Article summary of Machines and mindlessness: social responses to computers by Nass & Moon (2000) - Chapter

How do people react to computers?

People differ in their ability to use a computer. Because of this, they will approach a computer in many different ways. People who are experienced with the computer, move smoothly from mouse to keyboard and they use commands automatically. People who are new to computers strike every key because they are afraid that one wrong move will initiate an uncontrollable series of weird or unwanted events. E-mail users treat computers just like phones and gamers see a computer as a window to another world. How people use the computer depends on technological, individual and situational factors.

Despite the differences in computer use, all computers users know that the computer is not a person and it does not warrant human treatment. It’s hard to believe that somebody might reach another conclusion. Robots and dolls have faces, but machines have no faces and bodies and don’t look like a person. However, not all machines are like that. A car might be suggestive of a human face, because of its headlights (for eyes). Computers are not aware of human emotions and a computer doesn’t express emotions. A computer doesn’t even see itself or refer to itself as ‘I’. There is clear evidence between a computer and a human and there are not many people who say that a computer should be treated or viewed like a real person. The word for treating something that isn’t a human like a human, is anthropomorphism. Although people in studies do state that a computer should not be treated like a human, they actually do treat computers like humans.

The authors of this article argue that there is much evidence that individuals mindlessly apply social rules and expectations to computers. In the first part of this article, the writers describe how people tend to overuse human social categories. This is true for gender and ethnicity. In the second part, the writers provide evidence for engagement in overlearned social behaviours, like reciprocity and politeness toward computers. In the third part, the authors describe how people exhibit premature cognitive commitments with respect to computers.

What are mindless responses?

Mindless behaviour has been observed in a wide variety of social situations and it is the result of unconscious attention to a subset of contextual cues. The cues trigger certain labels, expectations and scripts and these focus attention on certain information and divert attention away from other information. People, who respond mindlessly, don’t construct categories based on all the relevant features of a situation, but they commit to overly simplistic scripts drawn in the past. Individuals are responding mindlessly to computers when they apply social scripts, scripts for human-human interaction and when they are ignoring the cues that reveal the essential asocial nature of a computer. The authors of this article have turned to the literature in experimental social psychology and found the use of scripts during human-human interaction.

Then, they replicated the experiments as closely as possible, but with one important change: no human-human interaction, but a human-robot interaction.

To elicit mindlessness, an individual must be presented with an object that has enough cues to make that individual categorize it as worthy of social responses, while it should also permit individuals to note that social behaviours were clearly not appropriate. One might wonder what cues might encourage a categorization of computers as social actors. There has not been a systematic investigation of this point, but there are a few characteristics that distinguish computers from other machines and that are closely associated with the human prototype. All these characteristics are incorporated into virtually every application on every personal computer.

What is the effect of an overuse of categories?

The first experiments focused on whether individuals would carry over human social categories to the computer realm. The first and one of the most powerful social categories the experimenters tested, was gender. They focused on three well-established gender stereotypes. The first one is that dominant behaviour by males is well received. Dominant men are seen as independent and assertive. However, dominant behaviour by women is not well received. These women tend to be received as bossy and pushy. The second one is that people who are evaluated tend to consider evaluations more valid if they come from a man than if they come from a woman. The third one is that certain things are seen as feminine and certain things are seen as masculine and people tend to assume that women know more about feminine topics and men more about masculine topics.

The researchers wanted to determine whether computers would trigger the same scripts and attributions with the gender stereotypes and therefore they designed an experiment involving computer voice output. Participants were told that they would use computers for three separate sessions. These sessions were tutoring, testing and evaluating. The computer had a pre-coded male or female voice and during the tutoring session, the computer verbally presented a series of facts on one of two topics: computer and technology (this is a stereotypical male topic), and love and relationships (a stereotypically female topic). After the tutoring session, the participant moved to a tester computer for the testing session. The tester computer had no voice and the participants had to complete a multiple-choice test, with each question having a correct answer. Afterwards, the participants were told to move to an evaluator computer. The evaluating computer used a different pre-coded male or female voice and reviewed each question, indicating whether the participant had given the correct answer and evaluated the performance of the tester computer. The evaluations were generally positive. The evaluator computer said things like ‘The tutor computer chose useful facts for answering this question’. The evaluator computer played two dominant roles: It evaluated both the performance of the participant and the tutor computer.

