How psychologists are using robots to study the “uncanny valley”

Mentioning the uncanny valley often brings one thing to mind — creepy dolls. But the faces of dolls in particular are a cultural touchstone for uncanny feelings — at least in part due to their (over)use as a spooking-device in hundreds of horror movies over the years.

As such, psychological research has been conducted on the subtleties that non-human facial structure and expression can have on producing feelings of unease in those who view them. However, the uncanny valley isn't just confined to faces, and its effects are not confined to just a horror movie device. For example, research from Burcu Urgen of Bilkent University demonstrates that biological-like motion can also trigger uncanny feelings, which poses real issues for those pushing the frontiers of robotics.

One of her studies from 2018 looked at the uncanny valley through the lens of expectation violations. For this experiment, 19 student participants from the University of California were strapped into EEGs and asked to view video clips of either a mechanical robot, an android (which was the same robot, but with a very realistic skin), or the human that the android was modelled after.

In these clips, each of the agents carries out a simple movement, such as wiping a desk or lifting a cup. In half of these trials, the video was paused on the first frame for between 600ms and 1500ms, in order to provide motionless control measurements, whereas in the other trials the video played instantly.

Analyses focused on a pattern of brain activity called the N400 — a relatively large spike in neural activity that occurs 400 milliseconds after seeing something that violates your expectations. The relationship between the N400 and expectation violations is well established, and has been demonstrated to be a good indicator that people's expectations have been violated during a variety of tasks such as reading, looking at faces, and interpreting gestures.

The research team found large spikes in the N400 during the android robot's movements, but not when it was static during those paused first frames. There was no indication of expectation violation when either the mechanical robot or human performed the same movements. Taken together, this means that the surprisingly mechanical movements of an otherwise human-like robot, but not the overall form or design of the robot, were what set it apart. This non-matching, or "incongruence", was the main issue.

"In our study, it was the incongruence of appearance and movement of the agent [that] led people to think that it would move biologically, since we know from our prior experience in the world that agents that look biological move biologically," Dr Urgen shared with us. "However, when it [didn't], it violated people's expectations."

But it may not be this way forever. The usual uncanny feelings may not arise once we come to understand that life-like robots don't always move in a biological manner (or indeed, that some mechanical-looking robots can actually move with fluidity and grace – just think of the dogs created by Boston Dynamics). "The main point about our "expectation violation" explanation of the uncanny valley supported by N400 data is that we use our prior knowledge when we are processing our environment, and constantly predict what comes next," Dr Urgen explains. "This implies that if our prior knowledge with robots changes, our expectations and what we predict will change as well. So, if we get more familiar with robots over long periods of time, it is possible that we may begin to move past the valley."

And this is exactly what's on the cards for Dr Urgen and her lab. In the near future, they plan to take a closer look at the effect of participants' age on how much robots violate expectations. For example, younger generations who have grown up around technology may be more likely to expect miraculously human feats from robotic agents than older generations. This could also feasibly vary by culture; one could envision a Japanese population, for instance, perhaps being more familiar with the idea of a robot moving biologically than one from say, Peru.

Future research could also look at other features that produce uncanny feelings, Dr Urgen says. "I think this expectation violation issue is not limited to appearance-movement congruence. Our real world experience is multimodal in nature. For instance, the congruence between appearance and voice may also be important."

Research into these factors informs robotic design and, in the near future, may provide a long-sought-after quantitative measure of the uncanny valley. In theory, N400 readings could be measured while participants view new robotic designs, allowing roboticists to more sensitively and quickly tweak their models into non-eerie territory.

For more on the uncanny valley, be sure to check out the features in December's issue of The Psychologist.

About the author

Emma L. Barratt (@E_Barratt) is a staff writer at BPS Research Digest