The newly released Wolfram Language claims to know everything about everything. The all-knowing language is the driving force behind computational knowledge engine Wolfram|Alpha. But it hasn’t gotten closer to obtaining perfect knowledge without help from real, human experts.
Nearly every one of Wolfram|Alpha’s algorithms and real-world object identifiers (the ones that make it possible for the language to understand what a “pint” or a “foot” is) have been vetted by people who truly do know everything. And the Wolfram Alpha team regularly consults with real experts on how to make sense of all the datasets in its knowledge stores.
Sometimes, that human intelligence comes right from the service’s community. Not long ago, a music composer reviewed some Wolfram|Alpha apps, and his feedback ended up catching the team’s eye. As it turned out, this composer’s heavy background in mathematics gave him unique insight into the type of musical smarts that Wolfram|Alpha was looking for.
“We ended up hiring him to enhance music theory content in Wolfram|Alpha, and he continued on to do some audio-related development in the Wolfram Language,” recalls Alan Joyce, director of content development at Wolfram Alpha.
Wolfram|Alpha draws upon its gathered knowledge to retrieve and compute a response rather than spitting out a list of related links that might contain the information you’re looking for, à la Google.
Say, for instance, that you wanted to know about love. You could enter “What is love?” or simply say “love,” and it would give you virtually the same result, replete with the usual suspects. You would see a histogram of word frequency usage and some definitions and synonyms.
But you would also get a list of a few phrases that might help you understand the context of the word, other words that rhyme with it, and crossword puzzle clues that could give you a leg up on Friday’s version from the New York Times. Wolfram needs experts to create results like these.
“In virtually all cases we do work closely with the original sources or outside experts in that domain to understand the key questions people ask in that domain, how to structure our results and so on,” says Joyce.
For the most part, Wolfram|Alpha operates from datasets that Wolfram gets from large repositories. But Wolfram usually does the asking and not the other way around. Joyce points out that the team does not normally take data from small-scale researchers or hobbyists.
“That said, we do regularly hear from researchers and other people who have interesting smaller datasets or algorithms that they would like to see added into Wolfram|Alpha. Sometimes those are simple additions to domains we already cover, like specialized data about countries or people. Sometimes they’re more complex projects,” says Joyce.
While hiring contributors doesn’t happen all too often, Wolfram Alpha takes in interns every summer to improve the tool through focused projects.
Last summer, an intern worked up a bunch of scenarios for Wolfram|Alpha that would have helped you win that game in the third grade where you had to guess how many Jolly Ranchers there were in the class candy jar. Called magnitude-of-estimation problems, he added the knowledge required for the engine to calculate, “How many pencils fit into a Boeing 747?” or “How many apples could you fit on a football field?”
But interns at Wolfram don’t necessarily get tedious jobs. “Sometimes the projects are highly technical. Sometimes they’ve got more of a fun, pop-cultural focus,” says Joyce.
Another intern last year got to work on the “Fun Curves” project where she took famous cartoon character sketches and turned them into mathematical formulae. Now, if you type “Thor curve” into Wolfram|Alpha, you’ll get a cool Thor rendering, plotted on a Cartesian coordinate system.
Despite how rarely unsolicited contributors leave their mark on Wolfram|Alpha, Wolfram always keeps its doors open to people who want to contribute. “We always encourage interested people to contact us if they have the skills and knowledge to add some useful new functionality to Wolfram|Alpha,” says Joyce.
Each time the Wolfram team works with experts (and a well-prepared intern) on better ways to understand how we search for knowledge, Wolfram|Alpha gets a little better at sifting through data and queries. All the while, the engine and its mature Wolfram Language take cues from their developers while they all learn to find coherence in it all.