Do People Want to Talk to Computers?

Do people really want to talk to computers? Let’s explore the question with a thought experiment that allows us to define “talking to computers” in a sophisticated, unrestricted sense. Ask yourself this: If C-3PO, the fussy, fretful, Golden-Droid of Star Wars fame were science fact instead of science fiction, how deeply would I want to talk to him? In other words, would you like to have a chat with C-3PO? Of course, most of us would jump at the chance to converse with that articulate robot. But can we therefore conclude that people really want to talk to computers? There are at least two motivational confounds at work in our theoretical thought experiment. First, the idea of a chat with C-3PO is probably inherently attractive to the star-struck among us and, as good designers of thought experiments, we would need to find a way to factor this variable out of our explanatory equation. Second, even with the aura of the silver screen out of the picture, most folks would still relish the opportunity to talk to the machine, due solely to the novelty of the prospective experience. Maybe we should recast our question more in these terms: If you could carry on an open-ended dialog with a robot that exhibited human-like physical abilities and personality traits, would you frequently like to do so? Casting our question as such, we eliminate the “movie stardom” confound and we can control for some of the inherent attractiveness of novel experiences by specifying that the dialogs would be frequent. What would your answer be? I suspect that most people would still assert some interest in sharing a conversation with such a hypothetical machine. Yet if this machine really existed and it could be used in an actual behavioral experiment, I would predict that the human subjects participating in the experiment would quickly lose interest in open-ended human-computer interlocution. Furthermore, understanding this phenomenon could be an important factor to success in the speech industry. Why would the frequency of human-computer conversation wane? In order to answer this question, we would be wise first to consider why the frequency of human-human conversation, under normal social and psychological conditions, is remarkably consistent. Basically, we must ultimately ask, “Why do people talk to people?” Paraphrasing B. F. Skinner from his controversial 1957 book Verbal Behavior, language, in all of its forms, is socially mediated operant behavior. People talk to people because, by doing so, they obtain reinforcers, particularly social reinforcers. That is to say, we talk to each other for the consequences of talking to each other. Conversation is always for a reason: it serves a purpose; it obtains a consequence. Sometimes the consequence is an actual primary reinforcer like food: a toddler might emit the utterance “dodos” because when said in the presence of its parent, it results in the appearance of some “Cheerios”. More often, particularly as the child’s language skills develop, the consequences of speech are social reinforcers, such as the attention of a friend or an attractive companion. Ironically, the actual content of a speaker’s utterances is often less psychologically significant than the social consequences of the utterances: a teenage girl might discuss football at length and with animated interest when in the presence of the quarterback that has caught her eye. Even if people could converse with C-3PO, interest would eventually wane because, compared to other people, C-3PO cannot compete as a source of social reinforcers. Thus people are unlikely to carry on open-ended conversational dialogs with computers until doing so provides access to the universe of reinforcers that are present when people talk with other people. So, all this psychobabble aside, do people really want to talk to computers? The answer, perhaps surprisingly, is a resounding “yes”. A world of research, both technological problems and social-modeling problems will have to be solved before computers can be as conversationally engaging as their human conversational competitors. But this end-all goal is irrelevant to the very real capacities of existing technologies to support reinforcing conversational interactions. The fact is, current technologies, if deployed in psychologically sound designs, can be sufficiently reinforcing to ensure recurrent, even enthusiastic use. The key is to design interactions that provide a progressively rewarding experience while adequately compensating for their own conversational shortcomings. People will gladly talk to computers, as long as talking to computers remains a rewarding experience. And those who focus on what the technologies do well today, as opposed to the promise they may someday hold, are likely to enjoy great success in the speech industry.