Standing
before a room of middle and high school students in Bloomington for the
National Science Olympiad tournament two weeks ago, Karl MacDorman, a
new associate professor for Informatics at IU-Purdue Univeristy at
Indianapolis, played a video of an android head. The disembodied head's
eyes rotate and its mouth chews as the students watching squirmed
uncomfortably.
"What's your impression?" MacDorman asked the students.
"Eeeeew! Creeepy!" they said.
This eerie phenomenon, MacDorman said, is called "the uncanny valley."
As
more humanlike robots are built, people respond positively because the
robots become more and more familiar, he said, as humanlike
characteristics stand out. But because of our "strong expectations of
what is human," MacDorman said, if the robot becomes "almost human,"
the response quickly turns negative and we notice the non-human
characteristics more clearly.
"If a robot walks around with
knees bent like this," MacDorman said while demonstrating the position,
"you think, well this is just a robot. But if something that looks like
a human being does it, you think that's odd, that's unnatural."
Many
theories attempt to explain why the effect occurs. For example,
MacDorman said, if you see a tree with blight, you don't say "eeeew,"
but if you see a human being with leprosy, it reminds you of a disease
and triggers a negative response. Also, he said some people believe
human-like androids resemble corpses.
"It could be that (very
human-like androids) are a reminder of our mortality," MacDorman said.
"(It can be) a reminder that we're all going to die, and it can be
quite disturbing for that reason."
Robotics researcher Masahiro
Mori first described the "uncanny valley" effect in 1970. Now that
androids, as well as computer-generated characters in video games and
movies, are becoming more realistic, it is a topic not only for
discussion but also for policy for 3-D animation studios like Pixar.
"Part
of the success of 'The Incredibles' was that Pixar made a conscious
choice to give the characters a cartoonish style," said Don Strawser,
who teaches T160, the History and Social Impact of Video Games, at IU .
The
"uncanny valley" effect is also a problem in video games, Strawser
said, because game makers try to make human characters as realistic as
possible without pushing it so far that the users will reject it.
"This
is really hard because you would have to animate every human gesture
and it's just not practical," he said. "Yet we don't want our cute and
cuddly cartoon world to turn into a soulless, zombie land."
Companies
like Pixar might have a policy about how to avoid uncanniness, but the
effect has not been widely tested experimentally, MacDorman said.
That's where his research comes in.
"Android science is the
idea that very human-like androids can be used in experiments to find
out what kinds of behavior are perceived as human," MacDorman said.
Evolution
certainly plays a role in our ability to determine humanness, he said,
and "we are very sensitive to what is human," particularly in facial
and body proportions.
For example, a one-millimeter adjustment
in eye width can mean the difference between seeing a person as ugly or
beautiful, he said. Other studies show that men prefer women with a two
to three ratio of waist to hip -- and women who have that ratio tend to
be the most fertile.
However, we are probably not as sensitive
to humanoid robots, he said. Though humanoid robots may have a head,
arms, legs or even eyes and mouth, only androids have human details
like skin, eyelashes and teeth.
"So, does appearance matter?" MacDorman asked as he showed the students a video of a robot moving with and without skin.
"People
have more sympathy for an android than they do for a mechanical-looking
robot ... It reminds you of yourself, your personal identity, but at
the same time it's a machine," he said. "People will turn off
characters they see on the T.V., but not want to turn off the lights in
a room with an android."
He and his colleagues have also shown
that realism is due to "non-conscious movements" -- the breathing
patterns and miniscule adjustments people make without realizing it.
"Even
when a person is doing absolutely nothing, they're never perfectly
still," he said. "In fact, if you held yourself perfectly still you
couldn't see -- everything would go optic-violet. People who are very
good at meditating have discovered that. Our whole visual system
depends on slight movements of the body."
To see if people could
be fooled into believing an android was human, MacDorman's research
team placed a seated android nine feet away, drew a curtain back for
two seconds, then asked participants what they saw.
Only 23
percent of people were fooled by the android, and these tended to be
older people, MacDorman said. But when the same experiment was run with
the android making "non-conscious movements," 70 percent believed the
android was human.
"In a two second period, most people were
fooled ... but at least it shows the importance of movement in creating
the impression of presence," MacDorman said.
The fact that older
people have more difficulty telling android from human was a problem
for his team when they presented a new android at the 2005 World Expo
in Aichi, Japan.
"Some older people, maybe 70 or 80 years old,
were asking where is it? Where's the android? Because they thought it
was a person," he said.
And extending that two-second period for
the 70 percent of people who were fooled is extremely difficult,
MacDorman said, partially because it is difficult to design and create
smooth movement.
"The motion needs to look humanlike, but the joint is not the same and not human underneath," MacDorman said.
Whereas
humans have 244 degrees of movement in their arms, current android
prototypes only have 43 degrees. Using air actuators in the joints
creates some of the best motion, but requires a compressor and pump
connected in another room in order to control the android.
It's
also a matter of cost. A few companies now make androids with realistic
skin and motors with smoother movements and facial expressions for
around $400,000, but initial prototypes cost around $1 million to make.
Another
experimental use for androids is to study the nuances of human
interaction. Making the robot appear more human not only gives it a
physical presence, but also a social context, he said. In a recent
experiment on gaze, MacDorman studied how people respond with their
eyes in conversation when they believe the android is autonomous.
Whereas
people in Europe and North America mostly look up and to the right when
thinking about what to say to another person, people in Japan usually
gaze downward.
"By world standards, the Japanese make a lot of
eye contact, about 30 to 40 percent of the time, although this number
is low compared to Europe and Asia," he said.
If the person in
the experiment believed the android to be self-controlled, the person
averted their eyes less often and made more eye contact with the
android. If the person in the experiment thought the android was
human-controlled, the subject responded more naturally by looking
downward while thinking of a response.
"This is the first time
breaking of gaze has been found to be related to what you believe is on
the other end -- human or robot," MacDorman said. |