Why Powerset (unlike Snap, Kosmix, Clusty, and Eurekster) will beat Google
High on the "will this startup tank after five months" checklist — I think it's number 14 — is "Does this company want to be the next Google?" Would-be search innovators fail for several reasons:
- They make the user click more than once. Searching isn't like browsing — users need to type a phrase, hit enter, and have all their best options one click away. That's why Ask.com's clustered Teoma search can only get press in USA Today and not a respectable tech outlet. It's also why Clusty failed to catch on despite constant hype.
- Portal creep: They have so many topical searches that they feel like directories. I haven't heard a thing about Kosmix, for example, ever since they took over $7 million in venture capital funding in January.
- They're too fancy for their own good. Snap.com's auto-preview feature is snazzy, but it's too slow. Maybe it'd work in two years, but can Snap wait that long?
- They try to get social. Please, I'm not going to convince my "community" to use Eurekster Swicki, and even when I do, I don't care what searches they use — basic search is not social. Tagging is, and that's what Yahoo's Del.icio.us already does.
- They're research projects that got caught up in Web 2.0 hype. See TagCloud, which readily admits this on its home page.
- Their only publicity is a spot on the Web 2 List.
So who the hell's Powerset and why will they beat Google?
Powerset's co-founder echoes many other would-be search kings when he calls Google search terms a "grunting pidgin interface." But he's actually thought of the answer, judging by a profile on news site VentureBeat: Take the phrases people already search with, and make them work.
Most Internet users have learned the halting language of search terms, but when AOL released the records of millions of real searches, it revealed that many users still try phrases.
The failed-or-failing search engines above come at search from the wrong end — they thought of a clever method and then justified it. Powerset tackles it the other way round: They want to parse the English language so the average phrase returns results relevant to it, not to its constituent words.
VentureBeat uses the search "books by children" as an example. Entered without quotation marks (as most users type it — people never learned to use quotes), the phrase gives sketchy Google results.
But it sounds like Powerset's ambitions reach much further than a three-word phrase. Co-founder Barney Pell, says VentureBeat, built a specialized language system that could take queries like "What were the top five products ordered over the past week?" and return a table of results.
Now, that would be a coup for general search — most impressive language systems, including the one Pell built, only work within strict topical boundaries. That's why when Ask.com tried to build a search engine using real questions, they opted to handcraft answers rather than automate the whole process. The load was just too much, and finding the real question among Ask's options was no better than poring through Google results.
But Powerset wants its system to truly understand those phrases. Such an engine would be an artificial intelligence. If the company can pull this off, it has a shot at rescuing the world from speaking Search Grunt.