2003年度 学術交流支援資金 活動報告書
プロジェクト番号：1-8
研究課題名：Social Affect Domain of Technology:
Entertainment Robots as a Possible Interactive Social Partner and Learning Tool
in the Home, School and Clinical Setting
研究代表者：徳田英幸 所属・職名：政策・メディア研究科／教授
Abstract
In this study we explore the child's
behaviors when playing with different types of entertainment robots. We are
interested on whether the different contingency levels and anthropomorphic
features of the robotic toys will have a difference on their behavior, animate/inanimate
understanding, and the child’s perception of intelligence toward these
artifacts. Preschool age children and college students will be shown three
robotic toys at once, each performing a different action. A series of forced
choice questions are asked that focus on the perception and interpretation of
the robot’s intelligence. We are also interested if the adult and child’s
perception of intelligence correlates with other general schemas such as
animate and biological understanding.
Introduction
Robotic technology has been a great success in
the industrial domain, encouraging the field to venture out into the private
sector. Home network appliances through ubiquitous computing are introduced
into the home, connecting major household appliances. Companies are starting to
produce affordable personal robots that are to manage these appliances, as well
as become social companions, and personal entertainment robots within the home.
The exhibit at the Robodex2003 conference showed that the animated future world
of "The Jetsons" with clean up robots, baby
sitter robots, robot pets, robot tutors and robot security guards in the home,
may not be as far into the future as we think. As prototype products are coming
out, very little research examines what impact these new technological
artifacts have on the kind of social relationship or feelings they might elicit
from humans, and how humans begin to adapt these types of objects into their
everyday experience.
Given where technology is now, what do entertainment/personal
robots contribute in developing social interaction, and social affection
between the child? There are two objectives to my research interest. One, is my
intent to study the social affect domain of technology and the impact it may
have on the home, school and clinical setting. The second objective is to
explore the notion of intelligence in these entertainment/personal robots,
since what differs greatly between entertainment and industrial robots is the
belief that a high level of intelligence is favorable when interacting with
humans.
Objective
1: Social Affect of Domain of Technology
As entertainment
robots have started to appear on the market, very few research question the
kind of social relationship or feelings that may come about from interacting
with them. On the other hand how could the robots make use of such relationship
built with humans?
In
this study we are interested in looking at the nature of the child and their
interaction with entertainment/personal robots, to see if there are components
that lead to a social-like relationship. We are interested in the child’s
perception of entertainment robots whether they see it as a machine, a
creature/artificial life capable of social affection.
More
specifically, we are interested in whether the difference in the robot’s level
of contingency and response, and the varied anthropomorphic features will
affect the children's behavior toward toys and their understanding of these
objects as animate or inanimate objects.
Through questions asked during the
two play sessions, we will explore how the contingency level and
anthropomorphic features of each toy affects the understanding of the object
as: 1) biological property 2) an agent, 3) a potential companion 4)Teach/learning
capabilities. Through the observational data, we will code the behavior and
interaction between the child and toy, to see different inner qualities in the
interaction that may lead to the development of social affect toward these pet
robots.
Objective 2: Perception
of Intelligence
As my second objective in researching this
topic, is to examine whether
intelligent features are noticed, or if it even matters. Research says that
humans tend to attribute intelligence to machines even when they are not. This
study explores if there is a difference in how children perceive and interpret
this so called “intelligent” feature in these new technological artifacts. The
difference in perception may have an affect on social relationships or
understanding they elicit with such technology.
In this study we explore if young children’s
perception of intelligence differ toward robotic toys that have specific
contingency levels and anthropomorphic features. We are also interested whether
children’s perception and interpretation of intelligence will correlate with
other general schemas such as biological/physical substrate, 2) agency 3) a potential companion, 4) Teach/learning
capabilities.
Methods, Techniques, or Modes of Inquiry
Participants
The study was
conducted at a
Materials
Three entertainment
robots (ERS-220A, ERS-210, ERS-311) were generously donated from Sony
Entertainment of America (
Study Design and Procedure
This study included a three-by-three
mixed model design (Table 1). The within-subject comparison was Action (dance,
kick, still), and between-subject comparisons were age (younger, middle, older).
