/* ---- Google Analytics Code Below */

Thursday, March 05, 2009

Wolfram Announces Intelligence Effort

I was a long time user of Wolfram's Mathematica in the enterprise. I also read parts of his epic book: A New Kind of Science (NKS), which is now online in its entirety. Neither of these two sources, though remarkable, are very friendly to the average business user. Mathematica requires the knowledge of a modeler. NKS understanding requires mathematical knowledge and the perception of a physicist.

In today's Wolfram Blog Stephen Wolfram announces a new mega effort called WolframAlpha to be released in May. He writes: " ... I had two crucial ingredients: Mathematica and NKS. With Mathematica, I had a symbolic language to represent anything—as well as the algorithmic power to do any kind of computation. And with NKS, I had a paradigm for understanding how all sorts of complexity could arise from simple rules ... ".

What he talks about next is the ability to combine existing data and algorithms to create useful knowledge. " ... It’s going to be a website: www.wolframalpha.com. With one simple input field that gives access to a huge system, with trillions of pieces of curated data and millions of lines of algorithms ... ".

What exactly this means is still unclear. As a former AI practitioner, this appears to mean that you will be able to enter something simple, like a natural language request and that the system will combine data and algorithms to produce results. He does not mention AI, but implies it. Semantic tagging is also mentioned as a method being used.

He goes on to say:" ... I’m happy to say that with a mixture of many clever algorithms and heuristics, lots of linguistic discovery and linguistic curation, and what probably amount to some serious theoretical breakthroughs, we’re actually managing to make it work ... ".

This is a project I will watch closely. It suggests something very big. Until we can see if natural language queries can be unambiguously entered by a non mathematician and converted into a algorithm and combined with mega data sets to provide useful results, it is hard to say. Or is this just a way to search for the parts and you as a modeler get to assemble them? In practice the assembly is usually the much harder part.

My own artificial intelligence experience in the enterprise says that the former is a very hard thing to do. If anyone could do it, and has the tools to work with, it would be Wolfram and his group. I remain skeptical.

I have asked for a look at the package, if I get such a chance I will follow up with an early look.

Update:I was left still pondering about how research from Wolfram's 2002 NKS book could be applied to this kind of problem, since it was mostly about how simple automata could be used to model physical processes. So I found my paper copy of his book to see if I had missed anything. Not unlikely since it is 1100+ pages long. No, not that I found with a quick look, though I guess anything can be modeled by anything ... it is just not efficient enough to use. Please point me to anything that may be insightful.

No comments: