Wolfram Alpha acts as ‘computational knowledge engine,’ not search engine

Peter Cohen, Macworld
20 May, 2009
View more articles fromthe author

Wolfram Alpha LLC has announced the general availability of Wolfram|Alpha, a “computational knowledge engine” available on the Web that shares code with Mathematica, Wolfram Research’s renowned technical computing software.

Alpha lets you input a query that requires data analysis or computation, and it delivers the results for you. It’s linked to terabytes of “curated” data — data that’s been hand-selected by experts working with Wolfram, who go through steps to make sure the raw data is tagged semantically and is presented unambiguously and precisely enough that it can be used for accurate computation.

Examples of how Alpha can current be used range from solving difficult math equations to doing genetic analysis, examining the historic earnings of public companies, comparing the gross domestic products of different countries, even measuring the caloric content of a meal you plan to make. You can find out what day of the week it was on your birthday, or show the average temperature in your area going back days, months or years.

Not a search engine. The first thing Wolfram Research co-founder Theodore Grey wants you to know is what Alpha is not: It is no “Google killer,” as it’s been called by some reports. In fact, Alpha is very, very different from a search engine.

“Search engines are like reference librarians,” Grey explained. “Reference librarians are good at finding the book you might need, but they’re useless at interpreting the information for you.”

Google’s strength, Grey said, is in providing you with links to pre-existing data that you may be interested in — news reports, online database entries, and other content that references the phrase or words you’re interested in finding.

Alpha, meanwhile, takes reams of raw information and performs computations using those data. It produces pages of new information that have never existed on the Internet.

“Search engines can’t find an answer for you that a Web page doesn’t have,” Grey explained.

“It’s been a dream of many people for a long time to have a computer that can answer questions,” said Grey. “A lot of people may think of a search engine as that, but if you think about it, what search engines do is an extreme limited subset of that sort of thing.”

Mathematica for the masses. Alpha is completely written in Mathematica, except for the Web-specific server-side code. It’s more than just a publicly accessible version of Mathematica, however. It employs an advanced heuristics engine that tries to make sense of your query without forcing you to understand how to make equations using Mathematica’s programming language-like syntax. Alpha allows you to input queries in natural language, and it does its best to figure out what you’re asking and provide you with the data you’re looking for.

That was one of the goals behind Wolfram’s recent private beta test and last weekend’s final shakedown — to make sure that Alpha could interpret many of the queries that it was getting. Grey says that going forward, Wolfram will continuously improve the Alpha engine to better understand what people are asking it. Wolfram plans to add additional data to search, as well. “Many people over the weekend went looking for sports statistics, for example,” he said — an area that Wolfram hasn’t gotten to yet, but will.

Making data work. To that end, Wolfram works with many contributors — Mathematica users and experts in their fields — who identify usable, publicly available data sets and work to make it digestible to Alpha.

“In order to make data computable, the tricky part is that there’s a lot of information — lots of collections of data where you can get a raw dump. But if you want to make a graph of quantities or compare data from another source, there are nagging problems, like the assumptions upon which these data were based,” said Grey.

So steps are required to take the raw data, transform it and make it ready to be transformed and compared with other sources. That process isn’t totally automated, but Grey says that Wolfram has found ways to “semi-automate” it. “It’s incredibly difficult and it’s a vast amount of work, but it’s not impossible or insane.”

Grey says the best way to understand what Alpha does is to use it, and to compare the results by inputting the same query in your favorite search engine and see what happens. The differences, Grey thinks, will be instantly apparent.

And if you’d like to see how Alpha works in the hands of an expert, there’s a screencast narrated by Wolfram Research founder Stephen Wolfram you might find informative, too.

Leave a Comment

Please keep your comments friendly on the topic.

Contact us