Afterwards, participant had to fill in (paper) questionnaires. The first set of questions were related to their assessments of the tutor computer’s performance and the second set of questions was related to an assessment of the evaluator computer.

The results showed that participants mindlessly gender-stereotyped computers. Male as well as female participants found the female-voiced evaluator computer to be less friendly than the male-voiced evaluator computer, even though their comments were identically the same! Also, they praised the comments from a male-voiced computer as more compelling than the same comments from a female-voiced computer. Also, participants thought the tutor computer was significantly more competent when it was male-voiced rather than when it was female-voiced. The male-voiced tutor computer was rated as more informative about computers, while the female-voiced tutor computer was rated as more informative about love and relationships. It is important to note that participants knew that the voice of the computer did not necessarily reflect the gender of the computer programmer. They also indicated (in post-experimental debriefings) that they thought that male-voiced computers were not different compared to female-voiced computers and that it would be weird to engage in gender-stereotyping with respect to computers.

What is the role of ethnicity?

The second category the researchers tested was ethnicity. An interactive video manipulation was used to provide participants with an ethnicity cue. Minority group members tend to exhibit much higher levels of ethnic identification and because of this, the researchers used Korean rather than Caucasian participants. The participants were given a couple of hypothetical choice-dilemma situations in which an individual had to decide between two different action choices. One option was more rewarding and attractive, but less likely to be attained. Participants read the description of the situation, made a decision and asked the computer agent- represented by a Caucasian or Korean video face- what decision it would make. The computer agents also provided a couple of arguments for the decision. After being presented with the decision and arguments, participants answered some questions about the quality of the agent’s arguments, the perception of the agent’s decisions and their own decisions. This was done for eight different choice dilemma situations.

The results showed that participants were using mindless stereotyping. The social category cue (ethnicity) triggered a series of expectations and attributions, regardless of the context in which the cue appeared. Koreans who saw a Korean-face agent, perceived the agent to be more attractive, trustworthy, intelligent, persuasive and as making a decision more similar to their own than Koreans who saw a Caucasian-face agent. This experiment was also repeated with a picture of a person, instead of a video. The participants were told they had a videoconference with an actual person (from the picture) and the results were the same as for the experiment with the computer. This showed the same results.

What is the effect of ingroup versus outgroup?

The researchers wanted to determine whether people would rely on an arbitrarily assigned social category, such as ingroup versus outgroup, when they interacted with computers. During games between groups, people were assigned to either the blue or a green team. The mere act of being labelled and made dependent on others leads to feelings of loyalty and perception that one’s team is much better than the other team. In one condition of the experiment, the researchers tried to create a feeling of shared identity between the person and the computer by reminding the person that he or she was dependent on the computer, giving the person a blue armband, referring to the participant and computer as the blue team and putting a blue border around the computer’s monitor. In the second condition, each participant was referred to as the blue person (armband) working with the green computer (border). Also, the participant was encouraged to focus on individual responsibility.

The manipulation of this study was minimal, but the results were major. Participants in the team condition (first condition) were more likely to cooperate with the computer, to assess the computer as more friendly and intelligent, to conform to the computer’s suggestions and to perceive the computer as being more similar to themselves, compared to the nonteam (second) condition. The research also showed that the mere matching of an armband and border mindlessly induces social responses. In post-experimental debriefings, the participants indicated that the labelling was irrelevant to their behaviours and attitudes.

The results of these three studies show that people do rely on social categories when interacting with a computer and that this is even the case when the cues associated with those categories do not have the same meaning or even make sense in a human-computer context.

What is the effect of overlearning?

Overlearning can also cause mindless behaviour. Overlearning is the use of deeply ingrained habits and behaviours. Once a script is initiated, some humans stop searching for additional context cues and respond according to the script.

What is the role of politeness?

Studies have shown that when individuals are asked to evaluate another person in a face-to-face setting, the evaluations tend to be positively biased. People tend to give polite evaluations in these settings. This can also include lying, because people are reluctant to hurt the feelings of that other person. The researchers used text-based computers for their study. Participants worked with computer-A and were then interviewed about computer-A’s performance.

The first condition had computer-A interview the participants. In the second condition, computer-B interviewed the participants. In the third and last condition, the interview was conducted through the use of paper-and-pencil questionnaires.

The results showed that evaluations were significantly more positive when the computers asked about itself as compared to the two other conditions. This means that people were even polite to a computer. In post-experimental debriefings, participants indicated that they believed that computers don’t have feelings and that polite behaviour to computers is not necessary. However, they show the same behaviors, which indicates overlearning.

