So far, the program has generated dishes like Swiss-Thai asparagus quiche and Austrian chocolate burrito. IBM served both out of a food truck that it debuted at a Las Vegas tech conference last week. They may sound like strange flavor combinations, but human taste testers have deemed them delicious. The truck's next stop is Austin, where it will be serving up more of these unorthodox dishes at the South By Southwest music festival.
"The goal is to help chefs figure out combinations they would not have thought about," says Florian Pinel, one of the IBM researchers behind the technology.
Chefs usually think about pairs of ingredients when pondering new tastes and combinations, Pinel tells The Salt. Occasionally, they'll think about three flavors that might work well together.
By contrast, "the computer can go through trillions and quadrillions of possibilities," Pinel says.
The chef starts by suggesting a main ingredient — say, lobster. The program then goes through its huge database of recipes and ingredient profiles, looking for other ingredients known to pair well with it in different global cuisines. The program evaluates the chemistry of the food and models human perception to try to predict which ingredient pairings will prove tasty and surprising. (One dish it came up with: a Cuban-style lobster bouillabaisse with squash.) This video explains how it all works in more detail:
The current prototype, which the chefs at ICE are testing out, produces a simple list of ingredients and a suggestion about how to cook them. For example, it might suggest making a pie filled with pork tenderloin, apples, garlic and allspice. This sort of system works well for chefs, who can figure out how to proceed from there, Pinel says.
And IBM is also working on an app for the general public, which will provide more detailed recipes for culinary novices, Pinel says.
For chefs, the program is a great way to generate new, unexpected dishes, says James Briscione, the director of culinary development at ICE and one of the chefs working on IBM's food truck.
"It has driven up some flavor pairings that we would not have thought of," he tells The Salt. "But we haven't come across an instance yet where something didn't taste good."
Take the pork belly moussaka that Watson's program thought up. "The ingredient list sounds horrendous," Briscione says. It involved cottage cheese, red bell peppers, dill, pork belly and cheddar cheese.
"I said, 'No freaking way! This sounds horrible,' " Briscione says. But in the end, the moussaka turned out quite mouthwatering. "Everyone who tasted it had to say, 'Wow.' "
Still, computers won't be replacing chefs any time soon. Briscione says that while computers are great at coming up with flavor combinations, they still have trouble with things like balance and texture.
"You can end up with a list of ingredients, and none of them has an element of crispiness," he says. That's where chefs have to use their expertise. And it's where human creativity comes into play.
"Two different chefs can get the same list of ingredients and come up with completely different dishes," Briscione says.
Curious foodies can taste IBM's offerings at SXSW Interactive from Friday, March 7, through Tuesday, March 11. You can also vote online for the types of dishes you'd like the food truck to serve.
Can't make it to Austin? You may be out of luck: IBM has no firm plans for the food truck's next stop. But if you're itching to try one of Watson's creations, you could give one of these computer-generated recipes, with instructions from ICE chefs, a whirl.
Copyright 2014 NPR.