Does the Uncanny Valley exist?

Erik sez, "This is an analysis of the shortcomings and potential irrelevance of the Uncanny Valley theory. It's based on reporting I did for the current Popular Mechanics cover story on social bots, as well as my face-to-face experience with some of these machines. In a nutshell, our argument is that it was a bad theory to begin with, and only applies to remote viewing of bots. In person, the Valley disintegrates."
According to all of the roboticists and computer scientists we interviewed, the uncanny is in short supply during face-to-face contact with robots. Two of the robots that inspire the most terror--and accompanying YouTube comments--are Osaka University's CB2, a child-like, gray-skinned robot, and KOBIAN, Waseda University's hyper-expressive humanoid. In person, no one rejected the robots. No one screamed and threw chairs at them, or smiled politely and slipped out to report lingering feelings of abject horror. In one case, a local Japanese newspaper tried to force the issue, bringing a group of seniors to visit the full-lipped, almost impossibly creepy-looking KOBIAN. One senior nearly cried, claiming that she felt like the robot truly understood her. A previously skeptical journalist wound up smiling and cuddling with the ominous little CB2. The only exception was a princess from Thailand, who couldn't quite bring herself to help CB2 to its robotic feet.

Royalty notwithstanding, the uncanny effect appears to be an incredibly specific and specialized phenomenon: It seems to happen, when it does, remotely. In person, the uncanny vanishes. There's nothing in the way of peer-reviewed evidence to support this, but then, there's almost nothing to confirm the uncanny effect's existence in the first place. As an unsupported theory that has morphed into a nerdy breed of urban legend, anecdotes are all we have to work with.

The Truth About Robots and the Uncanny Valley: Analysis (Thanks, Erik!)

(Image: Eva, a Creative Commons Attribution photo from Arenamontanus' photostream)


  1. I’ve been whining for years that the uncanny valley was pseudoscientific nonsense. All the things it covers are better explained by specific and distinct instinctive reactions to signs of death, disease, and so on.

    1. A reasonable hypothesis is that the uncanny valley results from the sort of instinctive reactions you cite. How does this make the phenomenon itself “pseudoscientific nonsense”?

      Two notes on the article itself. First, for most people the uncanny valley is experienced while watching visual effects on TV or in movies. So however interesting the case of robots may be, the results don’t bear on what most people think of as the “uncanny valley”. Second, calling the uncanny valley an urban legend is just sloppy. People experience the uncanny valley for themselves, no need for any legends.

      It is fine to study the relative creepiness of various robots, but this result is being oversold.

      1. Seconded- while the idea of the uncanny valley may or may not be exaggerated when applied to humanoid robots, it definitely holds true as a theory of art and character design.

      2. Illustrating multiple distinct phenomena with a single concept isn’t a problem, but if the concept is taken to be something other than useful convenience or taxonomy — for example, if it’s taken to be a unifying causal factor in all the distinct phenomena, as the ‘uncanny valley’ often is — it demands evidence.

        You could call it a “proto science,” I guess, but “proto semiotic circle jerk” is more its intellectual speed.

        1. I’m not sure what you’ve been reading, but I suggest you adopt my standard procedure when dealing with any proto semiotic circle jerk style material. Stop reading. In particular, if someone suggests that the “uncanny valley” is any kind of unifying underlying cause, they should be ignored. It rather signifies an effect, which is that we get creeped out at least a little, as well as a conjecture that we would not be creeped out if what we had seen had not approached a plausibly human appearance.

          As someone who works with facial modeling, animation and rendering every day, I would love to see some good psychological research done on the topic. However, it is technological extremely difficult to study, and unless you are planning on doing some of this work yourself, perhaps you should refrain from dissing those who try to address the issue as best they can in these unenlightened times. Unless, of course, they are just full of it.

  2. I propose an experiment:

    Measure, heart rate, blood pressure, ecg, respiratory rate, or any like biorhythm while a subject views: a picture of a robot, etc.,

    and then in another trial, the same robot in person.

  3. If only it can stop people from spurting out “Uncanny valley!” in the comments of any blog post featuring reasonably realistic CGI characters.

    If only…

  4. But, in general, take my advice, when you meet anything that’s going to be human but isn’t yet, or used to be human but isn’t now, or ought to be human and isn’t, keep your eyes on it and feel for your hatchet.

    Mr. Beaver

  5. Interesting. I never really assumed the uncanny valley effect was something that I or anyone would be experiencing while in the actual presence of a robotic humanoid. I always assumed it applied only to moving images of robots.

    1. “Since when did anyone say the Uncanny Valley applied to anything but CG?”

      Since the phrase was coined by a japanese roboticist named Mori in 1970, long before realistic CG humans were remotely practical.