What is the role of reciprocity?

In almost every society, reciprocity is encouraged in its members. This means that you should help somebody if he or she has previously helped you. Reciprocity has a really strong influence and some researchers think that it is the central characteristic of being human. In the first experiment, there were two tasks. In the first task, a computer helped the user and in the second task the participant was asked to help the computer. In task one, participants searched the web with a computer. The results of the searches were either extremely useful or not at all useful. In the second task, participants worked with a computer that was trying to make a colour palette to match human perception. Participants were told that by accurately comparing sets of presented colours would help the computer to create the palette. Participants could choose how many comparisons they would make. The more comparisons they made, the more participants would help the computer. In one of the conditions, participants performed task two on the same computer they performed task one on. In the other condition, participants used different computers for task one and two.

The results showed that participants acted according to reciprocity norms. Participants who worked with a helpful computer in task one and had to perform task two with the same computer, performed more work for the computer in task two than participants who used two different computers for the two tasks. The participants who worked on the same, helpful computer on both tasks even performed with greater accuracy on the second task. Also, when the computer in Task 1 wasn’t helpful and participants also had to work on this computer for Task 2, they made fewer comparisons than participants who used different computers.

Reciprocal self-disclosure

Research has shown that people are reluctant to convey intimate information about themselves to anyone but their closest friends and relatives. There is one notable exception to this rule: reciprocity. People will engage in intimate self-disclosure with strangers if they first become the recipients of such intimate self-disclosure from their conversational partner. People who receive intimate disclosure somewhat feel obligated to respond with personal disclosure of equal intimacy.

The researchers wanted to find out whether participants would engage in reciprocal self-disclosure with a computer. Of course, the computer would initiate the disclosure process by conveying intimate information first. Participants were then interviewed by a computer on a variety of topics. In the no-reciprocity condition, the computer asked questions in a relatively straightforward manner. In the reciprocity condition, the computer preceded each interview question with some parallel information about itself. For example, before asking the participant what his or her hugest disappointed in life was, the computer said that 90% of the computer users don’t use the computer to the full potential. Before asking what the participant has done in life where he or she feels guilty about, the computer told that it crashes on the most unpredictable time and that this causes great inconvenient to the user. The information provided by the computer was descriptive in nature and referred to factual matters. The computer never made statements that implied that it had emotions, attitudes or feelings and it also never referred to itself as ‘I’. The second condition was much lengthier than the first and the researchers therefore also decided to add a control condition in which the words used in questions were the same as in the reciprocity condition. Of course, this control condition did not include disclosures from the computer.

The results showed that self-disclosure tendencies were consistent with the norms of reciprocity. So, responses in the reciprocity condition included higher intimacy than responses in the other conditions. It thus seems that social scripts are activated even in a context in which they do not make sense, but also in a context in which the trigger for the script makes the nonhuman source of the information explicit (so when the computer gave information about its accomplishments/failures).

What is the effect of premature cognitive commitment?

Mindlessness is distinct from mere overlearning because mindlessness results from a single exposure to a stimulus, as opposed to repeated exposures. When an authority figure provides information, this information is often accepted uncritically. There will also be no attention paid to other aspects of the situation. This refers to premature cognitive commitment. The researchers wanted to determine whether this premature cognitive commitment also takes place when people interact with computers. Computers might naturally be perceived as authoritative in the content they produce, so the researchers decided to focus on a technology that does not produce content and is never seen as an expert: a television.

What is the difference in specialist versus generalist?

The researchers wanted to know whether the mere labelling of a television as a specialist would influence individuals’ perception of the content it presented. Participants were brought into the laboratory and watched fragments from news shows and situation comedies. Some participants were assigned to watch the generalist set and they were told they would watch an ordinary TV that experimenters used to show both news and entertainment shows. There was a sign on top of the TV that said that it was a ‘News and Entertainments Television.’

Other participants were appointed to the specialist condition and they were told that they would watch programs on two different televisions. They were told that they would watch the news on a television that was only used to show news programs and that they would watch entertainment on a televisions set that was only used to watch entertainment programs with. On top of the news television there was a sign that stating that it was a news television and on top of the entertainment television there was a sign stating that it was an entertainment television.

