For years Google’s search engine has ruled the Web story-finding world, but its interface is far from perfect: Even with the new voice-recognition front-end, it’s not like asking an expert to answer your question. Enter IBM with “Watson.”
The New York Times has a fascinating piece about this supercomputer today, and though it’s craftily built around a technology demonstration of Watson where the machine played a fake game of Jeopardy against human competitors and actually won, the intention of the research is absolutely clear: In pretty short order, tech like Watson will revolutionize Internet searching.
IBM’s expertise in supercomputers is nothing short of legendary, so maybe it’s not a surprise that Watson could defeat mere human health teacher and copy editor contestants in a quiz game. But the way Watson works is very interesting. Google‘s algorithms scan your search query and build a digital description of it: Google’s Web archive is then scanned, and how well matches “score” (in a complex piece of math) determines how high up the search list they appear. It’s clever, but mechanical, and every single Google user is aware that sometimes it takes a while to find something specific on Google, requiring a massaging of the search terms and careful scrutiny of the search results. But Watson is jam-packed with linguistic algorithms, with the goal of turning it into a “question answering machine.” It’s smart enough to listen to a query phrased in natural English, infer what you’re searching for, and then deliver a precise result.
Watson doesn’t use the Net as a database, and instead has millions of documents in its archive from which to draw its answers–that’s why it’s perfect for a TV quiz show-type task–but unlike previous efforts for intelligent question-answering (like the Wolfram Alpha technology) it doesn’t require humans to process, categorize, and link the data together. This is a huge stride in this sort of artificial intelligence, as if the machine did require this sort of filtering, and you tried to scale its database up to real-world sizes, you’d face an infinite-time task. When the machine is posed a question, it uses its supercomputer speed to quickly analyze the text using hundreds of different algorithms, and then it maps its answers from the database to best match to what the different algorithms think the question is.
The machine is going to get a share of real limelight as soon as this fall, as Jeopardy producers have been impressed by its skills enough to set up a special quiz with show veteran contestants. This is a nice piece of PR for IBM, of course, but there’s a much bigger implication. Watson is several years old, and supercomputing tech has evolved rapidly even in that short time–you can have a supercomputer on your desk now, and maybe soon in your laptop and then smartphone. When Web search engines embrace this sort of technology, searching for information about anything is going to be as easy as it appears in sci-fi movies. Though it’ll take a while for even this sort of super-smart AI to answer questions like “Who will win the World Cup this year?”