Cuil made the headlines in all the right ways: written mostly by a rare woman in technology, created by former Google staffers, claims to fix the problems that Google has and index more sites. Of the above, what stuck with most experienced observers was the claim of a better type of search. Cuil claimed to analyze page content, not popularity measured by number of inbound links as Google does, and that appeals to all of us who have laughed out loud at the results returned by Google.

Like the X-files, search technology is an exercise in seeking the unbelievable: that non-artificial intelligence methods can get us reliable search results. After all, the basic search -- does this phrase exactly exist in this collection of documents? -- has been around since the 1960s. Improving on it has been a fool's game, at least until Google "thought outside the box" by trying not to find the right pages but the most popular ones, a content model that extends to Wikipedia, democracy, eBay and other recent concepts.

What makes Google triumphant is that in a time of possibly bad options, it was a better than mediocre option, and continues to be. For most searches, you'll get WikiPedia -- plagiarized from better sources by an army of bored graduate students and drugged couch-surfing basement dwellers -- and the five most popular sites by inbound links with your approximate term. So far, good: it works, albeit badly in some circumstances, but it works well for the 90% who are searching for relatively cleanly defined things, like "Jessica Alba nude" or "nVidia GT 6800 driver."

The Google method isn't perfect, but its opposite, what we can call the "Altavista method" (remember them?) is to send a smart crawler into each page and to try to figure out based on what's on the page what it's about. Several companies have tried this. None have come up with a reliable method because this requires two tasks: first, parse the content, and second, assess its relevance. To do this correctly requires a smart computer that also has general topical knowledge and a good rule of thumb of how useful something actually is to a human.

We want to believe something like Cuil can exist, and succeed, but much as Google was a hack, Cuil is going to be a hack because it's a series of semi-smart scripts approximating what an AI would do, not being an AI themselves. We want to believe it can succeed, and in some areas, it has succeeded; searches are generally more relevant than Google for topics off the beaten path. It may not yet be there, but some muffled part of us wants to aim higher than for compromise, and we can thank Cuil for waking that up in us.