The Answer Man
In the early 1980s, after earning a doctorate in physics at age 20 and winning a MacArthur Foundation “genius” grant at 21, Stephen Wolfram started thinking bigger. He wanted to build an artificial intelligence that would act like the world’s wisest statistician, capable of understanding, researching, and solving any numbers-related problem. So he surveyed the state of technology to see whether this might be possible. “I decided, no,” he recalls. “We were not there yet.”
He continued working at several top-flight universities until he grew tired of academic politics and quit to create Mathematica, now the de facto software for scientists and engineers who need to solve complex equations. Around 1990, he checked again on that big idea. Still no. So for much of the next decade, he locked himself in a soundproof room and wrote A New Kind of Science, his best-selling, 1,280-page explanation of why our way of thinking about science is all wrong. Finally, around 2003, he judged that the advent of the Internet and the boom in processing power had at last made his answer engine possible.
Luckily, the time was right for Wolfram too. The enormous success of Mathematica, which now has more than a million users, meant that he had enough financial resources, and the years of developing the software gave his team the expertise in data-crunching needed for a project of such grand ambition. The result, WolframAlpha, debuted in May. “It’s really about getting specific answers to specific questions,” Wolfram says. Type in “highest average income country” or “yen exchange rate” or “atomic weight plutonium.” The program will return fact-based answers to these kinds of queries in a flash.
Wolfram has assembled more than 200 researchers to work on the project. They include specialists charged with curating data from the United Nations, the CIA and published scientific research, and programmers who write the six-million-plus lines of code needed to interpret queries and produce the best possible answers.
For example, say you enter “ISS from LA.” First, Alpha attempts to interpret what you want. Based on probability and context, it assumes that you’re interested in the International Space Station (if you typed in “ISS to JFK,” it would assume you meant ISS as the airport code for Wiscasset, Maine, and give you the distance from there to JFK airport). “LA” is trickier. Are you looking for the position of the ISS relative to Los Angeles or Louisiana? If you’re in California, it will see that in your computer’s geo-IP address, and assume Los Angeles. If you’re in New York, it will come to the same conclusion a different way, Louisiana being less popular than the City of Angels.
Next, Alpha will grab data from NORAD on the station’s last position, which is updated roughly every six hours. That’s not exact enough for Wolfram, though, so the engine takes that historical information and projects forward, solving a differential equation to tell you where the ISS is now and, if it’s not overhead, when it will be.
Wolfram says the current version is merely the first step in a larger effort. It draws from more than 10 trillion pieces of curated data from trustworthy sources. But he wants it to be able to draw on and compute all of the knowledge—facts, ideas, equations, algorithms—humans have compiled about the universe. Such lofty aims are “typical Stephen,” according to people who know him. “If Stephen hadn’t done it,” says computer scientist Gregory Chaitin, “no one else would have imagined that it was possible.”
Wolfram, 50, is known for being different, even among the eccentric world of software moguls. He considers himself a people person even though he runs his company remotely. (He insists that he just finds face-to-face meetings inefficient.) He looks you in the eye when listening but also has a habit of rapidly nodding, as if to say, “Yes, I’ve got it, move on.”
Although he won’t reveal his net worth, he happily spends millions on basic science research with no hope of turning a profit (a favorite field is cellular automata, a computer-based way of studying complexity). “I have one very business-oriented child who accuses me of being a mushy intellectual and not adequately businesslike,” he laughs.
But Wolfram has, in effect, been in the business of selling answers since he was a 14-year-old prodigy solving his fellow students’ physics problems for a fee. Alpha is essentially an automated version of that enterprising young geek.
At the same time, the resources of Wolfram Research, both financial and intellectual, were critical to making Alpha happen. “You have to have the vision to do it,” says Carnegie Mellon University computer scientist Klaus Sutner, “but you also have to have the means.” Alpha represents an investment of many years and millions of dollars—a project of its complexity could not have happened in academia or at a totally profit-driven public company, Sutner says.
Wolfram is convinced that this money isn’t lost. He believes Alpha will eventually pay for itself. It already shows ads alongside answers, and he says there are “a whole fleet” of additional deals in the works.
“My utopian point of view is that if you build something sufficiently useful, there will be a way to make money from it,” he says. “I just like to do projects where I feel like if I hadn’t done the project, the project wouldn’t have gotten done.”
—Gregory Mone
