It has been a while since there has been something new on the search scene. There was one highly publicized, incomplete junk that came out a few months ago called cuil (pronounced cool). As the name portends, it was not. At all.
Anyway - a new site called Wolfram|Alpha (name rather unfortunate) - is the new kid on the block bringing forth a really different approach to search. Instead of searching through pages to suggest potentially relevant pages, WA tries to answer quantitative questions with just the answer. If you need to know the population of the world, you just ask 'world population' and you get the answer of 6.53 billion people.
Search is tough. Understanding and algorithmically analyzing human fuzzy interaction is inherently difficult. And WA is trying to do two things at once - understand the fuzzy world-wide-web and obtain facts from them. This has long been the goal of a semantic web, that is nowhere near reality today. At the same time WA is trying to understand user searches to generate quantitative queries that can then be applied to the data that it has collected earlier.
First impressions - WA seems to be doing an ok job on the two entity intersections. If you are looking for 'world + population', you are good. Or you are looking for 'India + mobile phones' you are good too. But trying to do an intersection of 'world + population + mobile phones' seems to trouble the search engine.
Another disappointing aspect is its inability to interpret date and time as a dimension to queries that seem to work well on 'today'. For example, searching for gold + price works well. But trying to search for gold + price + any date doesn't compute. This seems to work for dow jones + any date, but not for gold + price. See similarities to the third level intersection problem observed above?
What it seems to be doing a good job is on the roll-ups. Try searching for Asia + cellular phones. You not only get the total estimate for Asia, but you also get a list of all the countries with their estimated cellular phone populations. Pretty interesting eh?
All in all, WA looks like a good for an alpha. It seems to be able to do simple queries and roll-ups. Not really good with anything complex, not to mention numerous glitches in the UI of the site. Also interesting will be the response from rights holders to their data being used by the engine. Granted 'facts' do not fall under the purview of copyrights. But what would happen if, say, results from surveys started to be incorporated into search results? That is if WA can one day show you the percentage of all people in United States, having cell phones with AT&T service expressing satisfaction with their service in a survey. What will the survey owners think of that? What will AT&T think of that?
All in all, am really excited to see where this goes. Here to the semantic web, without having to work hard at it.