|
|
|
|
|
|
Audio Excerpt |
Computers as Social Actors Clifford Nass Professor Stanford University nass@leland.stanford.edu
Ted Selker: And thank you very much Ted. At the conclusion of
Ed Fredkin's talk, we will invite all these wonderful people that have been
speaking today up here and continue the discussion for another 45 minutes.
Now I would like to introduce Clifford Nass, a sociologist. Clifford Nass: Even sociologists can say something about computers.
The title of the talk is actually slightly different than the one Ted put
down because he encouraged me to make the strongest and most aggressive
case I could. So the title is "Computersare Social Actors." Not
as or like or some other metaphor. I want to mean this in the strongest
possible sense. And throughout the talk you will see how strong I mean it.
Two commercials, one is the research supported presented today is supported
by the Center for the Study of Language and Information under Industrial
Affiliates Program of which IBM is one member. And the other commercial
is my book that comes out next week describes this research so you should
all rush to the bookstore next week.The basic ideance and that's the top
one there. Individual's interaction with technologies is fundamentally,
emphasis on fundamentally, social and natural, and in a minute I'll define
what I mean by social and natural. I can refer to in questions. Second point
if you'll see is these responses are automatic and unconscious. Simply put,
all of you in the audience will deny that you would do when you would see
people like you that is experienced computer users do up here. The reason
you'll deny it is because these are responses that you're not consciously
aware of and that you couldn't control. So what do I mean by fundamentally
social. What I mean is go to the social science section of the library and
the argument of this talk is the people who know the most by far about human
computer interaction or social science. Unfortunately, none of them know
that. Little did they know that they had been spending all their time writing
deeply about human computer interaction and just were not aware of it. So
what I am going to try and do today is show you how or where things should
be. So what I'm going to do is I'm going to the social science section of
the library and pull out a journal and find some paper concerning humans
attitudes or behaviors towards other people. So that would be social psychology,
sociology, communication and other disciplines. Here is the best part, pull
out a crayon and wherever you see the word "human" scratch it
out and scribble in the word "computer" or "communication
technology." Do that in two different parts of the article. First in
the theory section, which describes the predictions that these social scientists
have worked long and hard to come up with about how people will treat other
people. What I want to argue is now you have the theory how people will
treat computers. Now anyone can make up a theory. The hard part is proving
it true. So go to the method section where psychologists and sociologists
worked long and hard figuring out how we could design an experiment that
would prove this and just mindlessly go through and cross out a person and
plop in a computer. Steal their measures, steal their statistics, steal
everything. A you are doing is plopping in a computer. We are not talking
here about fancy computers. You will see here that these are all text based
,no artificial intelligence extraordinarily simple computers, but plop them
in. Then essentially specify what social characteristics you're going to
manifest and see how pathetic the social characteristics we are talking
about here are. We arere not talking about pictures, human representations,.
full AI. Computers refer to themselves as this computer rather than I. They
don't give people a name. Strip out everything that you would think would
make it social. Then run the same experiment and demonstrate using statistics,
quantitatively that the exact same rules work. Then draw a conclusions,
first for design and then for methods and then for markets, which I will
not talk about today. Let me give you an example of this. Take politeness.
If you go to social psychology section of the library your going to find
the following rule: when individuals ask about themselves rather than others,
the responses are more positive and more homogeneous. Put in another way.
If I say to you, "How do you like my talk so far," presumably
you're going to say a nice thing to me. But you can whisper something less
nice to the person next to you. So there's a social politeness norm. The
homogeneity one, just imagine it as people talking to me will not give their
honest opinion, they will give the opinion I want to hear, but across the
audience there will be a wide range of opinions about my talk, hopefully
all positive. So what we do is we go ahead and we do this experiment with
computers. We take a text-based computer, have it teach a person, in this
case a simple tutorial system, and then have the computer ask questions
about how well the subject perceived the performance--how helpful were the
information it taught, how useful, how informative, etc. In one case the
computer that asks the questions is the computer people work with. The other
case the people used a different, physically identical but different computer,
on the other side of the room. These are all experienced computer users
who understand that this computer would feel no worse when it heard about
itself than when this one heard about this one. They all thought correctly
in fact that they were programmed by the same person. We then ask people
various questionnaire assessments on ten-point scales about how much they
liked the computer, how much fun it was to use, how intelligent, etc. So
paradigm here tutoring testing evaluation and questionnaire, language computer
used text it was interactive in a very minimum way and when it gave you
a fact it would say, "How much did you know about it before it before
it gave you the next fact. In fact everyone got the same information and
it was text based. So what happened? Amazingly, responses when people were
asked by this computer that they just worked with were significantly more
positive. Significance here means statistical significance and not a fluke.