The entertainment robots are differentiated by the action they perform. The
order of action will be held constant while the robots will be counter-balanced
because of the different anthropomorphic features.
Table 1: Three-By-Three Mixed Model
Design
Action
(Within-Subjects)
|
Age
(Between-Subjects) |
Dance
Action |
Kick
Action |
Still
No Action |
|
Young
Group (37-48months) |
|
|
|
|
MiddleGroup (49-55months) |
|
|
|
|
Older
Group (56-61 months) |
|
|
|
Entertainment Robot Action:
Robot Dancing Action A : Robot plays music and dances.
Robot Kicking Action B: Robot perceives ball, approaches ball,
kicks ball.
Robot No Action C: Robot does nothing,
Procedures
In this study,
we examine 1) how young children's behavior toward toys, and to explore their understanding
of these objects as animate or inanimate objects, and 2) how children differ in
perception and interpretation of intelligence, preschool children were asked to
participate in a one-on-one videotaped session. During the session, several entertainment
robots that differ in contingency and anthropomorphic features perform different
action or no actions in front of the child. When the child enters in the room,
they will be given some time to settle in. The participants are then shown three
entertainment robots (two performing actions, one without actions) separately.
The experimenter then asks 14 questions in relation to the entertainment
robots. When the questions are completed, the experimenter notifies the participant
that the session is completed and is accompanied back to the classroom.
Measures:
In this study, we have four dependent
measures, 1) Open ended questions, to check what the child perceived and
interpreted from the action 2) Intelligence questions, which is a forced choice
question that addresses the intelligence potential of each robot. The questions
address the robots ability to initiate action to perceive, act based on their
perception, and changing their environment, 3) general schematic questions,
that explores the child’s understanding of these objects as a biological
entity, agent, animate/inanimate, a potential social companion, and 4)
background information.
The description
of the questions and examples are shown below.
Open-ended
Questions: to check what the child perceived and
interpreted from the action.
Examples:
“What did the doggy just do?”
Intelligence Questions: To check if child perceives the
features associated with intelligence.
Examples: “Do
you think any of these dogs will be able to tell the difference between a real
and a toy bone?”, “If you grab a leash, do you think any of the dogs will know
that its time to go for a walk?”
General
Schematic Questions: Questions that relate to physical biological
entities, agency, affinity and animacy.
Example: “Do
you think any of these dogs can grow?”(biological)
”If I forgot the remote control today, do you think any of the dogs would do
anything?”(agency)
“Which dog
would you like to show your parents?”(favorite, affection)
Background
Question:
Example: “Have
you seen a dog like this before?”
Exploratory Question on Teaching and Learning
To explore the possibility of the role
entertainment robots may play in the school setting, we explored the impact it
may have on learning. It is in our interest to see what aspects of the actor’s
(e.g. child, teacher) social interaction with the toy or object inspires the
foundational domains leading to an intrinsically motivating learning process.
Example: “Do you think somebody can teach these
dogs to shake hands?” “How would you teach them?”
The
score for each of the variables; Perception of Intelligence, Biological
attribute, Agency Attribute, Favorite/Preference Attribute, Teach/Learning
variable, was obtained by how the child ranks the robots for each of the
questions. For example, if asked “Do you think any of these dogs are smart?,
and if the child replies “yes”, we ask “which one?”. The first robot that the
child points to, will be given the value of 3, the second 2, and the third 1.
If the child says only one robot, is smart, that robot will be given the value
of 3, and the rest will be given the value of zero. If the child says that none
of them are smart, all the robots will be given the value of zero.
Results
Presently, we are still at the
level of preliminary analysis, but have found promising results. The tables
below (Table and Graph 2, 3, 4), show the mean of each action by age. In the
Table and Graph 2, we can see that when seeing the dance action, younger
children are more likely to attribute biological properties than middle/older
children. While middle/older children show a variation in how they attribute
various schemas to the dance action, younger children do not vary much across
the different variables.
