Skip to content
July 19, 2011 / Dr. Toad

The uncanny valley in the brain

We like things that look like us. We anthropomorphize animals, things and even abstract ideas.

Part of still from Finding Nemo

Eros, Greek god of love

So it’s no surprise that we like machines to look like us too. Only, there’s a catch. They can’t look too much like us without being completely like us. Somewhere between the lovable cartoonish humanoid looks of Mertz and the fully human looks of people, lovable or not, there is the creepy android look that is thoroughly unlikable.

This phenomenon: as the thing gets close but not quite close enough to apparent human-ness, all of a sudden the humanish-looking thing starts freaking humans out, is called the uncanny valley.

Masahiro Mori's uncanny valley graph 1970

Masahiro Mori's uncanny valley graph (1970)

Mark Brown for Wired UK reports on a new study led by Ayse Pinar Saygin of UCSD which found the brain’s uncanny valley response. It turns out that the motion-processing parts of the parietal cortex lit up when the subjects viewed videos of the spooky very realistic android Repliee Q2 (in the video above), including areas known for mirror neurons. This suggests that it’s the discrepancy between the biologically realistic appearance of the thing, and its biologically implausible, mechanical motion that freaks out the human brain.

Parietal cortex lights up when presented with video of realistic android

(A) Robot (B) Realistic-looking android (C) Human

It’s interesting that some roboticists think that the uncanny valley is nothing but disgust in the face of bad design. This finding may corroborate that view, if by bad design we mean bad — that is, unrealistic — motion algorithms. Similar abilities and looks are needed to climb out of the uncanny valley or the brain gets spooked by the incongruity.

Advertisements

4 Comments

Leave a Comment
  1. Ben / Jul 25 2011 4:29 pm

    That’s a fascinating new study that provides some exciting new debate fuel. First off, I love the Lovecraftian title of Saygin’s paper: “The Thing That Should Not Be.”

    As you probably know, the concept of the Uncanny Valley has been fairly controversial, largely because, as David Hanson has pointed out, that famous graph has been said to oversimplify the concept of “human-like” – and also because, until now, the graph was based only on hypothesized responses rather than hard data. But now all that could be poised to change.

    However, it’s still worth noting that some neuroscientists, such as Hickock and Pascolo, have called the continued to question the entire concept of a mirror-neuron system. Though this new uncanny valley research is definitely intriguing, research continues to unravel what role, exactly, the temporoparietal regions play in facial recognition and communicative expectations.

    It’s pretty thrilling, though, to see this discussion move out of the realm of hypothesis and into the field of serious neurophysiology.

  2. Ben / Jul 25 2011 4:48 pm

    [My apologies; my previous comment here was posted under the wrong WP login, and I wanted to make sure the link from my name pointed to my blog.]

    That’s a fascinating new study that provides some exciting new debate fuel. First off, I love the Lovecraftian title of Saygin’s paper: “The Thing That Should Not Be.”

    As you probably know, the concept of the Uncanny Valley has been fairly controversial, largely because, as David Hanson has pointed out, that famous graph has been said to oversimplify the concept of “human-like” – and also because, until now, the graph was based only on hypothesized responses rather than hard data. But now all that could be poised to change.

    However, it’s still worth noting that some neuroscientists, such as Hickock and Pascolo, have called the continued to question the entire concept of a mirror-neuron system. Though this new uncanny valley research is definitely intriguing, research continues to unravel what role, exactly, the temporoparietal regions play in facial recognition and communicative expectations.

    It’s pretty thrilling, though, to see this discussion move out of the realm of hypothesis and into the field of serious neurophysiology.

  3. Dr. Toad / Jul 25 2011 5:00 pm

    Hi Ben! I’m not sure what’s going on with your link — apparently that blog doesn’t exist.

    I agree with you that this had been an untested hypothesis for too long and now it’s fascinating to find solid experimental probing into it.
    I also remember clearly hearing someone in a talk give a great demonstration of how characters or robots apparently on the same level of being human-like or not could be more or less “familiar” or likable depending on the particulars of their design. Unfortunately I couldn’t remember who was giving the talk and what his examples were (apparently, I remember that it was a man and that it wasn’t David Hanson).

  4. Ben / Jul 26 2011 3:55 pm

    Yeah, I managed to fix the link after I posted two comments with the wrong one, but I do believe it’s fixed as of this post. Please feel free to delete one or both of my previous comments.

    I agree that we need more studies on the particulars that make a robot seem more or less human-like. Based on the research I’ve seen, someone needs to develop a set of uncanniness parameters incorporating attributes like visual abstractness, neoteny, and so on. Those may turn out to converge on a valley as well, but I suspect we’ll also discover some other interesting correspondences along the way.

What say you?

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

%d bloggers like this: