How Wolfram|Alpha Works

Wolfram|Alpha launched on May 15, 2009. Many early users expecting it to work like a search engine complained that the site wouldn't return results to some simple queries. But unlike Google -- and other search sites -- Alpha isn't an attempt to catalog the world's information. Instead, it's what Wolfram Research calls a computational knowledge engine.

Traditional search engines crawl the Web for sites and add them to their directories. Algorithms built into the code that powers these search engines rank pages higher or lower based on several factors. One is the number of people who click on the link from the search engine results page (SERP). Another is the number of third-party Web sites that link to that site. If you create a public Web site, there's a decent chance it'll end up on a search engine without your having to do a thing to get it listed - as long as there are links to it from other sites

Wolfram|Alpha doesn't scan the Web for material. Instead, it relies on licensed databases and content entered, tagged and cataloged by Wolfram Research employees. According to the company, there were more than 10 trillion individual chunks of data on the Alpha servers at the time of the site's launch [source: Wolfram Research]. Wolfram Research employees vet all information to ensure its accuracy before it's added to the Wolfram|Alpha databases.

To get that information, a person types a query into the search box on the Wolfram|Alpha home page and presses the equals sign to the right -- the equals sign is a clue that there's math going on in the background. Alpha then looks searches for corresponding data in its databases.

Results can be quite simple: Enter the name of a city, and you'll get the time of day and weather information, as well as population and elevation above sea level. Input several city names and you'll get a table that enables you to compare them. If you type in the name of an animal, you'll get the average size, alternate names, scientific name and even a breakdown of its taxonomy. Again, if you ask Alpha for the names of different animals, you'll get a table comparing them, and you'll even see how they're related taxonomically.

Scientific and mathematical problems are easy for Alpha to handle. In fact, you can ask the computational knowledge engine to derive a calculus problem and not only will it tell you the answer, it will also show you what steps are required to solve it.

But remember, Wolfram|Alpha isn't a search engine, which means that if you want to find pictures of celebrities, the cheapest price on a flight to London or articles written about a specific topic, you need to use a search engine. Alpha is designed for you to ask it a question and get a factually correct answer, along with a visual interpretation of that answer.

When Stephen Wolfram was just 15, he had his first scientific paper published. He earned his PhD in theoretical physics from the California Institute of Technology when he was 20. Wolfram received a MacArthur Prize Fellowship in 1981, and that's right about the time he started looking into the systems of nature and how complex they are.

Wolfram began using computers in 1973. In 1986 his company, Wolfram Research, released the first version of its well-respected Mathematica software. Mathematica is used in all sorts of industries, including engineering, science and finance, and is capable of high-end computation and modeling. The algorithms from Mathematica are used to calculate and display search results in Wolfram|Alpha.

In 2002, Stephen Wolfram published a book called "A New Kind of Science," in which he explains his belief that simple rules can explain complex problems. As Wolfram explained in his March 2009 blog post, the principles in his book and the computational power provided by the Mathematica software are what led him to believe that it was possible to create Alpha.

One of Wolfram's goals was to make it possible "to ask a computer any factual question, and have it compute the answer" [source: Wolfram]. But natural language processing isn't easy. According to Wolfram Research, the Web site uses algorithms and analyzing software to identify patterns in data. This helps it identify shorthand terms that people use to ask questions. This helps it guess what you're trying to find when you enter a query in its search box.

As an example, if you type in "GEC," Alpha assumes you're asking about the General Electric Corporation. If you separate the letters with commas, you get musical notes and a visualization of where they are on the keyboard. Click the "Play Notes" link to hear the progression -- and perhaps if you're an American, you can guess which major television network GE is the dominant corporate parent of. Wolfram|Alpha is guessing, by the way in which you enter your search term, what information you're trying to find.

Wolfram Research says Alpha is based on four pillars: data, dynamic computation, natural language understanding and computational aesthetics. The data are handled by Wolfram Research employees. The rest is done on the back end via Mathematica.

To crunch those kinds of numbers, you're going to need serious computer power. Wolfram Research has a supercomputer (really, two of them) built using Dell hardware customized by a company called R Systems. The machine, named R Smarr, is the world's 66th-fastest supercomputer (as of this writing). With 4,608 processor cores, R Smarr can perform 39.6 trillion operations per second. There are 576 quad-core Intel "Harpertown" Xeon chips inside and each R Smarr supercomputer has 65,536 GB of random-access memory, or RAM [source: Shankland].

If so many people want to pit search engine giant Google against the newcomer, why don't we take a look at the two of them and see how they stack up?

According to Wolfram|Alpha (which used Alexa.com as a source), at the time of this writing, google.com gets 3.8 billion page views per day. Alpha, on the other hand, gets 13 million page views. Google's No. 1 on the Web, compared to Alpha's 568th.

But that doesn't tell the whole story. Google makes its money from advertising. In 2008 alone, Google earned revenues of $21 billion -- 99 percent of that was advertising-related [source: Buley]. Wolfram Research is also using advertising to subsidize its site, using a contextual "featured sponsor" box that appears to the right of some search results.

Wolfram|Alpha may need to find other ways to generate income. There are around 250 people working behind the scenes on the site [source: Shankland and Needleman]. Curating data by hand is costlier than sending spiders out to catalog Web sites. It's also more time-consuming. The database Alpha used at launch took three to four years to build [source: Beaumont].

It's possible, too, that Wolfram Research may sell customized subscriptions to organizations that want to compare proprietary information with the Alpha database. There's also an application programming interface, or API, that will allow developers to blend the functionality of their sites with Wolfram|Alpha [source: Shankland and Needleman].

But Google co-founder and President of Technology Sergey Brin got a demonstration of the Wolfram|Alpha technology by Stephen Wolfram himself prior to the site's launch [source: Levy]. This suggests the two might find a way to partner. On the other hand, some think Google Squared, a search feature that will enable Google users to compare data in ways similar to those used by Alpha, shows a desire to crush the upstart. However, Google Squared evaluates data found online, not the exclusive proprietary databases, as Alpha does [source: Talbot].

As Stephen Wolfram said in his March 2009 blog post, Alpha is a massive project that will never be finished. Time will tell if it manages to carve out a niche in Internet users' hearts or whether its technology will end up underpinning search efforts on other sites.

For more on Internet search technologies and related topics, take a look at the next page.