Business - Written Sunday, April 26, 2009 by Jeff DeChambeau - 1 Comment

A Bicycle Built for Two-Dot-Oh

Google does a great job indexing words and phrases in unstructured data like web pages and scanned texts, but it isn’t yet able to deal with the concepts that those words and phrases represent. This means that Google is only good at answering questions that already appear in the documents it indexes — and that asking factual questions is more or less a category mistake, one that won’t return very good results.

This reality looks to soon be a thing of the past. In a couple of days, Wolfram (the people behind the MathWorld resource site) will be launching Wolfram|Alpha, a “computational knowledge engine” that can answer factual questions — essentially deriving new conclusions and results from existing web pages and documents (read the engine’s introduction here).  ReadWriteWeb recently got a chance to play with Alpha, and they seemed very impressed:

In today’s demo, for example, Stephen Wolfram searched for “internet users in europe,” or “weather oakland” – two queries that most users would also use in Google or any other search engine.

Where Alpha exceeds, is in the presentation of its “search” results. When asked for how many internet users there are in Europe, for example, Alpha returned not just the total number, but also various plots and data for every country (apparently Vatican City only has 93 Internet users).

Another query with a very sophisticated result was “uncle’s uncle’s brother’s son.” Now if you type that into Google, the result will be a useless list of sites that don’t even answer this specific question, but Alpha actually returns an interactive genealogic tree with additional information, including data about the ‘blood relationship fraction,’ for example (3.125% in this case).

Similarly, I.B.M. has developed a system that is also able to compute knowledge. In an amusing move, they’ll be introducing the technology to the world by having it compete on Jeopardy!:

In a demonstration match here at the I.B.M. laboratory against two researchers recently, Watson (the system) appeared to be both aggressive and competent, but also made the occasional puzzling blunder.

For example, given the statement, “Bordered by Syria and Israel, this small country is only 135 miles long and 35 miles wide,” Watson beat its human competitors by quickly answering, “What is Lebanon?”

Moments later, however, the program stumbled when it decided it had high confidence that a “sheet” was a fruit.

Missteps aside, these are exciting developments, and represent the introduction of computing to a higher level of information. Initially systems could only compute numeric data presented in very strict formats; this remained the paradigm for a very long time. Only recently has the development of extracting in-text relationships from so-called “unstructured data” made forays into mainstream technology — but such data still required human interpretation and understanding. These parallel developments from Wolfram and I.B.M. take the interactions between machines and information that next step further — which, despite being considerably more technically complicated, is likely to greatly simplify human interaction with computers.