杏吧原创

How can we evaluate complex human interactions clearly and precisely? By recruiting silicon-based research assistants, says psychologist Clifford Nass

鈥淭HAT鈥橲 ridiculous!鈥� exclaimed the software engineer, busy working on improving a spellchecker. I was visiting her workplace as a consultant on the strength of my reputation for making computers easier, more effective and pleasant to use. Because I had come to feel that users needed a kinder spellchecker, something closer to an encouraging teacher than a disparaging critic, I had suggested the system could not only correct mistakes but praise users when they spelled difficult words correctly.

鈥淚 don鈥檛 want my time wasted hearing about everything I do correctly,鈥� she said, scathingly. 鈥淚n fact, if you really think that鈥檚 a good idea, why doesn鈥檛 the computer go all the way and tell users their spelling is improving even if it鈥檚 actually lousy?鈥� The engineer thought she was being sarcastic, but the lead designer thought it was a brilliant idea. Before this had a chance to turn nasty, I moved the discussion to the subject of flattery in general. Do people like flatterers? Do flatterers seem insincere or insightful? Is flattery ignored or appreciated?

Since we found so little agreement, I decided to read the social science literature. After completely failing to find clear answers, I emailed my large network of social science researcher friends to see if any of them would run a study on flattery for me. To my surprise, they refused. Why?

Over a cup of coffee, I thought back to my undergraduate days when I avoided courses that didn鈥檛 focus on mathematics or science. I loved the idea that through ruthless clarity and ignoring unimportant differences, one could discover rich and complex rules. When I studied sociology as a postgraduate, I focused on areas like the information economy and macro-organisational theory, domains that provided objects to study that could be explained and predictions that could be made without oversimplification.

But just as many humanities professors have unpublished novels stashed in their desks, I had a worn copy of The Social Animal by Elliot Aronson, the social psychologists鈥� bible. My fear was of being unable to cope with the incredible variety among people: how could anyone distil social relationships to their essentials when people differed in so many ways?

The social scientists confirmed my misgivings. For an experiment to be clean and compelling, they wrote back, researchers must keep everything constant except for the characteristic they want to study. One strategy they tend to use is to hire a 鈥渃onfederate鈥� researcher, who appears to be just another participant in the experiment. Imagine, though, the difficulties maintaining the same facial expression, tone of voice, body language 鈥� whether the confederate was interacting with an attractive person, an ugly one covered in tattoos, an obnoxious jerk, a woman who looked like their mother, and so on. Worse, the characteristics of the confederate would compound the difficulties: after all, flattery has a very different meaning coming from a smiling person than from someone who is frowning, from a woman versus a man, and from someone in lab clothes versus someone in ordinary attire.

I felt crushed. All I needed were answers to what looked to me relatively straightforward questions about how people feel, behave, and think 鈥� surely the core of social science. The big problem was to find a reliable, unchanging, consistent, compelling confederate, someone who was social but not too social, able to carry on a constrained conversation without seeming contrived, and above all, able to make the interaction feel as natural as possible.

As my research with computer-human interaction continued, I discovered people were interacting with computers using the same social rules and expectations that they use when they interact with other people. For example, if they used computer software and were then asked to evaluate it using the same machine, they said nicer things about the software than if the evaluation was done using an identical computer across the room. They even felt team affiliation: give the participants a blue wristband, put a blue border around the computer and tell them that it is part of 鈥渂lue team鈥�, and they will like the computer more and think it is cleverer than when the computer has a green border. They will even gender-stereotype the computer: a female-voiced computer was seen as a better teacher about love and relationships and a worse teacher of technical subjects than a male-voiced computer.

鈥淧eople feel team affiliation for a computer鈥� they will even gender-stereotype it鈥�

After replicating a large number of social science findings unrelated to flattery, it slowly dawned on me I had the perfect confederate. Computers can do the same thing 24 hours a day, seven days a week, without deviation. They aren鈥檛 influenced by subconscious responses or unintended observations about the partner in their interaction. Without markers (faces, voices) of gender, age or other demographic characteristics, one computer is much like another. I could control out the complexity of two people interacting while ensuring the interactions were remarkably social.

Flattery became my first computer-as-confederate research topic. To discover whether flattery is distasteful brown-nosing or an effective strategy, I had people play a game with a computer. The computer then provided feedback on the participant鈥檚 performance. We told one group that the automated feedback they would receive would be highly accurate and based on years of research into the science of inquiry. A second group was told that because the evaluation software hadn鈥檛 been written, their feedback would be random and not connected to their performance. A third group received no feedback at all.

We found people are suckers for flattery: the group with the ostensibly accurate feedback and the group with randomly positive comments both ended up feeling equally positive about their performance, themselves, and even the competence and likeability of the computer, with both groups feeling much more positive than the members of the group with no feedback. The sole difference between the first two groups was that the first thought they were receiving genuine praise, while the second thought they were receiving flattery unconnected to their actual performance.

So did the flattered participants in the second group just forget that the feedback was random? 鈥淣o,鈥� they replied. A few even wrote a note to the effect that only an idiot would be influenced by comments that had nothing to do with their real performance 鈥� and these were graduate students in computer science! This suggests the following social rule: don鈥檛 hesitate to praise even if you鈥檙e not sure the praise is accurate. Those who receive the praise will feel good and you will seem thoughtful and intelligent for noticing their wonderful qualities 鈥� whether they exist or not.

My team at Stanford University, in Palo Alto, California, have now found almost 100 social rules that can make even the most socially inadequate person (or computer) seem more likeable, persuasive and competent. Using computers as confederates, we鈥檝e studied a number of highly charged areas such as employee evaluations (avoid the criticism 鈥渟andwich鈥� typical of workplace evaluation 鈥� extensive praise, blame, brief praise 鈥� and praise much more than is natural), and the complexities of emotional interactions (misery loves miserable company).

Even though the findings occur in the context of human-computer interactions, I believe the computer-as-confederate has freed me to address fundamental questions about human interactions with each other while remaining true to my desire for clarity and precision without oversimplification. Ironically, to experimentally study the quintessentially human, social scientists may well need to turn to a machine.