polar Express or more modern androids like Repliee Q2.
Perhaps one day we will become more accustomed to humanoid agents as they increase in ubiquity. Do the robot.
