Karl F. MacDorman &
Hiroshi Ishiguro talk with ScienceWatch.com and
answer a few questions about this month's Fast Breaking
Paper in the field of Social Sciences, general. The
authors have also sent along images of their
work.
Article Title: The uncanny advantage of using
androids in cognitive and social science
research
Authors:
MacDorman, KF;Ishiguro, H
Journal: INTERACT STUD
Volume: 7
Issue: 3
Page: 297-337
Year: 2006
* Indiana Univ, Sch Informat, IT 487,535 W Michigan St,
Indianapolis, IN 46202 USA.
* Indiana Univ, Sch Informat, Indianapolis, IN 46202
USA.
* Osaka Univ, Grad Sch Engn, Dept Adapt Machine Syst,
Suita, Osaka 5650871, Japan.
Why do you think your paper is highly
cited?
This target article with commentaries and rebuttal is seminal in the
establishment of android science as a field of inquiry. It makes the
strongest case yet for android science. Android science posits a
synergistic relation between developing very humanlike robots and
understanding human interaction and the processes behind it. Android
science provides synthetic and analytical methods to understand mechanisms
underlying interaction.
To make androids humanlike, we must study human interaction, and to
evaluate models of human interaction fully, we need to implement them in
androids. This lets us determine whether the models are "true to life" in
interactions with real people. By using androids instead of other kinds of
robots, we control for the unknown influence of nonhuman appearance on
interaction. This enables us to concentrate on the microdynamics of human
interaction, including motion quality, contingency, and learning, instead
of the appearance of the robot.
Does it describe a new discovery, methodology, or
synthesis of knowledge?
What's new about android science is its methodology: An android is used as
an experimental apparatus in psychological or neuroscientific experiments
to test hypotheses about human brains and interaction—or the android
acts as a testbed in which cognitive models are implemented and tested by
observing the android's interactions with people. Both of these approaches
demonstrate how androids can be used to test and refine social, cognitive,
and neuroscientific theories. Thus, the focus is on human beings, not
robots.
In addition to being a position paper, the article reviews the literature
on android science, which is still scant, and presents some empirical
findings on the uncanny valley. The robotics professor Masahiro Mori
graphed what he considered to be the relation between human likeness and
perceived familiarity: familiarity increases with human likeness until a
point is reached at which subtle imperfections cause a robot to appear
creepy. He called this the uncanny valley. According to Mori, movement
amplifies the effect.
So the uncanny valley is the phenomenon of "not quite" human entities
appearing creepy. Although there was a strong dogma against building
androids because of the uncanny valley, it turns out to be useful in
android science. A human form makes nonhuman responses observable, because
they look odd. That lets us know how to refine our models. A
mechanical-looking robot just isn't good at this.
Many people have been duplicating the uncanny valley figure. We recently
learned it had appeared on a television program called 30 Rock. A
student also posted to YouTube one of our presentations on the uncanny
valley.
Would you summarize the significance of your paper
in layman's terms?
Androids offer special advantages over human actors in helping to refine
theories about human interaction. For example, in reviewing a paper we
wrote with Shoji Itakura, Takashi Minato, and Stephen Cowley, "Assessing
Human Likeness by Eye Contact in an Android Testbed" in Proceedings of
the XXVII Annual Meeting of the Cognitive Science Society, Stresa,
Italy, 2005, we discussed an experiment in which an android sat with a
participant and asked him or her several questions, some of which required
thinking. People tend to break eye contact during conversations if they
must think about a response. How participants broke eye contact in the
experiment depended on whether they thought the android was controlled by a
human or acting of its own accord.
When Japanese participants thought the android was under human control,
they showed modesty by looking down—just as they would when
interacting with a real person. It was the first time anyone has shown that
what they believe about the mind of the other entity—human or
machine?—influences how they break eye contact. The results also show
the importance of cultural norms in automatic, nonverbal behavior. Previous
theories about why we break eye contact while thinking completely missed
this effect.
In addition, the paper explores alternative explanations of the uncanny
valley, from evolved cognitive mechanisms used in avoiding disease and
selecting mates to the social construction of personal and human identity.
The article also presents evidence that an uncanny android can produce a
terror management response. Each of us possesses a worldview that helps to
give our transitory lives meaning. If an uncanny android subconsciously
reminds us of our mortality, it can create a strong aversion to people who
threaten our worldview. In this sense, an uncanny android has something in
common with the terrorist attacks of September 11th, 2001.
Prof. MacDorman, how did you become involved in
this research with Prof. Ishiguro, and were there any problems along
the way?
My research at Cambridge in the mid-1990s concerned the problem of symbol
grounding and symbol emergence. Robots seemed like an obvious testbed for
exploring cognitive models of how infants and nonhuman animals develop
internal representations. However, I was also concerned with how children
come to use the external symbols of human language. When I met Hiroshi
Ishiguro (then at Kyoto University) and Minoru Asada at Osaka University in
1997, this seemed like a natural area in which to explore human-robot
interaction. Hiroshi Ishiguro proposed controlling for the effects of
appearance by using androids.
