We spoke with Rutsuko “Ruth” S. Nagayama, Professor of Psychology at Shizuoka Eiwa Gakuin University, for the latest Spotlight on Science Q&A. Here, she reflects on an article she co-authored in 2007 for PRESENCE: Teleoperators and Virtual Environments. “The Uncanny Valley: Effect of Realism on the Impression of Artificial Human Faces” has been one of the journal’s Top 5 Most Downloaded Articles in the past year. Read the article for free on our SOS page.
Can you talk a little bit about “the uncanny valley?”
The uncanny valley is a hypothesis about the psychological reaction when we see a robot, and was proposed by a roboticist Masahiro Mori in 1970. Mori argued that although it would be a good thing to make a robot's appearance more humanlike, people could feel uncomfortable with robots that were almost (but not perfectly) humanlike.
A graph used by Mori to explain his hypothesis is well known. In his graph, the horizontal axis represents how much artificial objects (robots, dolls, prosthetics, etc.) resemble real human beings. The vertical axis represents a kind of impression score of artificial objects as rated by human observers.
Mori predicted that the more closely artificial objects resembled real humans, the more comfortable our impression of them would be. But when their resemblance reached very close to real humans, we would have negative impressions of them. Mori depicted the occurrence of the negative impressions as a valley in the graph. This portion of the graph is called the uncanny valley. Mori warned that obviously robotic appearance is okay, but making them highly humanlike runs a risk of falling into the valley.
What are some of the challenges of trying to quantitatively measure the uncanny valley?
Mori’s graph is merely an explanatory diagram, not a plot of actual data. So, we decided to empirically test the hypothesis via psychological experiments. In our experiments, participants rated their impressions of face images with various degrees of human likeness. We experimental psychologists are familiar with measuring participants’ impressions on something and analyzing them, so carrying out the experiment per se was not difficult for us.
We expected that plotting the rated impression as a function of human likeness would reproduce Mori’s graph, but the valley did not appear in our graphs. In other words, observers did not have strong negative impressions of the face images. In our next experiment, we used face images that had extraordinarily large eyes. Participants rated negative impressions on those faces, but only when the faces resembled real humans. For faces that had obviously artificial appearances, the large eyes did not produce a negative impression. This suggests that the human visual system’s sensitivity to facial features is higher for realistic human faces than for artificial faces.
These experiments revealed that just thinking of the degree of human likeness is not sufficient to understand the uncanny valley. It should be interpreted as an interaction of human likeness and eerie facial features.
Since most participants in your experiments were Japanese, you discuss how a cultural background in Japanese style animation might have affected your results. How might the results be different if the experiments were conducted with Western participants? Could aspects of the uncanny valley change depending on culture?
The eerie facial feature we used in the experiment was an extraordinarily large eye size. Such large eyes are commonly seen in Japanese animation and cartoon characters. So, we mentioned in our paper that the Japanese cultural background of the participants might have affected the experimental results. At this time, however, we do not think the comparison between the Japanese-Western cultural backgrounds is a very compelling point of view, because large-eyed characters are not limited to made-in-Japan characters, and not all ANIME characters have large eyes.
Nevertheless, I’d like to suggest that an observer’s background and past experiences could influence the uncanny valley. Among young people in Japan, photo booths with facial-beauty enhancement functionalities are very popular. The other day, my daughter showed me photos taken at such a booth, and I noticed that her eyes were much larger than those used in our experiment! It was obviously eerie to me, but young people feel such large eyes are “cute.” Perhaps the meaning of the uncanny valley differs between generations.
Another example is an answer from an android developer to my question about his feelings on his android, which was, at least to my eyes, obviously a dweller at the bottom of the uncanny valley. He replied that he did not feel his android was eerie.
What do you hope people will gain from this study?
I would be very happy if our study can tell people the importance of examining through experiments or thinking based on real data. Mori's insightful hypothesis is very fascinating, and it has led to a vast number of conjectures on the physical appearances of robots, androids, game characters, etc. However, we found that the details of the uncanny valley were not as expected as in the original hypothesis. Experimental studies cost time and money but they are worthwhile.
What sort of a future do you see for the development of robotics?
An answer straight out of the textbook is that any future designers of humanlike entities should take the uncanny valley into consideration. This answer is based on a doctrine that things made by humans should be made to suit to humans. However, considering the cases of my daughter’s “cute” huge-eyed portraits and an android developer’s feelings about his android, humans are more adaptable than roboticists might think. Even if the design of a robot is not optimized for human psychological mechanisms, people may possibly accept the robot as a comfortable friend. Perhaps it may be okay that future roboticists develop robots without worrying too much about the uncanny valley.
Read “The Uncanny Valley: Effect of Realism on the Impression of Artificial Human Faces” for free on our SOS page.