They was also more homogeneous. We asked people who did the experiment if
they changed their answers to make that first computer feel better. Everyone
said, No don't be an idiot. No, only a moron would change their answers
to make them feel better, but that's exactly what they did. Not because
they are morons, but because they are humans and most importantly because
they are not evolved to twentieth century technology. And that is the point
of all our research. The human brain evolved in a world where something
filled the most minimal social queues. It used language, it interacted,
it filled the role. Our brain goes ahah, I know what to do with that--treat
it like a person. There is no on/off switch in a brain for media for computers
that says, "Computer don't do this stuff!" It's not the way human
brains are built. So what I going to do for the rest of the talk is show
more of these. Now hopefully its obvious that in a normal audience I would
go through all the different ways you would design if these things were
true. I'm going to assume for an audience like this you can deduce all the
designs implications having to do with politeness, crisis maxims, etc. If
not the book has them played out. Let me tell you about the other concepts
that I will be presenting. I could present a few more then if people have
questions about specific areas. Note that this is a rather strange way to
talk about human computer interaction. This looks more like you would talk
about human-human interaction. but that is exactly they way we should be
approaching human-computer interaction. So politeness--we can show that
people have personal distance queues. I will present flattery. We can show
that when a computer praises itself it is believed less than when a different
computer praises it. We can go through interface personality, imitation
when a computer changes to imitate you. People are flattered. They like
the computer more. Imitation is flattery. Just like with people. Blame,
that I will present, hopefully. Good-bad fundamental emotions, these are
the two basic emotions, and human-human interaction, negativity and arousal
they apply in human computer interaction. A specialist, simply by labeling
in this case even a television set, a specialist, people thought it was
more effective. Team mates I will talk about today. Gender female voice
computers or thought of as better teachers of love and relationship and
worse teachers of technical subjects then male voice computers, and perhaps
more depressing, people who are praised by male voice computers think that
they did better than people praised by a female voice computer. And this
is not my fault either. Multiple voices, two different voices on the same
computer are seen as two different social actors; therefore, when voice
number two praises voice number one people think the computer did better
than when voice number one praises itself. They also think its nicer. Vice
output/input--I will talk about. And these are various characteristics of
media presentations that effect people's perceptions, and I can go into
any of these if people are interested. I just want to give you a note here.
Let me cut to some more of the social rules now. Flattery, turns out people
are suckers for flattery. What that means is, when you were flattered, even
if you know you were flattered by a person, you will think you did just
as well as if you were sincerely praised. You will think the person is just
as good and nice, and you will like them just as much. Even though we all
deny it, we have terrible pejorative terms like brown nose, for people who
flatter us. Nonsense, we all love it. Conversely, unwarranted criticism
we are skeptics. When we get criticized, we evaluate it more carefully;
therefore, if I think the criticism is unwarranted I will reject it. This
psychologist called it Hedonecky symmetry, positive and negative are different.
You don't need to know that. What you need to know is that flattery is different
than criticism. So what we did was, we had people work with the computer.
In one case we told them that your evaluation will be based on a great deal
of study. The evaluation system is one of the best developed systems in
the world. In the other case we said we haven't had time... Good question.
In the case of flattery. Let me answer the question first in the case of
flattery. In the case of flattery, Western cultures are, I don't know the
literature on Eastern cultures. Some of the other concepts politeness is
a universal and some of the other ones do vary and this is important question--which
of these are cultural universals. I will argue that social responses to
technology are universal. The particular manifistationsdo differ. The flip
side is when you nationalize, when you design for particular countries,
your translation must include social translations as well as the more technical
translations for words. So the other case we tell people look have written
the evaluation software yet the evaluations you get are totally random.
And then people either praise or criticize everyone got it one to four cases
random definite, careful praise or criticism. Well, you probably guessed
flattery works just as well as true praise. Users thought that they did
better when they received no evaluation, and they thought they did just
as well as when quote and sincerely praised, their ---- was equal they liked
the computer as much they enjoy working with the computer as much and they
perceived performance the computers performance was perceived equal and
both was more positive than no evaluation at all. However, ---- criticism
just like with people is different than sincere criticism users perceive
performance as better for insincere criticism we reject criticism, we suck
up praise but reject criticism when its random. We also feel better about
it, we just we'll ignore it we don't like people who unwarranted we criticize
and we don't like either in performance or other things. This suggest that
an example a radical new spell checker. Spell checkers are bad thing as
for as flattery they only criticize and sometimes they criticize even when
you spell the word right, let's flip that around imagine a spell checker
that goes through that says ---, you spelled that correctly fantastic only
five percent of the country can spell -----, correctly and at the end it
says, "you're a much better speller than ever. ---- task people love
it, people love it, they think it just catches more long words, everything.
