If you were to open your fridge one night and find a measly 10 items inside, you might decide to order Chinese. But the truth is that you could craft 3,628,800 dishes simply by mixing those ingredients in various combinations. Factor in different preparation styles and methods of cooking, and the options quickly reach the quintillions.
But first you’d have to invite Chef Watson, a new web app created by IBM and Bon Appétit magazine. It’s designed to reveal the creative possibilities lurking inside your fridge and cabinets, bringing what IBM is calling “com-putational creativity” to the kitchen. To make a dish with the free app–which launched in beta on bonappetit.com in June–you simply type in the ingredients you’d like to use, the kind of dish you’d like to eat (paella or quiche, say), and the style you’d like to try (Portuguese, French, or even something called Mad Men). Then Watson goes to work, using its cloud-based processing power to produce not one but 100 full step-by-step recipes for you to try.
Chef Watson isn’t just a clever program: It’s part of IBM’s broader initiative to monetize its supercomputer. Three years after Watson defeated Ken Jennings at Jeopardy, IBM Research is harnessing the power of its landmark technology not only to generate innovative new recipes but also to potentially assist doctors in diagnoses and help investors see market trends early. “Ultimately, we want Watson tech to avail itself to everyone,” says IBM Watson Group VP Stephen Gold.
Chef Watson works by juggling three databases. The first is devoted to what’s called hedonic psychophysics, which is essentially the flavors people like and dislike at the molecular level. The second is Bon Appétit‘s library of 9,000 recipes, complete with ingredients, measurements, and step-by-step preparations. The third includes “chemoinformatics“–which marries the first two databases, informing the system which molecular flavor compounds are in which foods. In other words, Chef Watson can create combinations of ingredients that it knows will taste good, even if nobody has ever thought to put them together before.
Refining Watson’s back-end culinary intelligence was just part of the challenge. Designers then had to make the huge amount of information feel both accessible and delightful to click on, which forced them to answer a deceptively simple question: Just how many recipes should Watson give someone when the possibilities approach infinity? At its South by Southwest debut last March (where IBM distributed Watson-created dishes from a food truck, along with bottles of the program’s Thai-inspired barbecue sauce), Chef Watson offered users a daunting 5,000 recipes in response to any simple query. “Thousands was clearly too many,” says lead designer Jacquelyn Martino. So they decided to scale back to 10. “But then people said, ‘This is Watson! Seriously, only 10?’ ”
So designers settled on 100, which the app lists in order from “classic” to “unique.” But prerelease beta users had a new complaint. Because Watson is “thinking” at such a complex level, humans can’t always detect meaningful differences between recipes that, say, use slightly varying amounts of baking soda. “In Watson’s mind, there’s a subtle difference there,” Martino says. “In my human mind, I’m sorry, I just don’t see it.” In response, IBM engineers tweaked Watson’s algorithms to present recipes that feel more distinct.
As IBM sees it, the Bon Appétit app–which is designed to show off Watson’s commercial potential rather than generate any revenue–is just the tip of the Baked Alaska. The computing system could be taught to incorporate nutritional information and generate recipes that maximize flavor but minimize sugar, fat, or any number of highly specific dietary restrictions. Or it could be loaded with intellectual property from the likes of Kraft or Nestlé to generate ideas for products. “Think about something like a cronut,” says Steve Abrams, a director of IBM’s Watson Group. “The cronut is a pretty straightforward combination of two ideas, but you can imagine companies looking at it and saying, ‘I should have done that! Can Watson help me?’ ”
IBM is banking on a lot of companies asking that same question. In the past year, it has pledged $1 billion to the Watson project, opening a glitzy Manhattan experience-design lab and placing the Watson team inside it. IBM will also be investing $100 million in startups and businesses that connect to the Watson Developers Cloud. Eventually, IBM says, Watson could be the backbone behind all manner of apps; it could help doctors diagnose illnesses by digging through info on diseases, symptoms, and patient histories, or help financial analysts sift through global research and market data to see trends that the human mind could never spot–and even make predictions.
Of course, there’s a big difference between financial prognostication and bananas flambé. Still, Watson’s artificial intelligence could have a real impact on the culinary world. When Jared Wentworth, the chef-owner at Chicago hot spots Dusek’s and Longman & Eagle, was asked about Chef Watson, he thought it sounded “very cool.” Wentworth, a perennial earner of Michelin stars, says he looks up recipes all the time, often cross-referencing 10 or more before walking away with a workable basis for a dish. Watson’s ability to automate that task is appealing. And for home chefs, he says, the potential is even greater, because Watson can suggest flavor pairings that most of us would scoff at. “When I first started, you’d have to go into a restaurant and work for an extended period of time to gain knowledge,” he says. “Now it’s at your fingertip.”