Watson, IBM’s artificially intelligent supercomputer, is known for beating the pants off Jeopardy contestants in trivia. But soon it might also be known for giving foodies on Top Chef a run for their money at cooking.
This weekend in Austin, IBM showed off Watson’s Cognitive Cooking program, a system that taps into the supercomputer’s analytics and machine learning tools to whip up novel recipes, depending on user tastes and food preferences. It’s a novel application of Watson, designed to draw more attention to the IBM brand. (The company revealed the service at SXSW with a food truck that had customers lining up–even in the rain.) But the program also demonstrates the types of consumer-facing applications IBM could cook up using Watson in the future. “Let’s say you want to try to make new perfumes, or maybe you want advice for how to dress better, or if you want to create personalized travel itineraries—these are all pretty much like recipes,” says Florian Pinel, senior software engineer in the Watson group.
The Cognitive Cooking system is like Beats Music for food: Enter a bunch of preferences, and Watson will tap into an ever-expanding database of recipes to return the best match for a dish. “We ask you for a few inputs and recommendations, and then we start the creative process,” Pinel says. “There will be trillions of combinations, and we can narrow them down to a few hundred dishes to look at.”
On a tablet, Pinel walks me through the experience. “Which cushiness sound good?” Watson asks, offering a list that includes everything from Iraqi to Cambodian cuisine. Next up it asks for a particular ingredient you want worked into the dish, then offers more specific adjustment options, like, say, whether you want the dish to be vegetable- or meat-based. As a test, we put together a garlic-based, Armenian-style gumbo. Watson spits out a list of recipes. “Four vegetables, three spices, one two kinds of fish or seafood, and so on,” Pinel ticks off. “I think that makes for a decent gumbo.”
We’d have to cook it up to find out just how tasty it is, but Watson also makes predictions, before you start, as to how surprising the dish might taste, how pleasant it will be, and how well the ingredients pair together. In this case, Watson predicts our tastebuds will find it very surprising.
But will it taste good? “We’ve seen a few recipes where I look at it and I’m highly skeptical that it’s going to work,” says James Briscione of the Institute of Culinary Education. “But then Florian will say, ‘No, let’s try it.’ And we do, and it’s like, ‘Wow, this is pretty darn good.’”