Can we rewire our human distrust of robots that look like us?
Kyoto University’s Geminoid F robots carry a trace of the mechanism beneath their 'skin’
Photo: AFP/Getty Images
10:20PM GMT 25 Nov 2013
112 Comments
We humans have evolved to relate emotionally to inanimate artefacts, which is
strange when you think about it. Children play with dolls and toy soldiers
as if they were people. Adults talk to their cars. As long as they are
robot-like and “mechanical”, we are comfortable around them, and can display
affection (as for an old car). But when it comes to human-like robots,
something unnerving happens.
As they acquire more human-like features, our affection wanes and we begin to
get a creepy feeling. Our liking turns to revulsion. Androids that look too
human freak us out.
This odd phenomenon is called the “uncanny valley”, and it has befuddled
engineers and scientists who design robots and interactive software. The
term comes from the dip in a graph with two parameters: psychological
familiarity and human-likeness.
As human likeness increases, so does our familiarity, or affection, for an
artefact. As soon as the resemblance becomes too acute, though, familiarity
drops below zero – hence the “valley”.
Making robots look human is a major goal in robotics, and a key feature of
science fiction. A world of androids is how science-fiction writers and many
robotic engineers envision the future. The “uncanny valley” could
potentially spell the end to that dream. If our mechanical doubles make us
paranoid, we may need to think again.
Researchers have tried to find the cause of the uncanny valley. One of the
most interesting insights has come from an international team led by Ayse
Pinar Saygin of the University of California, San Diego (UCSD).
Saygin and her team conducted an experiment scanning the brains of 20 subjects aged 20 to 36 while they were looking at three different things: a human, a mechanical-looking robot, and a human-like robot.
Interpreting the results from the fMRI scans, the researchers suggested that the cause for the valley is a mismatch between at least two neural pathways: that of recognising a human-like face and that of recognising different kinds of movement. These pathways meet on the parietal cortex of the brain.
There, information from the visual cortex relating to bodily movement is integrated with information from the motor cortex that contains mirror neurons, the brain cells that register that what we are seeing is “one of us”. Alarm bells go off in the brain when there is a perceptual conflict between the human-like features of the robot and its inhuman movement.
This mismatch creates a feeling of revulsion similar to what we feel when looking at a movie zombie. We instinctively expect human-like creatures to have human-like movements. As Saygin says: “The brain doesn’t seem selectively tuned to either biological appearance or biological motion per se. What it seems to do is look for its expectations to be met – for appearance and motion to be congruent.”
Interestingly, something similar to the uncanny valley occurs in one of the most bizarre neurological disorders, a condition called Capgras syndrome. The afflicted believe that their spouse, friend, or relative have been replaced with an impostor.
French psychiatrist Joseph Capgras first observed it in 1923, with the case of one Mme M, who was convinced that her husband had been replaced by a doppelgänger. It is most common in patients suffering from paranoid schizophrenia, and can be dangerous.
Some patients become so paranoid that they turn on their families. In 1986 a paper reported the case of a man who, thinking his father was a robot, decapitated him to search for batteries and microfilm in his head.
The neuroscientist Vilayanur Ramachandran, at UCSD, has hypothesised that Capgras syndrome is also caused by a neural mismatch. In his book Phantoms in the Brain, he suggests that information processed in the temporal cortex (next to the parietal cortex of the uncanny valley) comes into conflict with information processed in our limbic system.
The latter, residing at the base of our cranium, is where our most basic emotions reside – fear, love and disgust. As with the uncanny valley, Capgras patients receive conflicting information from two neural pathways. The person whom they see in front of them is recognised by their temporal cortex as their father, say, but their limbic system does not signal the familiar emotional response. The “father” is therefore not “real”, so he must be an impostor, an android, a double from another planet.
This connection between Capgras syndrome and the uncanny valley may have a profound effect on the way AI develops. The famous Turing Test, in which a machine is quizzed by a human and passes if its interrogator does not realise it is a computer, blurs the borders between “real” and “artificial” on the basis of an emotional perception from a human observer.
If the human observer feels that the machine in the other room responds like a human, then it must be intelligent. Philip K Dick, the science fiction writer, took the Turing Test to a more disturbing level: paranoia about the “mechanical other”.
Dick anticipated the discovery of the uncanny valley. Rick Deckard’s task in the 1968 novel Do Androids Dream of Electric Sheep? (adapted into the film Blade Runner in 1982) is to decide if his love interest is a “real” human. And after he discovers that she is not, a new dilemma kicks in: can he love a “replicant” anyway?
Neuroscience and sci-fi point to the elephant in the room of robotic research. Humans are not straightforward, logical creatures but primates that evolved haphazardly over millions of years. Between 60,000 and 40,000 years ago, our species made a huge cognitive leap, the so-called “big bang of the modern mind”. We began to create art and bury our dead as if they lived forever.
We are the descendants of these first truly modern humans. Researchers believe that the modern mind came into being when neural pathways in the brain became integrated, probably thanks to the evolution of language. Our cognitive evolution was bottom-up, not top-down. Mutated genes and language gave us consciousness and intelligence. Robotic evolution challenges this kludge-filled mental “software” of ours. The uncanny valley seems like the edge of a cognitive horizon, beyond which our primal fears kick in. So is this the end of robots as we have dreamt them? Are our primate brains unable to cope with mechanical doubles?
Perhaps not. It may just be a temporary phenomenon, highlighted by machines such as the Geminoid F robot, created by Prof Hiroshi Ishiguro of Kyoto University. His robots have human-like bodies but their movements, although impressively human-like, carry a trace of the mechanism beneath their “skin”.
