March 11, 2009

Gird Your Loins Google, Here Comes Wolfram Alpha

Michael Feldman

Last week, Mathematica inventor Stephen Wolfram announced that he would be launching a new kind of Internet search engine in May, with the not-so-modest name of Wolfram Alpha. Actually it isn’t a search engine at all — it’s more like a fact-finding engine. Wolfram himself calls it a “computational knowledge engine,” which is meant to convey the idea that the software will be computing results, rather than just indexing Web pages based on keywords.

So presumably you could ask Alpha a question like “When was the last time SGI’s stock was above $10 a share?” and Alpha would go fetch — sorry, compute — the answer. (By the way, the answer is Oct. 8, 2008.) Usually queries like that would only be possible with highly-specialized database applications. Alpha aims to generalize that kind of capability using only the unstructured data that exists on the Web. If Wolfram has really succeeded in doing this, it would introduce an entirely new kind of Web interaction.

Well, maybe not entirely new. Even Google lets you do simple math calculations in its search box and allows you to find definitions of words by using a “define:” prefix on a keyword. But the scope of Alpha is much larger.

Wolfram provided few details on how his new software would be implemented and made no mention if he had a supercomputer in his basement to do all this computational heavy lifting. The basic idea is that you would present a question to the Alpha box (which coincidentally looks a lot like Google’s search box). Alpha would then untangle the semantics of the question, map it to the desired operation and then go distill the answer from the available data.

The last step is the most mysterious part. Whereas some people have suggested that the Web’s data needs to be systematically tagged to make it semantically friendly, Wolfram has apparently taken a different tack. He says the engine will be based on Mathematica and the computational approach outline in his 2002 book, A New Kind of Science (NKS) to provide the engine’s intelligence. In a nutshell, NKS describes a way of applying a general set of methods to computational systems, the idea being that a great deal of complexity is able to arise from a very simple set of rules.

If that’s not obtuse enough for you, Wolfram says the Alpha engine will “explicitly implement methods and models, as algorithms, and explicitly curate all data so that it is immediately computable.” I’m not at all sure what he means by “curate all data” other than reorganizing the Web data on the fly so that it’s more digestible to the software. Any way you look at, Alpha’s going to need a lot of smarts to do even basic fact finding, especially considering that there’s no way to verify the accuracy of data encountered on the Web.

All of this might be passed off as worthless hype, except for the fact that Wolfram has plenty of street cred in the industry. As the founder and CEO of Wolfram Research, he has built a highly successful business based on a very useful piece of software. Before his success in business, Wolfram was an accomplished scientist in his own right, having received his Ph.D. in particle physics from Caltech at the age 20. His NKS book opened to mix reviews, but the ideas presented in the text show he can still plumb the depths of computational theory and mathematical modeling.

I’m anxious to see how Alpha performs in the wild. If it lives up to even a fraction of the hype it has generated, Alpha is destined to become a common Web tool like Google and Wikipedia. And maybe someday, when your son or daughter asks you why the sky is blue, you’ll say: “Let’s wolfram it.”

Share This