WolframAlpha Briefing

Attended the WolframAlpha briefing today, which consisted of an overview by Steven Wolfram and a demonstration of his very ambitious interactive knowledge system. The system will be publicly released free in a few weeks.

You type in a free form query. The system analyzes your language (English only at this time) and interprets it and looks for data it has about any of the terms you used. Makes assumptions about the meaning of terms but allows disambiguation. Then it looks for algorithms that relate to the terms you used and performs them. Again using likely relationships. Produces an in-line report based on your query.

This is a little like when you type on the Google command line something like 'How many quarts are in a peck?' and you get the conversion. But WolframAlpha takes this much further, seeking to interpret complex language and have many kinds of science and data embedded in the system.

Initially the system will be fluent in only some categories of data. Wolfram's demo showed a number of areas where it lacked knowledge and failed. It uses public data only, but he suggested that enterprise data could also be included under license. The idea of tailoring it for a corporation or specific domain is intriguing. Could this be a way of delivering business intelligence to the corporate masses?

The system is not a Wiki, you cannot just add knowledge. He suggested that knowledge will be readily added from expert sources as it evolves.

It is a tour-de-force of Mathematica programming at very least. He suggests it is the first killer application of his New Kind of Science (NKS) idea. I remain skeptical of this in a general AI sense. It is easy to show lots of examples that work, but much harder to be generally useful. We all know language can often be ambiguous. Users will have to learn to re-phrase and resubmit their queries. Is that easier than repeating searches? It is if the algorithms add value as well. In the demonstrations, as can be expected, what seemed to work best were queries about the hard sciences.

After all of that I was more impressed than I thought I would be. I can see its value when focused on some data sources and approaches. It can only evolve. I would recommend anyone who works with business intellgence take a close look at it when it comes out in a few weeks. Steve may have something here.