Android science is a new interdisciplinary framework for studying human
cognition and interaction based on the finding that humanlike robots can
elicit the sorts of responses people direct toward each other. As a result
androids can be used as stand-ins for human participants in social,
psychological, cognitive, and neuroscientific experiments. This brings
several advantages. First, as an experimental apparatus, an android can be
more precisely controlled than a human actor. Second, unlike a video or
computer simulation of a human being, an android has physical presence.
Third, in comparing human-human and human-android interaction, an android
controls for the effects of appearance. For many experiments, an android
offers a good balance between experimental control and ecological validity
in supporting humanlike interaction.
There is always difficulty in conducting research in android science,
because androids have not reached the stage of being reliable, mass
produced products. In addition, experiments often require the integration
of many different kinds of equipment, including sensor networks, image
processing, voice recognition, and so on. Robotics research in the United
States is not well funded, but that will surely change as funding bodies
come to realize the enormous potential of android science.
Where do you see your research leading in the
future?
We have both been fascinated by the relation between the microdynamics of
human interaction and symbol emergence, relationship formation, and
cultural norms. One of us (Karl MacDorman) has recently received an IUPUI
Signature Center grant to start an Android Science Center in Indianapolis
along with our collaborator Chen Yu—the founder and leader of The
Computational Cognition and Learning Lab of The Department of Psychological
and Brain Sciences at Indiana University, Bloomington.
We are presently developing an android baby with anatomically correct
muscle-joint relations and opposing-pairs of elastic muscles to simulate
the appearance and movement of a human baby. We will use the android baby
and a similarly-sized robot to study how appearance influences timing and
contingency in infant and adult interactions.
Do you foresee any social or political implications
for your research?
Through hypothesis testing and cognitive modeling, android science can
contribute to basic science, such as our understanding of the perception of
human and nonhuman forms. However, it also results in principles for
interaction design. How do we create robots and computer graphics
characters that are more responsive and appealing? And the end-product of
this research—besides the basic science and design
principles—is a physical artifact that mimics human appearance,
dynamics, and interactivity as closely as current science and technology
allow. Such a device has myriad applications that would impact society.
"To make
androids humanlike, we must study
human interaction, and to evaluate
models of human interaction fully,
we need to implement them in
androids."
Hiroshi Ishiguro is already using his
Geminoid double to let him remotely sit in on
meetings. Our
collaborators at Kokoro Co., Ltd. have built an
android receptionist and museum guide.
Androids are useful for training and assessment, for example, as a
realistic replacement for medical dummies, with their ability to simulate
cardiac arrest or other medical conditions, such as changing speech,
behavior, and vital signs. Other applications include the socialization of
children with autism or other disabilities and training social workers and
counselors.
With the retirement of the baby boomers, the United States, Europe, and
especially Japan are faced with a labor shortage in healthcare and
eldercare. Androids can encourage exercise to fight obesity and coronary
heart disease and assist with rehabilitation after a stroke. In addition,
they can provide cognitive stimulation, to keep the mind nimble as we age.
And finally, they can be a source of companionship, alleviating loneliness.
However, as we better understand nonverbal behavior, androids could be used
to manipulate human judgment. Overly compliant androids could also feed our
narcissism. These could be ethical concerns in the future.
There is much to be said for the challenges posed by human relationships.
They may make us into better people in the long run.
Karl F. MacDorman, Ph.D. (Cambridge)
(Web)
Associate Professor, Human-Computer Interaction Program
Indiana University School of Informatics
Indianapolis, Indiana, USA
and
Adjunct Professor, Department of Electrical and Computer Engineering
Purdue University School of Engineering and Technology
Indianapolis, Indiana, USA
Hiroshi Ishiguro, Ph.D. (Osaka)
Professor, Department of Adaptive Machine Systems
Graduate School of Engineering
Osaka University
Osaka, Japan (Web)
and
Visiting Group Leader of the Department of Communication Robots
ATR Intelligent Robotics and Communication Laboratories
Kyoto, Japan (Web)
Note: links to external sites will open in a new browser
window. Clarivate and ScienceWatch.com does not endorse
external sites.
Keywords: androids, android science, new
interdisciplinary framework for studying human cognition and interaction,
humanlike robots, microdynamics of human interaction, human-robot
interaction, the uncanny valley, Masahiro Mori, human likeness, perceived
familiarity, human form, nonhuman responses, assessing human likeness, why
we break eye contact, android testbed, cultural norms, nonverbal behavior,
evolved cognitive mechanisms, avoiding disease, selecting mates, social
construction, personal and human identity, uncanny android, mortality,
symbol grounding, symbol emergence, cognitive models, infants and nonhuman
animals, develop internal representations, stand-ins for human
participants, social, psychological, cognitive, and neuroscientific
experiments, robots, computer graphics characters, Hiroshi Ishiguro,
Geminoid double, android receptionist, android museum guide, useful for
training and assessment, realistic replacement for medical dummies,
socialization of children with autism or other disabilities, training
social workers and counselors, overly compliant androids, our narcissism,
ethical concerns, challenges posed by human relationships.