OK, right...ok next one, so know I going to give you a run through froming
running throughout the libraries wonderful libraries paper you could run
through and grab things. So know I'm going to run through the personality
psychology section. Now personality personalities computers is one of the
great holy grail since I gotten into the field people have said oh, computers
have personality but it so hard its artificial intelligence and complex
representations and all this really hard stuff. Well it turns out couldn't
be, because humans can assess the personality of other humans within a minute
or two with great accuracy and we can't be using them enormous data base
with which to do it. So it turns out that psychologist have been very nice
being at first the human computer interaction that they are, and listed
for us the criteria that manifested various personalities let me just tell
you a few things about personalities written these are all written by psychologist
two aspects of personality that are fundamental there called the interpersonal
complex, dominance versus permissiveness your either dominant person or
permissive friendly or unfriendly those are two basic dimensions those are
thongnicle, those are cross cultural, manifestations dip a little bit those
have been tested in a about seventy different countries and they always
work. We can mark personality with similar cues and most important what
is called the law of similarity attraction, we like people who are like
us. That's the law of similarity and then were, "birds of a feather
don't flock together", but opposites don't attract. How should we manifest
personality on computers if we don't know any artificial intelligence, and
all we have is text easy. Speaker: Does this actually help in real success?
Clifford Nass: Well, it's a good question. The answer is that in some cases
we've studied and some cases we are about to. In cases where we have studied
it, people cooperate more with the computer. Cooperation has been associated
in other studies with learning. So to the extent people are cooperating
which happens with both similarity personalities and teammate, it does lead
to greater learning. We are about to embark on a series of studies that
focus on learning specifically. But we also know that positive affect leads
to all sorts of good things, including greater success. But to the extent
that you can manipulate positive affect, it leads to success as well. Speaker: In regards to the social aspect, have you done any experiment
in a sort of manage tois (French word) kinds of experiments. Because right
now we are now talking more about collaborative groupware. Have you done
anything with true psychology where you have people with mixed temperament
working together with the computer? Clifford Nass: We haven't but the areas that we are most excited
about are CSCW, computers for collaborative work. We have multiple computers
and multiple people. And the interesting question there is, "Do you
start getting stereotyping in-group out-group phenomena?" So that it
becomes we the people versus them the computer or you get the computer,
my computer and I are a team; therefore, defacto we are better, stronger,
smarter and not going to cooperate with these other people. We haven't started
that yet, but that is another avenue. Some psychologies are a 100 years
old roughly and big and we are small. So we are busily pulling off the bookshelf.
But people are welcome to come in and help. So similarly when they are similar
you get these blame and things. Now here is a very interesting one on control.
Most of the literature in HCI totally uninformed by human-human interaction
says, "Control is a good thing." My friend the antagonist, Ben
Schneiderman, is always saying that user control is the ultimate good. It
turns out that is not true because a list in terms of affect, if you think
people are going to succeed, then you want to give them control. They will
feel good. But if you think the likelihood of failure is high and you want
people to feel good, take away control. That way they can... [inaudible question] Good question. It depends on the particular study we are talking about.
In a case of the tutoring situation, it means a high score or low score
on a test. In the case of blame, it was a different type of test. We said
experts from the U.S. Army figured out the desert survival problem. We are
going to compare your score to theirs. So there is objective success. As
it turns out, my own particular bias is not towards objective success, but
perceived success. I personally am much more interested in how people feel
rather than their sort of learning, which is why learning comes last as
opposed to first. But certainly actual success is a valid variable. I think
it has been overemphasized, but it is certainly a valid variable. No, at
the expense of feeling of good, liking things. Those are important and good
things. Speaker: Do you advocate routinely giving personality tests to
users? Clifford Nass: It depends on what you mean by routinely. If someone
said to me, "I can make you work better with your computer, feel better
about the interaction, and have a good time," and all you have to do
is answer these 8 questions, yes I would feel comfortable with that. Now,
the potential for elicit forces, like Microsoft or something, capturing
this information and destroying the world are real. So there is that aspect.