Table
and Graph 2
Mean
Table for the Dance Action by Age
|
|
Young |
middle |
older |
|
Intelligence |
1.7 |
1.9 |
2.2 |
|
Biological |
1.7 |
1 |
1.4 |
|
Favorite |
1.6 |
2 |
2.3 |
|
Agency |
1.5 |
1.4 |
1.3 |
|
Teach |
1.8 |
1.3 |
1.9 |
In Table and Graph 3,
which looks at the kick action across age, you can see that all children
young/middle/older similarly attributed intelligence to the Kick action.
However, when it comes to biological properties there is a great difference between
younger and middle/older children. Again, for the action Kick, younger children’s
attribution across the variables, differ less compared to middle/older
children.
Table
and Graph 3:
Mean Table for the Kick Action
by Age
|
|
Young |
middle |
older |
|
Intelligence |
1.9 |
2 |
2 |
|
Biological |
1.8 |
1 |
1 |
|
Favorite |
2.1 |
1.8 |
1.4 |
|
Agency |
1.6 |
1.4 |
0.89 |
|
Teach |
1.6 |
1.5 |
1.2 |
Table and Graph 4:
Mean
Table for the Still/No Action by Age
|
|
Young |
middle |
older |
|
Intelligence |
1.9 |
2 |
2 |
|
Biological |
1.8 |
1 |
1 |
|
Favorite |
2.1 |
1.8 |
1.4 |
|
Agency |
1.6 |
1.4 |
0.89 |
|
Teach |
1.6 |
1.5 |
1.2 |
Compared to the other two tables, in Table and Graph 4, you can observe
that the mean score for the robot with Still/No action, there is little
difference between age, as well as across the different variables. From this,
one can suggest that children have been making judgments based on the robot’s
action, and not just based on the robots anthropomorphic features.
Taking a more closer look at
the 14 questions (3 Intelligence, 3 Biological, 3 Favorite, 3 Agency, 1
Teach/Learn, 1 experimental question), we conducted a data reduction analysis
to see whether children associate certain questions with different variables.
In doing so, we ran a principle component analysis separated by action to see
if certain questions in a given variable correlate across to different
variables and whether they can be grouped into smaller number of composite
variables.
Action: Dance
For the action Dance, a
principle component analysis was run using the 3 questions from each of the
variables; Intelligence, biological, favorite, and agency, plus one additional
teach/learning question and 1 experimental question totaling to 14 questions.
Although the KMO (.578) may be inadequate, the
The total variance shows that
there are four groups extracted with eigenvalues greater
than 1. Looking at the cumulative percentage, it seems that over 2/3 of the
variance is accounted for by the four factors. The analysis shows
that the 14 questions can be sorted into 4 groups.
The four groups are sorted out with the following questions.
Principle
Component Analysis for Action: Dance
|
Group 1 |
2 Favorite questions, all Intelligence
questions |
|
Group2 |
2 Agency questions, 2 biological questions |
|
Group3 |
1 Favorite, Teach/Learning question |
|
Group4 |
1 Agency question, 1 biological, 1 experimental
question |
When looking at the four groups, and examining the content
of the items, we found that for the action Dance, children tend to associate
their favorite robot with Intelligence, Biological properties with Agency. One
can assume the possibility that for the action Dance, children attribute
intelligence to their favorite robot, and attribute agency to robots which they
see as having biological properties.
Action: Kick
For the principle
component analysis action Kick, we again took the 3 questions from each of the
variables; Intelligence, biological, favorite, and agency, plus one additional
teach/learning question and 1 experimental question totaling to 14 questions.
In this analysis, the KMO (.793) was adequate, as well as the
The total variance shows that
there are four groups extracted with eigenvalues
greater than 1. Looking at the cumulative percentage, it seems that over 73% of
the variance is accounted for by the four factors. The analysis shows
that the 14 questions can be sorted into 4 groups.
The four groups are
sorted out with the following questions.
