The Search for Answers

Stephen Wolfram computes answers to never-before-asked questions. 

June 8, 2011 | “Wolfram|Alpha knows about lots of things,” said its creator, the British mathematical prodigy Stephen Wolfram, CEO of Wolfram Research and author of A New Kind of Science. Wolfram delivered the opening keynote to a packed house at the 2011 Bio-IT World Conference & Expo in Boston in April. 

Wolfram’s quest for the past 30 years or so has been to try to take knowledge and make it computable. As he told Bio•IT World, he judged that “it wasn’t totally crazy to build a systematic system that could take a large swath of the world’s knowledge and set it up so, if you could ask a specific question, you get an answer.” 

Launched a few years back, the Wolfram|Alpha “answer engine” is based on four principles: First, it gathers the underlying data. Foraging from the web doesn’t work, Wolfram said. Rather, the engine goes to primary sources. “We’ve developed a curation pipeline to get the data and make it computable.” The next step is to inject human domain experts. 10 million lines of code from Mathematica (Wolfram’s flagship software product) computes answers from the data. The third step is to use natural language processing to handle the queries. “We want humans to walk up to Wolfram|Alpha and ask questions,” he said. The final piece is computational aesthetics and data presentation.  

Ingesting the raw data—finding the definitive sources and making them computable—is in some ways the easy part. “95% of the work ends up [being] taking that raw data, setting it up so it’s all validated, organized, and correlated, so you can answer questions based on the data.” In the life sciences, the tool can return everything from the melting temperature of oligonucleotide sequences to matches with the human and other genomes. There is also a vast body of public health and medical test data.  

Wolfram said his answer engine is “possibly the most complex software ever assembled.” Every answer is computed dynamically. “We’re not just taking text, we’re trying to understand the question... we’re taking knowledge and data we’ve completely curated and trying to compute the answer to the question. We have to turn the human utterance, or image, or spreadsheet... and turn it into a precise symbolic form, completely understandable to the machine, and use algorithmic methods that have been specifically set up to compute based on the question. We can compute answers to questions never asked before.”  

“It’s a pivotal time for the computer industry right now,” Wolfram told us, and said he is interested in very large-scale systems modeling and the future of biomedicine. “With all the sensors we have available, we can get all these data on a particular person flowing into the system... Can we tweak this particular piece and now we can immediately compute the consequence of eating more whatever? Or in three years, you’ll have this issue that will cause you to take this drug?” 

Another intriguing idea is what Wolfram calls algorithmic drugs. “Right now, we make drugs with particular targets and purposes... Systems like molecules can do very non-trivial computations. When you have your drug going around [the body], can you have it not respond in a very standardized way to a target, but have it compute what to do on the fly, so to speak?” 

Wolfram has as good a shot as anyone of figuring that out. •