Almost one year ago I wrote about Wolfram|Alpha, a Web service offered by Wolfram Research, which the company describes as a “computational knowledge engine”.
Almost one year ago I wrote about Wolfram|Alpha, a Web service offered by Wolfram Research, which the company describes as a “computational knowledge engine”.
At the time I summed up the services as “quite obviously brilliant” with the rider that, as a friend of mine commented, “it’s boring and boringly bad at explaining why it’s boring.”
My main complaints were two-fold: First was the lack of a natural language interface and, second, the odd “holes” in its knowledge.
The service is about to celebrate its first birthday and I recently had the opportunity to talk to Stephen Wolfram, the co-founder of Wolfram Research and the primary motivating force behind Wolfram|Alpha.
Stephen told me that the service had roughly doubled what it knows about in the last year and that they are finding that the more knowledge they add the easier it is to do so. This is because Wolfram|Alpha is built as a collection of customized knowledge domain and sub-domain specific modules on top of Wolfram Research’s Mathematica engine and they now have so many specialized modules that they have, in effect, templates that can be easily adapted making knowledge acquisition far simpler.
The way this architecture functions is by breaking your query into symbols (this is what Mathematica is exceptional at) that are then handed to the modules which each try to do something sensible with the data. The resulting response page is a collection of the output of all the modules that succeeded in responding.
Stephen pointed out that when you’re looking for knowledge, traditional search engines simply “forage” for answers without any way to determining the context of the question. Thus, “what is $100 in 1990 dollars” as interpreted by Google gets a page full of results that don’t actually answer the question while Wolfram|Alpha’s response is spot on:
Input interpretation: $ 100 (1990 US dollars) in 2010
Result: $ 165.55(2010 US dollars)
Average rate of inflation: 2.55% per year
Wolfram|Alpha also tries to be relevant in the user’s context. For example, if you enter “What is the temperature in Reading” the service will attempt to geolocate you and if you are in the U.S. the service then the answer will be for Reading, Pa., rather than Reading, England (although Wolfram|Alpha also offers you a link to the answer for the U.K. location in case that was actually what you meant).
Stephen sums up the goal of Wolfram|Alpha as being “to provide immediate access to knowledge” rather than what search engines provide, to wit, access to data. As he points out, if you want to know what angle the sun would be at your location on July 1 at 3 p.m. you could go and find the celestial mechanics equations and the ephemeris data and figure it out but when the algorithm has been built to do the equation and tested to make sure it’s right, why would you bother? In the same way that, for example, Google and Bing have democratized access to data, Wolfram|Alpha is democratizing access to computation.
My complaints from a year ago about a lack of a natural language interface have been largely addressed by improvement in the original knowledge modules as well as through the addition of new modules so my question, “what is the square root of minus 1” that Wolfram|Alpha surprisingly couldn’t answer is now correctly answered as is the rather over-brief query “population CA NY FL” (the even trickier query, “population CA + NY + FL” is also nailed).
The progress of Wolfram|Alpha is spectacular in such a short period. The service now boasts an API and an iPhone and iPad applications.
Eventually Wolfram|Alpha may become partially ad-supported but the big bucks are being made through a deal with Bing which makes the search engine way smarter. Other revenue options will include a Wolfram|Alpha enterprise appliance for organizations that want to ensure that their intellectual property is kept private and for fee integration with applications through the API service.
So, what could Wolfram|Alpha become? We’ll ponder this next week.