This technological mismatch is what confuses our neural pathways and causes us a Capgras-esque revulsion. However, once movements become even more human-like, the familiarity graph rises again from the uncanny valley; acceptability returns steeply to normal. We seem to be at ease with androids that have human bodies and human movements, even if we know they are not human. As we traverse the uncanny valley another basic instinct comes into play: empathy.
To turn the idea of human-like robots on its head, consider Stephen Hawking, a human who speaks with a robotic voice. Speech engineers had suggested to Prof Hawking that he replace the voice synthesizer with a more advanced version. He declined; his mechanical voice had become a part of his personality. We cannot imagine him speaking in any other way; it is possible to mix human and mechanical characteristics without getting trapped in uncanny valley. Eventually, human-like robots will make us love them too.
George Zarkadakis is a science writer, novelist, AI engineer and web entrepreneur. Twitter @zarkadakis
http://www.telegraph.co.uk/
Saygin and her team conducted an experiment scanning the brains of 20 subjects aged 20 to 36 while they were looking at three different things: a human, a mechanical-looking robot, and a human-like robot.
Interpreting the results from the fMRI scans, the researchers suggested that the cause for the valley is a mismatch between at least two neural pathways: that of recognising a human-like face and that of recognising different kinds of movement. These pathways meet on the parietal cortex of the brain.
There, information from the visual cortex relating to bodily movement is integrated with information from the motor cortex that contains mirror neurons, the brain cells that register that what we are seeing is “one of us”. Alarm bells go off in the brain when there is a perceptual conflict between the human-like features of the robot and its inhuman movement.
This mismatch creates a feeling of revulsion similar to what we feel when looking at a movie zombie. We instinctively expect human-like creatures to have human-like movements. As Saygin says: “The brain doesn’t seem selectively tuned to either biological appearance or biological motion per se. What it seems to do is look for its expectations to be met – for appearance and motion to be congruent.”
Interestingly, something similar to the uncanny valley occurs in one of the most bizarre neurological disorders, a condition called Capgras syndrome. The afflicted believe that their spouse, friend, or relative have been replaced with an impostor.
French psychiatrist Joseph Capgras first observed it in 1923, with the case of one Mme M, who was convinced that her husband had been replaced by a doppelgänger. It is most common in patients suffering from paranoid schizophrenia, and can be dangerous.
Some patients become so paranoid that they turn on their families. In 1986 a paper reported the case of a man who, thinking his father was a robot, decapitated him to search for batteries and microfilm in his head.
The neuroscientist Vilayanur Ramachandran, at UCSD, has hypothesised that Capgras syndrome is also caused by a neural mismatch. In his book Phantoms in the Brain, he suggests that information processed in the temporal cortex (next to the parietal cortex of the uncanny valley) comes into conflict with information processed in our limbic system.
The latter, residing at the base of our cranium, is where our most basic emotions reside – fear, love and disgust. As with the uncanny valley, Capgras patients receive conflicting information from two neural pathways. The person whom they see in front of them is recognised by their temporal cortex as their father, say, but their limbic system does not signal the familiar emotional response. The “father” is therefore not “real”, so he must be an impostor, an android, a double from another planet.
This connection between Capgras syndrome and the uncanny valley may have a profound effect on the way AI develops. The famous Turing Test, in which a machine is quizzed by a human and passes if its interrogator does not realise it is a computer, blurs the borders between “real” and “artificial” on the basis of an emotional perception from a human observer.
If the human observer feels that the machine in the other room responds like a human, then it must be intelligent. Philip K Dick, the science fiction writer, took the Turing Test to a more disturbing level: paranoia about the “mechanical other”.
Dick anticipated the discovery of the uncanny valley. Rick Deckard’s task in the 1968 novel Do Androids Dream of Electric Sheep? (adapted into the film Blade Runner in 1982) is to decide if his love interest is a “real” human. And after he discovers that she is not, a new dilemma kicks in: can he love a “replicant” anyway?
Neuroscience and sci-fi point to the elephant in the room of robotic research. Humans are not straightforward, logical creatures but primates that evolved haphazardly over millions of years. Between 60,000 and 40,000 years ago, our species made a huge cognitive leap, the so-called “big bang of the modern mind”. We began to create art and bury our dead as if they lived forever.
We are the descendants of these first truly modern humans. Researchers believe that the modern mind came into being when neural pathways in the brain became integrated, probably thanks to the evolution of language. Our cognitive evolution was bottom-up, not top-down. Mutated genes and language gave us consciousness and intelligence. Robotic evolution challenges this kludge-filled mental “software” of ours. The uncanny valley seems like the edge of a cognitive horizon, beyond which our primal fears kick in. So is this the end of robots as we have dreamt them? Are our primate brains unable to cope with mechanical doubles?
Perhaps not. It may just be a temporary phenomenon, highlighted by machines such as the Geminoid F robot, created by Prof Hiroshi Ishiguro of Kyoto University. His robots have human-like bodies but their movements, although impressively human-like, carry a trace of the mechanism beneath their “skin”.
This technological mismatch is what confuses our neural pathways and causes us a Capgras-esque revulsion. However, once movements become even more human-like, the familiarity graph rises again from the uncanny valley; acceptability returns steeply to normal. We seem to be at ease with androids that have human bodies and human movements, even if we know they are not human. As we traverse the uncanny valley another basic instinct comes into play: empathy.
To turn the idea of human-like robots on its head, consider Stephen Hawking, a human who speaks with a robotic voice. Speech engineers had suggested to Prof Hawking that he replace the voice synthesizer with a more advanced version. He declined; his mechanical voice had become a part of his personality. We cannot imagine him speaking in any other way; it is possible to mix human and mechanical characteristics without getting trapped in uncanny valley. Eventually, human-like robots will make us love them too.
George Zarkadakis is a science writer, novelist, AI engineer and web entrepreneur. Twitter @zarkadakis
http://www.telegraph.co.uk/