But on the other hand, there is certainly an issue of privacy and privacy
is always an issue. Same thing in human-human interaction. The more you
reveal about me, the more potential power I have over you and the more miserable
I can make your life. Also the nicer I can make your life. Which is why
marriages are some of the happiest things and divorces are some of the most
unhappy things. They are both based on too much information, which can either
be used for good or ill. And I think the same rules apply. Information is
a very complicated thing. What I am saying though, if you trust, and I think
the issue of trust is an important one, giving information can optimize
an interface and make people work better, happier, etc. Speaker: It seem very clear that these effects are true for short-time
interactions with the computer. We have longitudinal studies. I know people
have dated bots and muds for about 6 months and eventually figure out that
they are dating a computer. In interacting with the machine initially we
are going to use all that wiring that you talked about, but over 6 months,
a year and long-going interaction, the differences between human-computer
interaction and human-human interaction, one would think, are going to become
apparent, so have you studied that. Clifford Nass: No. It is a great question that takes a lot of
money and a lot of time. But it is a great area. There has been some studies
with respect to Bob over time use, do people get tired of it. Speaker: It seems it is a little dangerous to extrapolate from
a very short-term interaction and conclude that that is true for long-term
human-computer interaction. Clifford Nass: Well that is a good question. The answer to which
is that types of rules you are using are unlikely to be extinguished (that
is a fancy psychological term). Things may go away, because there is little
motive not to. That is it is only when you make some egregious error that
things get bad. I know, even in the case where you are dating a robot, that
in and of itself is probably not an egregious error because you have had
this social interaction which is what you wanted presumably by doing this.
But when it becomes an egregious error is when someone comes up to you and
goes, "ha-ha-ha, you are stupid." Now, does that protect you the
next time. No. You will then encounter someone else who will manifest social
rules and you 're evolved brain will fall for it just as much as it did
the last time. Not because you failed to understand, but because that is
the way brain is built. So I think the emphasis on these affects going away
are actually much overestimated. We can really parse--our brains are too
small. Not only evolved a particular way, but are too small to continually
filter through. Let me give you my favorite example of this. When you go
to a horror movie--you get very scared. How do you calm yourself down? The
best way is to say "It's only a movie, it's only a movie." Now
why doesn't your brain say, "You moron, what the hell else do you think
it was!" The answer is your brain has a great answer for that--real
life. Because your brain was built to rules of human evolution. When is
saw something, it didn't sit there and go, "Is it real?" Until
? we didn't have to worry about that stuff. So the point is that your brain
is built to accept all this stuff as real and what is hard is to filter
it. Now what happens when you say, "It is only a movie, it is only
a movie." You don't get the plot, you miss out all the cool stuff because
your brain is not big enough to keep on going, "It is only a movie,
something happened, it is only a movie, something happened." Because
our brains are built the way they are that you can do this sort of what
we call discounting for media. Let me just zip through one more of these
and then if people have particular concepts they are interested in, I will
do it. There is a lot of work on voice input. A lot of companies, including
IBM, are going into having software that has voice input. But one of the
things that we know in real life is that when we speak with someone, there
is an increased social presence. That is why it is harder to break of with
someone, if you have to talk to them it is harder to fire someone. It is
easier to write a letter. That may not be nice, but it is certainly easier.
We wanted to see whether people would feel emotionally funny in the same
way with computers. The particular rule we explored here is the more social,
the more social conformance. If you are alone in a house you can eat your
dinner, you can beans with a wooden spoon, with a towel wrapped around you
over the sink. You can't do that with a lot of people around or you get
in trouble. So you conform more with larger social groups. So we wanted
to see whether when people spoke to a computer as opposed to use text code
whether they would in fact be more circumspect. So we had an input mode
of text and voice and output mode of text and voice, and we asked people
personal questions about themselves. People were told they would not be
recorded and everyone believed that. They just though it was a speech recognition
system. We did not have a speech recognition system, so recorded them. But
we destroyed the tape immediately after beaus it is unethical. And what
we were interested in here is the extremity of responses. It turns out that
is not socially normative to give extreme responses, more so for dominance
and submissive. But to say absolutely to personal questions is considered
not okay. And also people tend to give the more socially accepted response.
So we asked people a variety of survey questions. One involved feminist
issues and one involved political issues. We did two different experiments.
And the moral to the story is summary both, voice input led to less extreme
responses and more socially acceptable responses. So this should be a lesson
not a moral--sorry about that. That is people were more circumspect in speaking.
