Updated: Stephen Wolfram, creator of the Wolfram/Alpha search engine, on Tuesday demonstrated his much ballyhooed "computational knowledge engine" at a talk at Harvard University. Wolfram likened his effort to reproducing a global reference library and said Wolfram/Alpha will launch in "a few weeks.
Updated: Stephen Wolfram, creator of the Wolfram/Alpha search engine, on Tuesday demonstrated his much ballyhooed "computational knowledge engine" at a talk at Harvard University. Wolfram likened his effort to reproducing a global reference library and said Wolfram/Alpha will launch in "a few weeks."
Wolfram's demonstration was one of the few times his search engine, built on algorithms from his Mathematica project, has been seen since word leaked out about it weeks ago. ReadWriteWeb had a nice overview over the weekend. Wolfram covered multiple topics such as data accuracy, Wolfram/Alpha's business model and the bridge between analytics and search.
One worry: The Wolfram/Alpha search engine seemed slow, according to Wolfram. What will happen when the public starts poking around on it? "Running a little slower than I'm used to seeing," he said. "When it's alive in the world it'll be quite a bit zippier than this."
Is Wolfram/Alpha a Google killer? Probably not. Wolfram/Alpha's approach, however, is notable and it's easy to picture it being used in the enterprise. Wolfram's demonstration could be summed up as an intersection between analytics and generic Web search. Wolfram/Alpha gets you an answer instead of pointers to potential answers. "It tries to tell us useful information based on what it can compute," said Wolfram. "The goal is to provide expert level access to anyone at anytime."
Wolfram also gave some insight to his business model. "This Web site will be a free site. We will have corporate sponsors that will have things on the side here. We know a lot about the specific questions people are asking and know what kind of knowledge people want. There is a lot of vendor information" (that Wolfram can monetize eventually) and a subscription professional site, he said.
Here's a screen from the Webcast, which was buggy most likely due to a bevy of watchers.
"What we're trying to do is take all the things that can be computed about the world...and try and package it to the point where we can just walk up to a web site and have it deliver the knowledge we'd like to have. Like interacting with an expert it will understand what you are talking about, do the computation and present to you results."
Four big pieces are behind Wolfram/Alpha:
Curated data: Free, licensed and feed data. Running through human and automated process to verify the data and make sure it's "clean and curatable." At some point, you need a human domain expert.
Algorithms: Wolfram/Alpha uses a bevy of algorithms in 5 million to 6 million of Mathematica code.
Linguistics: The goal is to interpret free-form language processing. Wolfram said Wolfram/Alpha uses various components and techniques to figure out what people are actually asking. Part of that process is filtering out fluff. "We've been pretty good at removing linguistic fluff," said Wolfram, he said people eventually get to the point where they speak as if they were talking to an expert. "People quickly begin to just type in concepts as they come to them."
Presentation: Algorithms try to pick out what's important to the searcher. Again, Wolfram noted that human-aided algorithms are needed.
Instead of delivering up a bunch of links, the Wolfram/Alpha search engine tries to put a narrative around a user's question and allow them to drill down. Indeed, the result presentation features graphics and other computational features. Think part calculator, part search engine.
Among the demonstrated searches:
Wolfram did a search on the GDP of France and got a plot of the GDP history and some history. The next search had the GDP of France divided by the GDP of Italy and Wolfram/Alpha delivered an answer.
The weather in Lexington, Mass. as a search term delivered a summary of the temperature and plotted it as a function of time.
A search on "medical LDL 180" dove into a public health study showing that level of cholesterol put someone in the 95.9 percentile in the U.S. Further refinements of the search---like "male age 40"---will yield a chart for life expectancy.
How will Wolfram/Alpha do? Probably pretty well for certain applications---especially the academic variety. Meanwhile, the presentation is interesting. At the very least, we can learn from Wolfram's latest pet project.
In the question and answer session, Wolfram addressed data accuracy and said "there will be a mechanism to contribute data, audit data and have it flow into the system." The big question will be how quickly Wolfram/Alpha can absorb the vetted data.
Other odds and ends:
Wolfram/Alpha will have a variety of levels of APIs---presentation, underlying XML for mashups and individual results from its databases and computations. There is a first draft of the API documentation as of March 27.
There is a plan to have a professional version of Wolfram/Alpha to upload data. This move would open an enterprise revenue stream.
Wolfram said his search engine will footnote data where there is scientific disputes.
Here's what Wolfram said when asked about the semantic Web. "If the semantic Web turned out to be an overwhelming thing our job would have been much easier," said Wolfram, who also noted that much of Wolfram/Alpha's data isn't available on the Web.