Watson cooks up computational creativity
Florian Pinel talks about food and future uses for cognitive computing
Watson works on a lot more than facts. Now, the cognitive computing system is sifting through ingredients, helping to find new recipes for food, business and other applications.
It all began about two years ago, says Florian Pinel, lead software engineer on the computational creativity project, in IBM’s Watson Group. “We wanted to see if computers could also be creative, instead of just reasoning about the world as it exists, as Watson did during Jeopardy! Could it help people be more creative themselves?”
Forget about Watson painting the next Mona Lisa or composing a new sonata. Computer creativity has its limits. That’s why the team looked for a field that had a “scientifically creative” focus, rather than pure creativity like music or art.
“With data analytics, a computer can examine various creations and, in effect, measure them against standards. But if you just have a computer create random things, it has no way to evaluate them. You can’t really say that it’s fully creative,” says Pinel. “With scientific creativity, on the other hand, we have a lot of data available that allows us to evaluate the creation.”
Annual U.S. restaurant industry sales--$683.4 billion
Restaurant locations in the U.S.--990,000
Restaurant industry employees--3.5 million
Restaurant workforce as part of overall U.S. workforce--0%
Restaurant industry share of the food dollar--47%
1965 U.S. per capita beef consumption - 74.6 lbs
2014 per capita beef consumption - 53.6 lbs
1965 per capita chicken consumption - 33.7 lbs
2014 per capita chicken consumption - 85 lbs
Starting with cooking
“Cooking is one domain of scientific creativity. You would think it is more strictly creative, but it’s not.” Pinel should know; in addition to an M.S. in Computer Science and Engineering from Ecole Centrale de Paris, he has a Culinary Arts diploma from the Institute of Culinary Education (ICE) in New York City.
The first step was to educate Watson about cooking. The team looked at cuisine pages on Wikipedia as well as some 30,000 online recipes, nutrition facts from the U.S. Department of Agriculture, information about food at the molecular level, even details about what flavor compounds are in what ingredients. All that went into a database that essentially became Watson’s memory about cooking. The team also collaborated with ICE, Pinel’s alma mater.
Once the data was in place, the team could choose a few parameters -- ingredients, cuisine (Italian, Asian, etc.) and a dish (taco, croissant, etc.) -- to generate billions of trillions of combinations. Watson would then evaluate them based on novelty and predicted quality. After that winnowing, the team worked with chefs at ICE to test 20 or 30 recipes, Pinel says. Some of those were served recently from an IBM food truck at the South by Southwest (SXSW) festival in Austin earlier this year.
Looking at future applications
Pinel admits that inventing new recipes is unlikely to become an important business for IBM. But he does believe in a future for apps that incorporate the feature. “We’ve been working on dietary restrictions. Some people, either because of personal beliefs or medical conditions, can’t eat certain foods. That often limits the variety of their dishes. With Watson we can create similar recipes that taste the same. For example, we could generate a ‘shrimp’ recipe that tastes like shrimp, but doesn’t have same level of cholesterol as the actual seafood.”
But Watson’s recipes will likely resonate far beyond food. “The algorithm that we’ve developed can help you create new products that are made with smaller pieces. You could think of making new fragrances, perfume, for example,” says Pinel. “In fact, if you’re any kind of manufacturing company that uses smaller items to make a bigger product, you could use this approach.”
Generating new business processes
It also has potential for generating new business processes. “Recipe instructions are very much like a business process, when you think of it,” he says. Imagine a company that needs a new process and has a number of business objects to work with. “Then by pressing a button it could generate new business processes to fulfill its objective.”
Take insurance claims, for example. A claim involves multiple steps–filing, assessment, reimbursement. Each is a business object. The Watson system would look at existing processes and objects and try to combine the latter into new processes. It would also be able to evaluate the results based on such criteria as cost, speed and efficiency.
Financial portfolios and more
A similar approach could create unique financial portfolios, Pinel says. “Think of a portfolio as a recipe. Watson could generate a financial portfolio with a mix of assets that no one else has. It could also evaluate the result against the individual’s tolerance for risk, need for growth and other goals.”
Likewise, Watson could offer highly personalized itineraries, based on a traveler’s personal interests, available time and budget. In fact, he says, any industry that mixes ingredients, chemicals and components could benefit from Watson. The one exception is pharmaceuticals, since the results of new drug combinations can only be evaluated in clinical trials.
What the average American eats in a year (in pounds)
Red meat 105
Fish and shellfish 16
Dairy products 607
Four and cereal products 194
Caloric sweeteners 130