      The Uncanny Valley is a hypothesis, and has shown, under certain conditions, to be true to various degrees. My personal opinion, as someone who regular works to make those super-realistic CG people that so often inspire the invocation of “oooo. uncanny valley!” is that it’s generally an experience in which we notice the imperfections in something we are expected to believe is human/alive. Speaking to a robot, with the knowledge that it is, in fact, a robot, puts a whole different spin on things.

      This is a good starting point for rigorous testing of the hypothesis. Much like the Turing Test, acurate physical simulation of a human isn’t something that has a single solution. It is, and I suspect, will continue to be a constant quest which will continue for many, many years.

    2. Exactly, it’s got nothing to do with robots, and all about CG. Which can be thought of as “remote viewing”.

  6. I’ve always thought the uncanny valley was silly. I’m glad to see people beginning to take a rigorous approach to it.

  7. I don’t know about the various “theories” of the Uncanny Valley, but I know damn well that robots that mimic human actions are inherently deceptive on an instinctual scale, and consequently promote highly irrational behavior and have a strong potential to be very dangerous.

    There’s a short story (the name of which escapes me), a snippet of which I believe was on BB at some point in the past, which is where I know it from.

    In the snippet, one character has the other use a hammer to destroy a small, robotic beetle on a worktable. The first pair of strikes don’t finish it off, and just when the woman is about to make her third strike, the robot starts to wail. Suddenly she drops the hammer in horror, the librucants and fluids leaking from the machine resembling blood, the high pitch wail resembling death cries.

    Instinctual reactions, behaviors and emotions that are hardwired into us, are very powerful. When machines play with those instincts, we are robbed of our dignity. We can resist only with strength of will and discipline, and our composure and rationality crumbles in the face of strong instinctual messaging. Allowing people to produce robots that expressly manipulate those instincts is very wrong in my mind.

    ~D. Walker

    1. We lose our dignity when we allow our instincts to control us, regardless of what is manipulating them. Saying that we’re powerless to manipulation of our instincts is nonsense, all it takes is willpower.

    2. You bring up an excellent point. People who think they can just avoid manipulation through willpower are lying to themselves.

      If something is flashing; you will look at it, the first time at least. If water drips on you, steadily, hour after hour, while you are tied down, you will undergo torturous effects. Try to find someone who can remain calm if a speaker were to follow them around playing sounds of a screaming baby or a tortured animal.

      It is entirely logical to understand that we do not always function logically, no matter how much willpower we put into it.

      People seem to have this view that they can just ignore good marketing (I actually suspect that marketers support people in this lie). It’s just BS. Certain things will find chinks in your willpower, and when billions of dollars are being spent on finding those chinks – like advertising to your children or friends so the market message appears to come from them – anyone can be susceptible. I never really thought about your example with robots, but it is definitely something to start trying to defend against, even if the defense is the sometimes lacking power of our own wills.

  8. More anecdotes for the nerdy urban-legend mill:

    When looking at a Bunraku doll (in person) or watching the film Polar Express, I get the sense that I’m looking at something wearing a human face that is not at all human. There is no animation from within.

    But– when interacting with a robot that was definitely a candidate for the Uncanny Valley– dead eyes, rigid face, unnatural mouth (in a study at a local university), I was surprised not to have this reaction. I later found out that the “robot” was actually being remotely piloted and voiced by a set of human operators. It was a human personality without a human face.

    In interacting with the “robot” in the study, I felt as if my brain was trying to re-map its definition of “person” and smooth out the sense of the surreal. Because the “robot” was actually sentient, this was uncanny in a completely different way. I kept getting worried that I was going to upset or frighten the robot, which was about the size of a small child.

    Uncanny Valley may be a good catch-all term, but I get the sense that there’s more than one flavor of Weird going on in these experiences. We may need to expand our vocabulary.

  9. I often refer to the Uncanny Valley (it’s in my bag of interesting bits of knowledge), but usually in reference to things other than robots. Does anyone remember the puppets from Jim Henson’s Dark Crystal?

    I think the “they’re almost human, but not quite” explanation perfectly explains why they are so incredibly creepy. While I acknowledge that actually interacting with an uncanny creature might make it less disturbing, I don’t think that disproves the theory at all.

  10. Human faces and body language are incredibly high-bandwidth channels we use to constantly update our theory of mind about other conscious individuals.

    Until there is actually something behind one of those animated faces – CG or mechanical – we will continue to perceive something dreadfully amiss.

    I know there are benefits to machines interacting with people using this existing “language”, but I think it may not be as necessary as some think. R2D2 has no human features, expressions or speech, but we can feel a kindred connection with just a little body language and sound we can map on to this far richer human vocabulary.

  11. My impression from this is that in person, they simply look fake enough that they’re not in the valley.

    Unlike a mime. Brrr.

