Social-Psychological Research

Among the most fascinating of the many experiments that have been conducted on the subject are the ones done by Stanford University professors Clifford Nass and Byron Reeves and documented in their book, The Media Equation: How People Treat Computers, Television, and New Media Like Real People and Places . The following is a distillation of some of their findings.

The professors ran a series of studies to determine if the "rules" of social psychology (that had been tested thousands of times over the past century or so) would hold true not only for person-to-person interactions, but also for person-to-media interactions. They took some tried-and-true social psychological interactions (based on famous experiments dealing with flattery, reciprocity, expert opinion, and so on) and tested them using a person and a computer (or another media device, such as a television). Please note that these experiments did not deal directly with people interacting with speech-recognition systems, but rather people interacting with computers in a social way. Their results showed that people interact with media in the same way that they interact with other people and these media become social actors (that is, entities that emulate aspects of human behavior). Testing later conducted by the professors on speech recognition bore out the conclusion that speech-recognition systems are yet another media with which people interact adhering to the rules of social-psychological interaction.

The Flattery Experiment

In one experiment, when a computer randomly flattered people using a program (displaying words on the screen, such as "good job"), the participants in the study still felt flattered by the computer ”even when they knew it was only the result of a randomizing program.

What are we to make of this? That lots of flattery in a speech-recognition system is a good thing? Well, no. I think any person subjected to prolonged episodes of random flattery will eventually be less moved ”or perhaps even increasingly irritated ”by such unwarranted praise. But it would be a different story if we designed a system to do what we would expect real people to do, that is, to flatter someone when that person has endured some arduous task to completion or made a sound, well-informed decision.

For example, if a person took the time to help a speech-recognition system learn more about their preferences, then perhaps a bit of flattery would be in order. I would also consider adding some flattering remarks to a speech-recognition prompt if a caller agreed to allow the client's affiliates to send them marketing material.

But I wouldn't stop there. I would then go on to do what real people also do: explain why I'm flattering the caller, as in the following example.

SYSTEM:	You did a great job of rating all those products! Would you mind if we released your name , e-mail address, and preferences to some carefully selected partners of ours so that they can send you advertising?
CALLER:	OK.
SYSTEM:	That's great ”we appreciate your allowing us to send you marketing material. Partner advertising helps us keep our costs low and pass along the savings to you.

The Reciprocity Experiment

When someone helps another person (either by giving them information, something tangible , or assistance), the person who has received help will feel obliged to reciprocate the aid. The experiment that tested this principle proved that people would try harder at a task if a computer only tells them that it is trying harder as well ”even if the computer doesn't change its behavior at all.

Now, just for the record, I don't believe people ”or even computers ”should make a habit of lying, but as Machiavelli once pointed out, the end does occasionally justify the means. It's particularly useful when a system needs to motivate users to try harder at a task.

For example, in the Wildfire Communications, Inc. speech-recognition system, Wildfire, callers can teach the computer how to recognize their names when they call the system. If the system experiences difficulty recognizing a caller, it will ask that caller to use another method of identification (for example, saying or entering a phone number). At this point, the system attempts to learn once again how the caller pronounces his name, so that the caller won't need to enter his telephone number again in the future.

WILDFIRE:	Sorry, I didn't recognize you. Would you like to take a moment to teach me how you say your name? Yes or no?
CALLER:	Yes
WILDFIRE:	OK, please say your name .

The system goes on to collect the name a couple of times before saying, "Thanks, that will help for the next time you call."

Now, in reality, Wildfire does actually modify the way it recognizes the caller. However, when people call in and attempt to identify themselves the system may still not successfully recognize them. But, as this example shows, callers may be motivated to try a task again if the system asks them and promises to reward them in the future (in this case, by not requiring the caller to say or enter a phone number). The promise of future benefits establishes reciprocity ”"scratch my back and I'll scratch yours" ”and that invites cooperation.