Going along with the crowd is the American way: We flock to make the same choices others have already made. Four out of five dentists recommend Crest! Nine out of 10 shoppers choose Kellogg's! See the movie the rest of the country is seeing this weekend! Every marketer knows that in public spaces, people naturally flow to the storefront that's already got a crowd in front of it. Why should we expect the Internet to be any different?
A couple of new software tools aim to translate this behavior pattern to the Web -- and bring some order to the medium's frantic chaos -- by relating the choices you make online to those made by hordes of other people. Alexa is a compact utility program that sits next to your Web browser; it tracks your travels and offers suggestions for where you might go next -- based on where previous visitors to the same page have decided to proceed. Direct Hit, a new search technology company, is applying a similar logic to the frustrating experience of using Web search sites. Most search sites currently spit out thousands of results to any simple query, requiring you to sift through endless lists of marginally relevant pages; under Direct Hit's approach, when you search for, say, "yahoo," you'll get results based on the sites that previous searchers actually chose to visit.
Which, of course, is great if you're actually looking for the Web's phenomenally popular site by that name. But what if you're a fan of "Gulliver's Travels" and are seeking info on Jonathan Swift's Yahoos, instead?
The logic behind this new software -- which might be called "groupthink-ware" -- is unimpeachable: Technology has created a chaos of data proliferation; why not use technology's own data-collection capabilities to organize it?
Why not let other people's searches and paths guide yours? Here are two good reasons -- one pragmatic and one philosophical.
The practical problem is the one that's typical for Web technologies: The system rarely works the way it's promised. Direct Hit's software isn't publicly available yet, though the company says some major search sites will be offering it later this summer, so it can't be put to the test. Alexa has been in distribution for months now, though. The early results I got from it were spurious, and I hesitated to write about it for a long time: A product like Alexa needs time to mature -- it's only as good as the database of previous usage it collects.
Today Alexa is nearly a year old. It is much praised, and deservedly, for its innovative interface and its mold-breaking approach to Web navigation. Its site recommendations are only one feature of a more complex design that also involves access to a vast Web archiving project. (The name is a cheeky reference to the library of Alexandria, the Hellenistic era's version of the Library of Congress.)
But the site recommendations remain the chief benefit Alexa offers most users -- and, alas, they remain awfully spotty. In theory, Alexa ought to provide the most reliable recommendations from heavily traveled sites, where it has a reasonably big pool of data to draw from. So why, when I check the technology-news headlines at CNet's much-trafficked News.com, does Alexa suggest I take a look at the entirely non-technology-oriented Worldwide News Agency of Agencie France Presse, or the entirely non-news-oriented page called "Pack Tools," "an in-house tools page for the UCSB Many Wolves authoring collective"?
Presumably because some statistically significant group of users who preceded me went on to those pages from CNet's. But who are those users? And what do I have in common with them? Alexa's approach to Web navigation may at first seem to bank on the same principles that govern the world of "collaborative filtering" software from companies such as Firefly and NetPerceptions -- tools that match a profile of your likes and dislikes in, say, music and books with those from a database. But the collaborative filtering tools depend on your providing reasonably extensive information about yourself; the more they know about you, the better they can serve you (and the better their databases can serve others). Alexa, unlike the collaborate filters, has only one data point to go on: the page you're reading. That's not enough to provide genuine personalization.
Today, evidently, the Alexa database remains too small to provide reliably useful recommendations (though it's certainly capable of coughing them up some of the time, and it remains a fun little program). But a different problem looms if and as Alexa becomes more popular. If millions of people embrace it, it will find itself promoting an ever narrower set of least-common-denominator choices -- road signs to the most well-trod paths across the Web. And if Alexa, or something like it, becomes really popular, it could get dangerous: Its recommendations could create a self-reinforcing feedback loop, giving mass preferences a snowballing momentum that could never be displaced. What began as a tool for enriching the experience of using the Web would then devolve into a structure for narrowing it.
A few years ago, an ill-fated company (led by an exec who went on to lead the Microsoft Network) tried to produce "interactive movies": They'd wire up an old-fashioned movie theater with pistol-grip controls and let the audience vote on what should happen at key plot crossroads. It was a dumb idea in many ways, but the ludicrous failure taught some interesting lessons. When a group is large enough, and you have some historical data on its previous choices, you can pretty much predict its future behavior under the same circumstances.
For "interactive movies," this meant that 50 people in a theater would always choose the plot-branch that offered a hint of sex. The movie wasn't interactive at all; each crowd always made the same choice. For the Web, there's a similar principle: individual choices exhibit all the quirks and unpredictability of our unique personalities, but aggregate choices have a dull sameness to them.
That's why I don't pin much hope on any software that offers recommendations based on mass behavior. It's less likely to be a helpful tool than a popularity contest whose outcome you know in advance. And surely the Web has enough of those already.