    1. My impression from this is that in person, they simply look fake enough that they’re not in the valley.

      This is probably a very good explanation of the difference.

      I don’t see how someone can simply say there is no uncanny valley in person. Humans are not creepy, and puppets made from corpses are. It might be easy to avoid, but it’s definitely possible to make a more convincing facsimile of the second than the first if you put work into the wrong features.

  12. The idea that the uncanny valley doesn’t exist in person is rubbish. Countless mimes, clowns, and people in costume have been scaring the crap out of people since before recorded history.

    One of the most scary things I’ve ever seen was a man (I presume, anyway) dressed in a donkey costume. This costume was good enough to look like he’d just skinned the donkey and was wearing it as a disguise. It stood there silently, virtually motionless, looking like the zombie donkey from Hell, just itching to eat my soul with its dead eyes. Regardless of knowing it was just a man in a suit, my brain was screaming “Flee, don’t be fooled, this is a beast from the very bowels of Hell! Let the others in the flock fall prey to its unholy hungers. Flee!!!!”.

  13. What about this study showing an apparent uncanny valley effect in monkeys? Obviously there might be other explanations for this finding, but it does seem to be a bit of scientific evidence which favors the idea.

  14. I agree with t3hmadhatter at #19. There are definitely human-created, non-human 3D likenesses that are freaky. Perhaps the Uncanny Valley has a lower or different threshold of humanity from the original theory.

    It could also be the case that personal interaction with a robot overrides the Uncanny Valley. Humans are practically trained to make social interaction easy. We are constantly trying to make conversations make sense, whether we are speaking to other humans, dogs, or inanimate objects. Robots developed for social interaction, such as the ones highlighted in the article developed by humans to interact with humans.

  15. I’ll tell you what’s giving me uncanny creeps… corporations just won citizen rights from the Supreme Court today. The court thinks corporations are people. I am having revulsion fits! As would the constitution, if it were sentient (now that would be weird!)

    Corporate personhood… horrible idea! Join the resistance!

  16. The author claims there is some evidence against the uncanny valley effect and none for it, without doing a literature review.

    This is a very anecdotal article claiming scientific authority.

  17. I call BS.

    The author provides essentially anecdotal evidence to support his theory. So I’ll rebut with my own, personal anecdote.

    I went to the Word Expo in Japan in ’05. Since it was in Japan there were tons of robots everywhere. One greeting people near the entrance was this one: (Search YouTube for “kokoro robot” and you’ll find lots of appropriate videos.)

    Being in its presence freaked me right the hell out. It triggered my flight instinct, which freaked me out even more. I knew it was just a robot but something in my deep brain wanted me away from it.

    There was another one, I can’t find reference to it, but it was a little higher up the uncanny valley slope. It was a full body woman wearing the usual ‘campaign girl’ outfit. She freaked me out a lot less, but made me uneasy. Even when she was idling her chest moved to simulate breathing, and her eyes and head would survey the room casually, her eyes would turn and lids flare a bit if it heard a loud noise.

    I really like robots. I build robots from scratch. These robots made me, at best, uncomfortable.

  18. What bugs me is the conceit that the Uncanny Valley hypothesis declares something must be creepy, when instead it is best seen as a reason why something seems creepy to an individual. It describes the curve from “that reminds me of” neutrality to “oh, that’s cute how it imitates” partial acceptance before dipping into “okay, that’s really creepy” rejection, before reaching full acceptance.

    It’s not something you can measure, because individuals each have different criteria, and the Uncanny Valley is also coupled with a sense of danger, something that also is different between people. If something remains nonthreatening, then the Uncanny Valley may not be triggered.

  19. somebody’s missing the point: yes, the uncanny valley is a label referring to “specific and distinct instinctive reactions to signs of death, disease, and so on”

  20. Most people who’ve seen Eva and her compatriot Jules, here at BRL have commented how freaky they are and that’s without even seeing them move. This article seems like a load of nonsense to me – having done some work with humanoid robotics. We’ve got a whole bunch of interviews with people saying that they are more comfortable with robots making errors if it matches with how human-like they look. Essentially, the more human-like the robots appearence the more human-like the expectation of behaviour; and the more freaked out you are when it gets it wrong. The uncanny valley is a fairly subtle effect that would require much better experiments to dissprove it conclusively.

    Also I wonder who actually took that photo, given it’s not from our website.

  21. Isn’t it possible that, in addition or as an alternative to all the things mentioned above, this robot is simply climbing the far wall of the valley? No one has even considered the fact that it might be so realistic that it is again starting to look more realistic instead of less?

  22. Still, two out of three robotic overlords agree that Uncanny Valley makes the best ranch dressing ever.

Comments are closed.