Can Computer Personalities Be Human Personalities?

Clifford Nass, Youngme Moon, BJ Fogg, Byron Reeves, & Chris Dryer

Department of Communication
Stanford University
Stanford, CA 94305-2050, USA
phone: 415-723-5499
e-mail addresses: nass@leland.stanford.edu & youngme@leland.stanford.edu

Abstract

The present study demonstrates that (1) computer personalities can be easily created using a minimal set of cues, and (2) that people will respond to these personalities in the same way they would respond to similar human personalities. The present study focuses on the similarity-attraction hypothesis,which predicts that people will prefer to interact with others who are similar in personality. In an experiment (N = 48), dominant and submissive subjects were randomly matched with either a dominant or submissive computer. When a computer was endowed with the properties associated with dominance or submissiveness, subjects recognized the computer's personality type along only that dimension. In addition, subjects not only preferred the similar computer, but they were more satisfied with the interaction. The findings demonstrate that personality does not require richly defined agents, sophisticated pictorial representations, natural language processing, or artificial intelligence. Rather, even the most superficial manipulations are sufficient to produce personality, with powerful effects.

Keywords:

agents, personality, individual differences, computers are social actors

INTRODUCTIONM

It has become a commonplace idea that believable agents must have personalities. Yet, attempts to create personalities of agents virtually never draw on the enormous psychological literature on personality. This is ironic, because psychologists have been defining personality, and identifying how particular people will respond to various personality types, for many decades.

The present study turns to the vast personality literature in the field of psychology to conceptualize personality. The goal is to find the minimum set of cues necessary to create a computer- based personality, and to determine whether these cues are sufficient to produce powerful effects. Specifically, using a laboratory experiment, we seek to demonstrate that if a computer is endowed with a set of personality markers, users will be able to identify that personality, and respond to it in a manner predicted by personality theory in the field of psychology.

DEFINING PERSONALITY

Psychologists have identified two major interpersonal personality dimensions, the dominance/submissiveness dimension, and the affiliation (warmth/hostility) dimension [3].

This study focuses on the dominance/submissiveness dimension of interpersonal behavior. Behaviorally, dominance is marked by behavior that is self-confident, leading, self-assertive, strong, and take-charge. Submissiveness is marked by behavior that is self-doubting, weak, passive, following, and obedient [3].

We seek to demonstrate that computers can be given a personality along this single personality dimension, with powerful results. This leads to the following hypotheses:

Hypothesis 1: Subjects will perceive a computer with dominant characteristics as being dominant, and a computer with submissive characteristics as being submissive.

Hypothesis 2: Computers with dominant characteristics will not be perceived differently with respect to affiliation or competence, compared with a computer with submissive characteristics.

THE RELATIONSHIP BETWEEN PERSONALITY AND INTERPERSONAL INTERACTION

A pervasive finding in the psychological literature is that individuals prefer to interact with others who are similar in personality to themselves [2].

The present study attempts to demonstrate that the similarity- attraction relationship applies equally to human-computer interaction. Specifically, we have the following hypotheses:

Hypothesis 3: There will be a cross-over interaction between the user's level of dominance and the computer's level of dominance such that users will be more favorably disposed toward the computer, and more satisfied with the interaction, when the levels of dominance/submissiveness are similar.

METHOD

Procedure

Subjects were defined as dominant or submissive, based on responses to a standard personality test [1]. Subjects were randomly matched with a computer with either a "dominant" personality or a "submissive" personality in a 2 x 2, balanced, between-subjects design.

Upon arrival, the subject was asked to work with a computer on a problem-solving task that has been used in a variety of studies involving interpersonal interaction [4].

The dependent variables were measured using a set of paper-and- pencil questionnaires. Upon completion of the interaction with the computer, subjects were asked for their perceptions of the computer and of the interaction.

Manipulation

The dominant computer used strong language expressed in the form of assertions and commands, displayed a high confidence level, and always went first in the interaction. Conversely, the submissive computer used weaker language expressed in the form of questions and suggestions, displayed a low confidence level, and always went second in the interaction.

In sum, the fundamental information conveyed by the computer was not manipulated. Only the style of communication was manipulated. All responses were pre-programmed. There was no natural language processing or artificial intelligence employed.

RESULTS

All analyses are based on a full-factorial model.

Consistent with Hypothesis 1, the dominant computer was perceived as significantly more dominant than the submissive computer (F(1,44) = 15.92, p<.001), based on a 9-item index; conversely, the submissive computer was perceived as being more submissive than the dominant computer (F(1,44) = 24.74, p<.001), based on a 5-item index.

Consistent with hypothesis 2, subjects did not rate the computers differently with respect to affiliation (F(1,44) = 1.32, p<.26) or competence (F(1,44) = 1.31, p<.26), based on 5-item and 12- item indices.

Hypothesis 3 predicted that subjects would prefer to interact with a computer that was similar in personality to themselves, i.e., dominant subjects would prefer the dominant computer, whereas submissive subjects would prefer the submissive computer. The results supported this hypothesis. When the person and the computer had similar personalities, the computer was given higher affiliation ratings (F(1,44) = 4.28, p<.05; 5-item index), and higher competence ratings (F(1,44) = 11.60, p<.001; 12-item index) compared to when the person and the computer had different personalities.

In addition, subjects matched with similar computers found the interaction to be more satisfying (F(1,44) = 5.18, p<.;<.05; 7-item index) compared to subjects matched with dissimilar computers.

Finally, when the subject and computer shared personality types, subjects were more satisfied with their own performance, compared to when the subject and computer had different personality types (F(1,44) = 9.02, p <<.005).

FIGURE 1: Mean Levels of Competence and Satisfaction by Subject and Computer Personality

DISCUSSION

There are numerous implications of these findings. The key theoretical implication is that personality is powerful and easy to manipulate; even in its simplest form, it can provide the cues for complex social behavior. In contrast to the prevailing idea that the creation of personality requires natural language programming, artificial intelligence, complex graphical environments, or richly defined agents, this research demonstrates that even the most rudimentary manipulations are sufficient to produce powerful effects.

Another implication is that humans respond socially to technologies. Indeed, this study provides additional evidence that the "Computers are Social Actors" paradigm [5] is a powerful and accurate description of human-computer interaction. The CASA studies have demonstrated that the social rules guiding human- human interaction apply equally to human- computer interaction. And like these CASA studies, the present study presents evidence that human-computer interaction is fundamentally social and interpersonal.

References

1. Bem, S.L. (1974). The measurement of psychological androgeny. Journal of Consulting and Clinical Psychology, 42, 155-162.
2. Byrne, D., Griffitt, W., & Stefaniak, D. (1967). Attraction and similarity of personality characteristics. Journal of Personality and Social Psychology, 5, 82-90.
3. Kiesler, D.J. (1983). The 1982 interpersonal circle: A taxonomy for complementarity in human transactions. Psychological Review, 90, 185-214.
4. Lafferty, J.C., & Eady, P.M. (1974). The desert survival problem. Plymouth, Michigan: Experimental Learning Methods.
5. Nass, C.I., Steuer, J., & Tauber, E. (April, 1994). Computers are social actors. Paper presented to CHI'94 conference of the ACM/SIGCHI, Boston, MA.