21st: Tough Room for the 'Toons

Despite the boos and catcalls, Microsoft keeps sending its animated little helpers out into the spotlight to perform their artificial-intelligence tricks.

By Andrew Leonard
Published May 1, 1997 11:00PM (UTC)
main article image

There isn't a tougher gig in software show business than the role of lead cartoon in a Microsoft production. In 1995, the debut of Microsoft Bob, a package of easy-to-use software applications with animated helpers -- the "Friends of Bob" -- was greeted with near-universal derision. The marketplace rejected it with a resounding thump. Most companies wouldn't have given the project a second look.

But not Microsoft. Microsoft is nothing if not persistent. It's back with a sequel to Bob in the form of Office97's Office Assistant -- a stable of cartoon characters who serve as the front end to Microsoft's interactive help system. And once more, the reviews on the Net haven't exactly been raves.

"Too cutesy." "Obnoxious." "Pesky." "Annoying." "Rude." "Abysmally cheerful." To say that not everyone is crazy about the critters would be a slight understatement.

"After a couple of times, Mr. Paper Clip stops looking cute," wrote Barry Connolly in the microsoft.public.office.misc newsgroup. "In fact I rather think he starts looking snide, sinister and perverted."

"By the time I hit the 'Help' button," says another dissatisfied customer, Erinnyes Dragon, "I'm already irritated because I can't figure out how to do whatever it is I'm trying to do. Having that sappy, winking, grinning, boinking thing capering on my screen is like a slap in the face."

Cuteness isn't the only issue. The Office Assistant is merely the tip of the cyberg. Behind it lurk an ambitious framework of artificial intelligence (AI) technologies that dramatically raise the stakes for all computer programs. The Assistant is just a rudimentary stab at solving some age-old AI problems, but we'd better pay attention: Judging by Microsoft's record, it won't be going away, no matter what the critics say.

The Office Assistant does represent progress in some key arenas. It is interactive. If questions are phrased properly, it can offer decent help. The interface is also much improved. The paper clip and other incarnations of the Assistant (an Einstein cartoon, a paper cat, a buzzing "hoverbot") are far less insulting to the user's intelligence than the moronic Friends of Bob were.

Representatives of Microsoft shrug at the carping. Steven Sinofsky, who led the Office97 design team, dismisses Usenet barbs as "the least indicative of the vast majority of data points (on the Office Assistant) from purely a statistical point of view in terms of the sample size and cross-section of users."

"Overwhelmingly, the feedback we have gotten on Office97, specifically the Assistant, has been very positive," says Sinofsky. "Several corporations have told me personally that they are upgrading to Office97 because of the Assistant."

Microsoft is convinced that there is a future in the so-called social interface. Citing research that indicates users "love" interacting socially with their computer, Sinofsky argued that "people want to treat a computer like other people."

They also want to be able to escape, which Sinofsky conceded is one of the most important things Microsoft learned from the Bob debacle. The Friends of Bob could not be turned off; the Office Assistant can, although not without some effort. Microsoft has indeed taken a step in the right direction.

But it has also made a step in a decidedly Orwellian direction. The hype for the Office Assistant declares that it anticipates user needs, that it monitors what the user is doing so as to be able to suggest help before the user actually asks for it. The Assistant is watching you.

Eric Horvitz, a research scientist at Microsoft who specializes in the underlying technology that the Office Assistant uses to accomplish these functions, downplays the ability of the Assistant to do really intelligent work in its current incarnation. There isn't really any formal natural language processing occurring in the Office97 Assistant. Right now, the Assistant -- employing a modeling technique known as Bayesian inference -- guesses what the user might be asking about based on statistical probabilities associated with keywords. And other Assistant features, such as the option to have it provide periodic tips, are essentially randomly generated.

But there's no question that Microsoft is headed in an artificially intelligent direction. As Horvitz stresses, this is only "Version 1.0." Future revs of the Assistant will be monitoring our every keystroke, tracking how long we "dwell" on a particular menu or toolbar icon, pondering what kind of questions we have asked in the past, and guessing what we might be trying to do in the future.

Some software consumers might find such features the answer to their dreams. Others will no doubt be dismayed at the concept of constant software surveillance. Regardless, nearly everyone will find their expectations raised by the promises Microsoft makes for its software. And that's a big reason why Bob, the Office Assistant and all their future descendants have such a brutal gig. As soon as a software program starts using normal speech and is promoted as understanding your needs and desires, you expect a great deal more from it. When the software fails, you will be correspondingly more upset. Unsolicited bad advice is much more annoying than no advice at all.

Horvitz and Sinofsky stress that current user complaints are mostly the result of immature implementations.

"Keep in mind that the Assistant is really very much a first pass," says Horvitz, "and is quite limited from the point of view of goings on in research."

"No feature is ever done, and the Assistant is no different," says Sinofsky.

Of course, this is exactly what AI scientists have been saying for 40 years: Just wait until the bugs are worked out and everything will be fine.

"The better tools we build for managing the Assistant and driving dialogue based on it, and coding things like query expansion," says Horvitz, "the better the tools will become. AI has been a tough challenging battle for a couple of decades, but I think we are pushing the frontier. We are doing some wild projects that the free economy wouldn't support directly. There is a lot of innovation going on and we are going to come up with some breakthroughs. I'm very optimistic."

They always are.

Andrew Leonard

Andrew Leonard is a staff writer at Salon. On Twitter, @koxinga21.

MORE FROM Andrew LeonardFOLLOW koxinga21LIKE Andrew Leonard

Related Topics ------------------------------------------