The Internet still has the capacity to startle. Of late, I've been oohing and aahing over Gnutella, a free technology that allows users to swap files across the Net. Gnutella resembles Napster, the MP3-file-sharing software that has many music executives, artists, and copyright advocates in a tizzy. But unlike Napster, Gnutella (gnutella.wego.com) doesn't limit itself to MP3s--though that file format is certainly a staple there--nor does it rely on any central computer or computers to orchestrate the action. As part of a "fully distributed information sharing technology," every PC running the tiny Gnutella client is part file server (offering selected files, directories, or drives to others within GnutellaNet), part "mini-search engine" (querying other users' computers for data). Architecturally, the open-source Gnutella is elegant and robust; culturally, it's revolutionary, potentially terrifying, and maybe even capable of turning the Internet on its head.

Gnutella was developed by engineers at Nullsoft, an America Online company. (AOL, no fan of the free-content distribution model, was quick to pull the plug on the project, but not before a beta version had hit the Net.) An open-source initiative that just keeps gaining momentum, Gnutella hearkens back to the earliest days of the Net, when shared information and one-to-one contact were all the rage. Those days are long gone. From the Net's high-tech 1969 birth as ARPANET, through the introduction of the Web in 1991, the global network has grown less personal and more centralized. Eventually the graphical browser came along, advertising showed up, portals were born, and audiences started flocking to only the best-known Web sites. The scrappy, friendly, insider's Internet went underground; the Web flourished. For those seeking more democratic days, Gnutella offers welcome redress.

Many critics condemn Gnutella as simply a vehicle for piracy, providing access to files without concern for copyright. Fair enough. I don't want to pooh-pooh those very real concerns, which certainly must be addressed. But Gnutella itself is not the culprit. It's simply a protocol, a morally neutral technology, much like the telephone. Sure, some people use the phone to scam credulous victims or to interrupt our dinner with pitches for products we don't want. Yet the phone lets us call 911 in an emergency or communicate with friends half a world away. Gnutella too has enormous potential to change how we interact. In particular, its future application as a new kind of search engine has plenty of people excited--and many search engines executives worried. In fact, Gnutella's search skills are on currently on display at InfraSearch (www.infrasearch.com), an early "proof of concept" that hints at the huge potential below the surface. (See the CNET News story, "Napster-like technology takes Web search to new level" at news.cnet.com/news/0-1005-200-1983259.html for details). Of particular appeal: InfraSearch can return results from dynamically generated pages drawn from a searched site's own database. First, though, the site must choose to run Gnutella's search agent software. No conventional search engines, which rely on "crawlers" or "spiders" to canvas the Web for data, have InfraSearch 's dynamic page search capabilities, nor can they be as accurate or timely. As the CNET News story reported, no less an authority than Mosaic/Netscape creator Marc Andreessen proclaimed that Gnutella would do "for search what the Internet did for communications."

Now, I hate to disagree with a visionary as credentialed as Andreessen. But if I were a search portal CEO, I wouldn't be overly concerned about Gnutella, which relies on the simple, laudable premise that people can be trusted. It assumes people will do the right thing.

If the last six months of viral infestations, denial of service attacks, and other insidious schemes has taught us anything, it's that good will is in scarce supply. Within the Gnutella universe, all the files swapped, all the pages downloaded, all the images viewed originate from someone else's computer--it's the basis of the technology's decentralized model. For the program to be helpful, users must assume those pages and files are on the up and up. They may not be.

For example, yesterday I was playing around with Gnutella and typed in the word "dinosaur" (I have a 2-1/2 year old at home with a major Stegosaurus fetish). Half the results led to a deliberately mislabeled HTML page that turned out to be an ad for porn videos. No dinosaur fossils were in evidence, though there was no shortage of skin over bone. Or consider the consequences of an increasingly common prank called "Napster bombing." I heard of cases in which an MP3 file purporting to be from teeny-bopper Britney Spears turned out to be a track from rude hip-hopper (and Spears critic) Eminem. How do you know files you're downloading on Gnutella aren't similarly mislabeled, or worse, harboring nasty viruses? If the recent "resume" for Janet Simons could turn out to be a hard-drive destroying worm, imagine what might be lurking in other innocent-sounding files. I'm describing a worst-case scenario, but decentralized content, though tempting in an egalitarian sort of way, is just plain dangerous.

Though InfraSearch is still in rudimentary form, it makes similar assumptions about human behavior. Sites that run the InfraSearch software can "respond to queries the way [they] want." Yikes! We're relying on sites to represent their content honestly. Suddenly I'm imagining marketing departments running amok here, with spin doctors, hucksters, and maybe even larcenous e-tailers getting equal time. Should we expect honesty just because people join a community? Plus, the IntraSearch formula assumes sites will participate, essentially buying into the Gnutella model. Some will; other won't, which will translate into a limited community, at least compared to the vastness of the Internet.

I could be wrong. In fact, I hope I am. I believe in the technology; it's humans I distrust.