Principle
Component Analysis for Action: Kick
|
Group 1 |
All Favorite questions, 2 Agency questions |
|
Group2 |
All Biological questions, Teach/Learning
Questions |
|
Group3 |
All Intelligence Questions, 1 experimental
question |
|
Group4 |
1 Agency question |
When looking at the four groups, and examining the content
of the items, we found a different result from the action Dance. In action Kick,
children tend to associate their favorite robot with Agency. The children
seemed to be able to attribute Biological properties independently from all
other variables. The same results were seen with Intelligence. The children
seem to perceive the action Kick as a clear sign of intelligence, and enough
attribute biological properties to the robots. On the other hand, children seemed
to attribute agency to their favorite robot.
No Action: Still
For the principle
component analysis no action Still, we again took the 3 questions from each of the
variables; Intelligence, biological, favorite, and agency, plus one additional
teach/learning question, and 1 experimental question totaling to 14 questions.
In this analysis, the KMO (.700) was adequate, as well as the
The total variance shows that
there are four groups extracted with eigenvalues
greater than 1. Looking at the cumulative percentage, it seems that over 74% of
the variance is accounted for by the four factors. The analysis shows
that the 14 questions can be sorted into 4 groups.
The four groups are
sorted out with the following questions.
Principle
Component Analysis for No Action: Still
|
Group 1 |
1 favorite question, 2 biological questions, 2
intelligence questions, 1 experimental |
|
Group2 |
2 favorite, 1 teach |
|
Group3 |
1 biological, 1 intelligence, 1 agency |
|
Group4 |
2 agency questions |
When looking at the four
groups, and examining the content of the items, we found a more scattered
result from no action Still. In no action Still, the association that children
make seemed to be somewhat scattered, The scatter of the different variables
across the four groups may be due to the reason that the robot performed no
action, and therefore the children had no action to base their answers, leading
to a scattered grouping of questions across the different
variables.
Summary
In this study we have made the first attempt to look
at the nature of the child and their interaction with entertainment/personal
robots, and to see if the child’s perception of entertainment robots whether
they see it as a machine, a creature/artificial life capable of social
affection. The two objectives to this research are 1) the social affect domain
of technology and the impact it may have on the home, school and clinical
setting and 2) explore the notion of intelligence in these
entertainment/personal robots, since there is a belief that a high level of
intelligence is favorable when interacting with humans. We explored these objectives through
questions that relate to the entertainment/personal robot as a 1) biological
property 2) an agent, 3) a potential companion 4)Teach/learning capabilities.
We are also interested children’s perception and interpretation of intelligence
will correlate with such general schemas.
As
a result of our preliminary analysis, we found that children’s age may be an important
factor in how they perceive and interpret action. For the action Dance, younger
children are more likely to attribute biological properties than middle/older
children. While middle/older children show a variation in how they attribute
certain properties to the dance action, younger children are pretty much
constant across the variables. For the action Kick, you can see that children
of all age, similarly attributed intelligence to the Kick action. However, when
it comes to biological properties there is a great difference between younger
and middle/older children. Again, for the action Kick, younger children’s are
pretty constant across variables with little variation. For the no action
Still, all children were pretty close in that very few properties were
attributed to the no action Still.
Through principle component
analysis, we found that for the action Dance, children may attribute intelligence
to their favorite robot, and attribute agency to robots which they see as
having biological properties. For the action Kick, children seem to perceive
the action Kick as a clear sign of intelligence, and enough attribute
biological properties to the robots. Also we saw a slight pattern in that
children tend to associate their favorite robot with Agency when seeing the
action Kick. For the no action Still, the association that children make seemed to
be somewhat scattered, implying that children made little association between
the action and the question.
Since we are still at the preliminary analysis
stage, it is difficult to say what role entertainment/personal robots can play
in the home, school and clinical setting. However, what we can say is that the child’s
age, has an effect on how one can interpret robotic actions. In exploring how
certain actions create specific associations within a child, can contribute in
designing certain actions that will be fit for a given context or environment.
It is in our interest to continue this study and data analysis to filter out
how the different actions elicit different attributions in children and their
interpretation of various artifacts.
Acknowledgements from
the Research Group
Our greatest appreciation to Keio University
for the Kokusai-Kouryuu Shien
Shikin Grant, and providing us with the opportunity
to do continuing collaborative work with the students and faculty at Stanford
University. Also we would like to thank Sony Entertainment of America (