Last week, fans of the game show Jeopardy were treated to a three-day competition between IBM’s latest supercomputer – Watson – and two of the show’s best known champions, Ken Jennings and Brad Rutter. I found myself conflicted as I watched the show. I started rooting for the super computer as it’s truly a marvel of human engineering and shows remarkable progress in the field of artificial intelligence, but then I was disappointed when Watson would take over and not let the super-smart humans even chime in with a response… After all, if the computer wins, does it mean that computers are now smarter than humans?
As it turns out, even though Watson won the show (and hopefully that wasn’t a spoiler), it still isn’t smarter than some of the smartest humans. Watson was designed to perform well within the particular scenario of a Jeopardy game where, within 3 seconds, it could process its database of various topics to come to a reasonable response (actually, several responses ranked by statistical likelihood of being correct) and click on the response buzzer. The fast processing also allowed Watson to determine “contextual” relevance of the responses and, in some sense, to understand the nuances of natural human language. But for many questions, the speed to click on the buzzer was really all that Watson had as an advantage as the human competitors also knew the answers. In several instances, Watson was stumped and, in fact, at the end of the first game, it totally botched up the Final Jeopardy response by guessing a Canadian city for the category U.S. cities.
Listening to a post-game analysis of the show on NPR’s “To the Point”, I learned about something called a Turing test. This is a test of a machine’s ability to demonstrate human intelligence. It was conceived by the British mathematician Alan Turing in 1950 (yes, more than half a century ago). During this five minute test, a human judge is tasked with posing questions to two participants (one is a human, the other a computer) and needs to determine by the end of the test which of the two respondents is the computer. Over the years, this test has been run countless times and though computers are getting smarter, they are still not able to imitate human conversation to such a level of sophistication to fool the human judge. The IBM scientist who worked on Watson conceded that Watson would not be able to pass the Turing test.
So is Watson only good for game show playing? Shortly after the Jeopardy publicity stunt, IBM announced some of its first agreements to deploy Watson in the real-world – one of them is the medical sphere where a Watson-like computer can act as a physician’s assistant. For example, a patient with a specific medical issue comes in to see the doctor. While in the examining room, the doctor can pose questions to the patient to determine a possible diagnosis. Meanwhile, Watson would be listening in, and providing additional inputs for the doctor. Watson can easily pull up and process the patient’s medical records, as well as the latest medical and pharmaceutical research that the doctor might not have had time to read, and present those insights to the doctor for additional consideration as he forms his final diagnosis. Sounds like a very useful application of supercomputing power!
Of course, I then began wondering if the world of marketing analytics would one day benefit from computers like Watson. Certainly, much of our industry already relies on mass computing power to capture clickstream data, process it, and render it in pretty charts and graphs. What, of course, is missing and where the human element comes in is to determine what the data means and if a change to a specific banner ad, an update in SEO keywords, or a link on Twitter caused an improvement in traffic and conversion rates. We spend a lot of time testing these changes and inputs, but sometimes we don’t have a complete view into all of the inputs… so we re-test hypotheses, tweak websites, and see what the data tells us after all those tweaks. But wouldn’t it be really cool to use faster processing power of a supercomputer to look at multiple data streams quickly and help us reach insights sooner to dazzle our clients? Perhaps, Watson and super smart computers aren’t all that bad.