Voice output had no affect on this. Interestingly also, we asked people
how honest they felt they could be, depending on which condition we are
in. Here is a case where people actually, even though they denied this,
when you looked at the scores for people who use voice input, they actually
felt they could be less honest. So somewhere they were able to access the
rule that they being circumspect. Not just doing it unconsciously. So just
let me summarize. We will have a few minutes for questions. First of all,
very important. People can't tell us what they think and feel. If we did
focus groups on all these experiments, I wouldn't be presenting here today
because the results would people aren't polite, they are not subject to
flattery, they don't believe in personalities, they don't get personal.
Because when you ask people, that is what they tell you. You are not wrong,
they just don't know. You have to use methods that get people to say what
they thinking and feeling, and if that thing in some of our other stuff,
we actually use things that really don't ask. We are using electroencephelograms
and electrocardiograms and skin conductance to get at responses. Second
point someone was asking, quality is really a perceptual issue. What is
good and what is bad is really perceptual not technological. Tells what
people perceive is good or bad that is important. Third point is people
are human first. People always say that experts are different which is why
we run all these studies with experts. <*** recording tape changed ***> Ken Kahn: Ken Kahn here, I was wondering if you could comment
on lessons that all of us might learn from the experience with Microsoft's
Bob. Clifford Clifford Nass: He won't let me, but you can grab me afterwards. Ted Selker: That's too big a question and we are really out of
time, Phil I need your comment though, please go ahead I want you speak. Phil Agre: This may be as big a question as Ken's, Cliff I found
your presentation ethically troubling all the way down, I want to .... Clifford Clifford Nass: It's not my fault (laughs in the background) Phil Agre: No, I think it is. At least, it's my concern. Let me
just try a scenario on you. In the literature you are talking about is a
great deal of research on the conditions under which people are more likely
to obey instructions. What do you think about imbedding those principles
in user interfaces. Are you comfortable with that? Clifford Nass: Okay, I think I can give you a really short answer.
It is critically important and socially valuable to know all the terrible
ways that people can be manipulated. That is critically important that is
not to say, nor if I advocated at all in this talk, that we necessarily
should use those methods. The discovery that people can be manipulated is
one of the most important social findings in the 20th century and I'm also
delighted we know that. I'm also delighted that we know we should avoid
it, that's good too. There is no ethical component to the discovery that
these things exist, there is an ethical component in using them and I am
not advocating which ones you use and which ones you don't. That's for the
individual .. Ted Selker: Except, except when you are in your consulting role. Clifford Nass: Well but even there, I'll give you a really short
anecdote: male characters are trusted more than female characters. Ted Selker: So the character in us, for example if I am designing
a user interface I really want to focus and work on tasks and be oriented.
Now if I've got this little guy over here, that's like disorienting me.
I'm sorry guys, that's not really helping with my task. Now when is it appropriate
to have an avatar helping me in a task, and that has to do when the task
is generally social probably plus I'm sure we can learn about that. Clifford Nass: No, it's the same thing as sometimes when I want
to know what the meaning of a word is, I look in the dictionary. Sometimes
I go to the guys next door, not for reasons of speed but I feel like being
social with the guy next door. Even though I may be working on a task I
may just feel like it. Similarly social things should be there, social manifestations
should be there when you feel like it. With that said, one lesson from Bob
is the characters there where way over the top. They spent their life saying
look at me I am a character, look at me I am a character, we don't like
that in people and we certainly don't like in software either. So social
presences that are available when we want them and not when you don't are
the people we like the best and those are the people we should model. Ted Selker: This is a fantastic talk, Thank you so much. Clifford Nass: Thank you and I'll be glad to talk to people afterwards.
And buy the book. Ted Selker: Buy the book, I know there is a lot more people interested
and a lot more questions about .... Clifford Nass: The media equation. Ted Selker: Ooh... Clifford Nass: Mediated life equals real life. Ted Selker: Anyway, no more speaking, lets get the microphone
of him. I want..... Clifford Nass: Cambridge University Press, but it's in regular
bookstores. Ted Selker: No advertising, ...give it to him... Clifford Nass: Copyrighted. (laughter) Ted Selker: I really hope that the discussion about this talk
has more life in the discussion period following this talk. I trust and
I expect that we will have more chance to talk about all these issues and
I think they revolve around a lot of things that we were going to be seeing
more and more reasons to think about. With that we are going to talk about
something very different, we are going to talk about....
|
|
[ IBM home page | Almaden Research Center | IBM Research | (C) | (TM) ] |