Emerging technologies: the social interface model.

It's tempting to write off Microsoft's "social interface" (originally embodied in the Microsoft Bob shell program) as a truly dumb idea that no amount of Redmondian marketing muscle could salvage. In fact, Microsoft itself seems to have put the idea on a back burner. Two years ago, Bill Gates insisted that social interfaces were "the next major step in interface design." Then Bob took a nose dive and Gates switched to predicting that Web browsers would be the next great interface model.

But we suspect social interfaces are far from dead. Gates himself may be more interested in leading an Internet blitzkrieg, but Microsoft still supports an active research effort on man-machine interaction (the so-called Persona project) and hosts an annual Lifelike Computer Characters conference in Snowbird, Utah for social interface academics. And in a sense Bob's quasi-intelligent cartoon characters have gotten a promotion: They are now Office 97 "assistants," smarter, more responsive, and certainly less obnoxious than their smart-ass predecessors.

Moreover, the new generation of smart characters embody some of the most interesting and advanced technology to be found in Office 97. The Office assistants understand natural language questions, they learn by observing a user's patterns of behavior, and they can propose trouble-shooting solutions by using Bayesian reasoning techniques (a kind of algorithm that assigns probabilities to various types of evidence). At least one of the nine Office 97 assistants, the Einstein character, demonstrates very credible human animation--something that Bob's developers two years ago felt was too demanding for the prevailing hardware and software installed base.

It seems clear that the next few generations of assistants will get even smarter. Social interfaces have become a convenient repository for much of Microsoft's most advanced work in such areas as speech recognition, artificial intelligence, reactive 3-D animation, and speech synthesis. (A surprising amount of this work has been published on a little-known Web site that Microsoft maintains at www.research.microsoft.com.) These technologies probably need another generation or two of processor evolution to function on a standard desktop PC, but--unless Moore's Law finally hits a wall--it's quite possible that we'll be carrying on a reasonably intelligent conversation with our computers some time within the next four or five years.

Which raises the real question about social interfaces: What will we talk about with our PCs? One scenario, which comes from researchers on the Persona project, envisions computer assistants that "ask questions to clarify their understanding of an assignment, describe their plans and anticipated problems, negotiate task descriptions to fit the skills and resources available, report on progress, and submit results when they become available." In short, software assistants could become model corporate employees, doing most of the busy work that occupies live human beings (who presumably will now have more time to hang out at the water cooler and engage in real social interaction).

The trouble with this vision, at least to us, is that social interfaces tend to appeal to the least experienced PC users. The fundamental idea behind the social interface idea (as described by Stanford behavioral researchers Byron Reeves and Clifford Nash) is that human beings feel subconsciously that computers are a kind of intelligent life form, not necessarily alive but somehow smart enough to communicate on human terms (Softletter, 1/17/95). So it makes sense--good interface sense--to offer novices an environment that conforms to their expectations.

But once a user starts up the power curve, those expectations change. The computer quickly becomes more of a tool to be mastered, and the limits of its ability to participate in high-level judgments become much clearer. It's hard to imagine that serious users will often "negotiate" with their word processors, or that they'll want traditional desktop tools to "clarify their understanding" of a task, any more than a carpenter wants a hammer to understand hitting a nail. In the end, expert users will probably not interact very often with the artificial personalities that inhabit their computers.

That's pretty much the early feedback that Microsoft has heard about its Office 97 assistants: Novices appreciate the helpful characters, but experienced users simply turn them off. Novices who bought Microsoft Bob had a similar reaction. One of the most common complaints about Bob was that the software applications that came with the shell program--such as a checkbook program and a mini-database--were underpowered. The social interface experience was interesting and helpful at first, but ultimately Bob users decided they wanted software that did something more practical.

In fairness to the social interface enthusiasts, of course, it's always tough to predict exactly how a new interface concept is likely to impact personal computing. Lots of smart people once insisted that the graphical interface of the Macintosh (and later Windows) merely got in the way of "real" command-line interaction with their computers. And only now, some two years into the new age of the Internet, are software developers beginning to see that browsers and HTML tools have profoundly transformed the way almost everyone thinks about PCs. We're not entirely convinced that social interfaces and artificial personalities will have a similar impact--but we do think there's the seed of an idea here that might just be the catalyst for some important new applications and start-up companies.