Computers are Social Actors

This lecture taken from

``Computers Are Social Actors: A Review of Current Research''

by Clifford Nass, Youngme Moon, John Morkes, Eun-Young Kim, and B. J. Fogg

In Batya Friedman, ``Human Values and the Design of Computer Technology''

and from a lecture by Clifford Nass at NSF HCI Grantees Workshop, August 1997

Do People Respond Socially To Technology?

Of course. We mutter at the washing machine, swear at the computer, talk back to the telephone (just the bell part, I mean).

traditional academic response:

1. it's an aberration (lack of knowledge, or psychological or social dysfunction)
2. social behavior is directed at the human creator of the program or machine

Some Social Behaviors

• politeness norms
• response to personality types (dominant/submissive, friendliness)
• flattery
• gender stereotypes
• team affiliation

The Subjects

• Stanford undergraduates in communications classes (in published paper)
• Stanford computer science graduate students (mentioned in lecture)

Politeness Norms

Tutoring, testing, evaluation task. Following task completion, subject interviewed about the performance of the computer.

Conditions:

1. interview conducted by same computer
2. interview conducted using pencil and paper
3. interview conducted by an idential computer in another room

Responses were significantly more positive and more homogeneous for condition 1.

Politeness Norms (2)

These were all simple text-based interfaces.

In post tests, all the subjects said (sometimes vehemently) that it would be absurd to engage in polite behavior toward a computer.

Another set of experiments: voice output, using same voice or different voice.

Design Implications

Product evaluation questions should not be asked by product itself.

Nor should they be asked by the technology used to test the product (despite convenience). Use pencil and paper, or a different technology.

Users may expect politeness from computers. Most systems avoid direct insults, but for example error messages are often impolite.

Cultural issues: internationalization may require more than just translating the interface text.

Attraction and Personality

Social psychology: one of the best predictors of whether two people will like each other is to find how similar they are.

Personality trait tested:
dominance/submissiveness

Personality programmed using very simple preprogrammed text-based cues.

Example:

• ``You should definitely rate the flashlight higher. It is your only night signalling device.''
• ``Perhaps the flashlight should be rated higher? It may be your only reliable night signalling device.''

Attraction and Personality: Results

Users identified dominance and submissiveness in the programs.

Users were able to detect similarity of the computer's personality to their own.

There was strong evidence that subjects preferred interacting with computers that shared their personality type. (This was consistently true for all personality types.)

Flattery

``People are phenomenal suckers for flattery.''

Experiment: a guessing game. Test conditions:

• generic feedback (``begin next round'')
• sincere praise (``Great job! You seem to have an uncommon ability to structure data logically.'')
• flattery (same, but subjects told that evaluation portion of program not yet written, and computer feedback had nothing to do with their performance)

Result: subjects in both sincere praise and flattery conditions felt much more positive about themselves and the computer. No significant difference between sincere praise and flattery.

Flattery: Design Implications

Most current computer applications are heavily geared toward critical feedback.

Add positive feedback? Even noncontingent feedback?

Categories of software for which this might be important:

• training and tutorial software
• software that enhance user creativity
• software for performing unpleasant tasks

Gender Differences

Stereotypes tested:

• dominant behavior by men vs. women
• evaluation coming from men vs. women
• knowledge about various topics (experiment: knowledge about computers and technology, knowledge about love and relationships)

Gender Differences: Results

Female-voiced computer in a dominant role was evaluated more negatively than a male-voiced computer in the same role. It was perceived as significantly less friendly. This was true for both female and male subjects.

Evaluations from a male-voiced computer was regarded as significantly more ``competent'' than from a female-voiced computer. This again was true for both female and male subjects.

The female-voiced computer was perceived as a better teacher on the subject of love and relationships; the male-voiced computer was perceived as better for computers and technology.

Gender Differences: Theoretical and Design Implications

• The tendency to gender-stereotype is so deeply ingrained in human psychology that it extends even to computers. Vocal cues alone elicited this response.
• Choice of gender of a computer's voice is an important design decision. Choosing a male voice or a female voice cannot be a neutral decision.
• Computer voices may indicate much more than gender -- for example, age, social class, geographic location. This may create expectations about how the computer will behave.
• Example: what voice should be chosen for a CD-ROM with medical advice for pregnant women? Male or female? How old should it sound? Accent? Two or more voices?
• Should computer agents conform to stereotypes at all? Should we design agents that challenge stereotypes? (Example: synthetic gender-neutral voice. Also see ``The Left Hand of Darkness'' by Ursula Le Guin.)

An Alternative Explanation?

Is the social behavior directed at the human creator of the program or machine?

Social psychology predicts no -- people orient toward proximate sources (``blame the messenger'').

Experiment: half of subjects told they were working with computers, half told they were working with programmers.

Result: subjects who were told they were working with computers perceived the tutor to be significantly more friendly, effective, playful, and similar to themselves.

If social behavior were directed at the programmer rather than the program, there should have been no difference.

Implications for Agent Research

Maes: ``The agents have deliberately all been drawn as simple cartoon faces, in order not to encourage unwaranted attribution of human-level intelligence.''

Nass: ``Furthermore, these studies suggest that it does not take extremely sophisticated technology to generate social responses.''