Book Review by Paul Dourish
Apple Research Laboratories
Research meetings around the Bay Area and the country for the last couple of years have, every so often, played host to two Stanford professors with an unusual thesis -- that people treat computers as social actors. The professors, Byron Reeves and Cliff Nass, are from the Department of Communication at Stanford, and for a number of years they have led a research project entitled "Social Responses to Communication Technology", bringing together interests they had previously been pursuing separately into viewer responses to television and film, and user responses to computers. "The Media Equation" was published in 1996, summarising the results of this long-term project.
The equation of the book's title is "media equals real life". The conclusion with Reeves and Nass draw from the series of experiments which they outline in the book is that people treat and respond to media (computers, televisions, etc.) in just the same way as they treat and respond to other people in everyday social interaction. The computers and televisions are treated as social actors; and the rules which people apply to everyday social interaction with other people apply equally well to their interactions with media.
This final statement is partially suggestive of their research methodology, which they outline in the first chapter. The methodology is, on the surface, quite straight-forward. Go to the library and pick up some social science textbooks. Pick a finding describing a social rule governing people's interaction with each other. One example, which is used in the book, is "people like other people whose personalities are like their own."
So you have a rule which describes social behaviour. According to the media equation, media and real life are the same, so people should also like media whose personalities are like their own". Now find the experiment by which the original rule was uncovered, and re-run it, substituting media (a computer, a television or even a picture) for one of the people. The claim is that, when you run the experiment, you will find that the results work out just the same with media as with people, supporting the media equation.
Subsequent chapters go through a series of rules and experiments, following this general pattern to show how the experimental results remain stable across people and media, and so building support for the media equation. So, for instance, they design programs which seem "extrovert" and "introvert", and have people use these programs and then fill out surveys about what they thought of the programs. Sure enough, extrovert people prefer the extrovert programs and introvert people prefer the introvert programs.
Examples of the rules which are used in these experiments include: when people ask a question about themselves, they will receive a more polite answer than if the question was asked by a third party; people place more trust in experts than in non-experts; people who change to make other happy are preferred over those who always try to make others happy; and people cooperate more with teammates than with others; people prefer those who flatter them to those who criticise them. For each, their experimental results show the same results for people's responses to media as were found in the original experiments on people's responses to other media. Some of the results are quite startling. For instance, their experiment on perceived expertise involved people rating the quality of the same news and entertainment segments in two conditions. Some people saw all the segments on a television which had a sign identifying it as a "news and entertainment television"; others saw the news segments on a "news television" and the entertainment segments on an "entertainment television". Those who watched the segments on the "specialist televisions" rated the news segments as more informative and objective, and the entertainment segments as more relaxing and enjoyable than the same segments were rated by people who'd watched them on the "generalist television".
From results like these, Reeves and Nass draw the conclusion that the media equation holds. However, despite the barrage of results which they present, I found this conclusion deeply troubling. The evidence was intriguing but still unconvincing. I'm not questioning their results, but I definitely question whether these results provide a warrant for their conclusion. There are a number of issues I'd like to raise here.
The first concerns the nature of the "social rules" which they apply in the first place. The "social science" texts from which they draw them strike the reader as more likely to appear on the psychology shelves of Waldenbooks than on the shelves of the Stanford library. The common features of these rules are that, first, they're exceptionally broad, ignoring the rich context in which social behaviour arises; and, second, that their foundations are vague and unexplored. The very nature of these experiments suggests that the term "people" is not very well founded in these rules; and questions about just what constitutes "preference", or what might be meant by "flattery" or "expertise" are similarly questionable. The social psychology which underpins these rules (and it's interesting that their perspective on "social rules" is drawn entirely from social psychology, never from sociology or anthropology) is never explored. This is particularly problematic because, within the context of a book for the layman, the work is presented with all the hallmarks of a strongly scientific orientation (hypothesis, methodology, experiment, conclusions); but when we look at the content, the scientific facade falls away quickly. If the rules themselves are questionable, then we have to question what sort of conclusions might be drawn from them when they're abstracted into some completely different domain.
Second, I have a number of concerns about their research methodology. Sociologists have been consistently critical of the sort of laboratory studies on which Reeves and Nass' work seems to be based. Social action, after all, is something which unfolds in everyday interaction, and the experimental situation is hardly everyday. The social aspects of the experimental situation (the decontextualisation of action, and the roles of experimenter and subject) tend to outweigh the social elements of what is actually done in the experiment. This, in many ways, is the real lesson of Millgram's famous experiment into authority, in which people were persuaded to administer (the believed) potentially fatal electric shocks to an experimental subject (actually another experimenter), precisely because the context of "doing science" would override people's natural responses.
The third issue is, for me, the most significant. The conclusions drawn from the experiments described in the book are often, simply, subject to some dispute and interpretation. For instance, consider the experiment concerning generalist and specialist televisions. To me (perhaps particularly as a recently-arrived foreigner), this experiment seems to say more about American broadcast television than it does about people's responses to technology.
There is also a concern about how people impute social behaviour to the media or to those seen to shape it -- the programmer, in the case of the computer systems. This is a question they specifically address, but not nearly to my satisfaction. If I read an article which annoys me in the newspaper, scrunch up the paper and throw it angrily away, I don't think that's a reaction to the newspaper as a social actor. When my social behaviour depends on whether someone's office door is open or not, I don't think I'm responding to the door as a social actor. The story in the newspaper or the state of the door carries social significance, for sure, but that's not the same thing; but for the purposes of this book, it would seem there is no form of social significance which is not human-like. Reeves and Nass seem to systematically underestimate the intelligence of their subjects; and when they point out that the rules seem to apply even when people understand that computers can't think, be angry or choose to act arrogantly, my reaction is that they undermine the validity of the formulation of the rules themselves.
Of course, they are not blind to these problems. For instance, concerning the specialist television experiment, they conjecture that their subject might be conceiving the segments as originating on specialist television networks, rather than televisions (and they conduct a second experiment which supports this). So they conclude that "media" in this case means television networks, and the media equation remains intact. However, this seems like another case of underestimating the intelligence of their subjects, as if they simply mechanically follow the rules (what Harold Garfinkel called the view of man as a "judgmental dope"). If people presume that news segment on the News Hour are more objective than those offered by ABC, are they blindly responding to the news agency as a person, or are they applying what they know about the political and financial elements of news production and television scheduling? The conclusion which Reeves and Nass draw seems to suggest that this plays no part in the reaction of their subjects whom, they continually emphasise, are intelligent and well-educated (being largely, I believe, Stanford undergraduate and graduate students).
In this vein, they conclude, "The research showed that people are more simple than we often imagine." This seems to me like an assumption behind their work rather than a conclusion from it, and it raises an interesting point. The authors and their students have worked with Microsoft and they refer in their preface to "the great pleasure of seeing some of our work included in their products". The primary example which they point to is "Bob"-- a system which most people seemed to regard as an insult to their intelligence. Somehow, I'm not surprised.