The results showed that there was a premature cognitive commitment to the notice of expertise. Participants in the specialist condition thought that the news segments were significantly higher in quality, more informative and interesting than did participants in the generalist condition. However, all of them viewed identical news segments. Participants in the specialist condition also thought the entertainment fragments were funnier and more relaxing compared to participants in the generalist condition. This shows that even meaningless assignments of expertise can result in mindless acceptance of content.

What can be said about depth and breadth of social responses?

The researchers did not only want to establish the breath of mindless responses across areas of social psychology, but they also wanted to establish a rich set of results within a domain of psychology: personality.

People usually think that it takes tremendous computing power to create personality in a computer. In this study, the authors made participants work with a computer that displayed a dominant or submissive personality style. The dominant computer used assertive, strong language during a task (you should definitely do this) and the submissive computer used more equivocal language (perhaps you should do this). The dominant computer also expressed high confidence in its actions during the task (confidence level of 80%) and the submissive computer expressed low confidence (confidence level 20%). First, participants themselves were categorized based on whether they had dominant or submissive personalities. The participants were then paired with a computer that either matched or mismatched their personality. Just like the similarity-attraction principle, the researchers found that dominant participants were more attracted to, conformed more to and assigned greater intelligence to the dominant computer, compared to the submissive computer. Submissive participants reacted the same way to the submissive computer compared to the dominant computer. In other studies of these researchers, they found that people are more willing to purchase items via internet when using a matched-personality computer.

What are alternative explanations for the effects?

The experiments in the previous sections have shown that mindlessness is quite a good explanation for the wide variety of social behaviours that the researchers have observed during human-computer interaction. However, over the years, alternative explanations have also been proposed for these behaviours. These explanations allow for mindful responses on the part of the individuals. However, the authors of this article don’t think that these alternative explanations are correct.

Anthropomorphism

If individuals believe that computers are essentially human (anthropomorphism), human-appropriate responses to computers reflect a reasonable application of social rules and behaviour. Anthropomorphism was the standard explanation for social responses to computers. The authors reject this assumption of anthropomorphism, because participants in their studies were adults, who were experienced with computers. During the debriefing, they insisted that they would not respond socially to a computer and they denied the specific behaviours they had exhibited during the studies.

However, there are some individuals who do develop strong relationships with computers (or other objects). Some individuals become emotionally attached to an object and may give it names or talk endlessly to it. However, these responses are not evidence for anthropomorphism, because anthropomorphism involves the thoughtful, sincere belief that the object has human characteristics. The individuals in this studies are unaware of their behaviours and apply a couple of social rules mindlessly. There are differences in anthropomorphism, cherished objects and the responses described in the current paper, reflecting ethopoeia. Ethopoeia is defined as giving a direct responses to an entity as if it is a human while knowing that the entity is not human.

Orientation to the programmer

Some people think that social responses to the computer are not social responses to the computer at all, but instead are social responses to an invisible human behind the computer. This is usually the programmer. The argument is that individuals frame the interactions with computers as interactions with imagined programmers. Programmers are people and it is therefore (according to this argument) not surprising that individuals display social responses. However, the authors of the current article don’t agree with this. The majority of the participants in their research have indicated that they did not have the programmer or any other human in mind during the interaction. Also, in the studies described above that used more computers, the participants indicated that they thought that both computer programs were designed by the same programmer (not two different programmers). If people acted socially to computers because of the programmer, why would they then display different behaviour towards different computers who they thought had the same programmer? This suggests that mindlessness is a better explanation for social responses compared to an orientation to the programmer.

Demand characteristics

This argument against mindlessness focuses on the extent to which the experimental situations or the questionnaires encourage users to demonstrate social responses. Some researchers think that the participants believe that they are asked to pretend that they are in a social situation. They might assume that in order to engage in the experimental task, they are expected to forget that they are dealing with a computer. The writers of this article do not think that this argument is correct. They state that participants were interacting with a simple text on a screen, and there were no sophisticated input modalities, so the computer never referred to themselves as ‘I’, never referred to the user by name, and never implied that they had feelings or emotions. Every effort was made to create questionnaires that did not suggest human traits. So, it can not be demand characteristics that encourage the social responses.

The authors think that their conclusions are rather strong and that they have provided strong counter arguments against the alternative explanations. They do suggest that further research should focus on individual differences. Further research should also focus on direct comparisons with human-human interactions. Half of the participants should interact with the computer and the other half should work on a computer, but they should be told that they are interacting by computer with a person who’s sitting in another room. Then, the differences in responses should be examined.