When Computers Cook: IBM’s Chef Watson Teaches You a Thing or Two About Flavor

Chef Watson’s Portuguese Lobster roll. (Photo: SourceBooks)

When first encountered with the idea of computer-generated recipes, I assumed the result would be home cook-friendly. If the recipes were the result of science, and not art, then surely the average cook would be able to follow the recipe.

“Cognitive Cooking with Chef Watson,” by IBM and the Institute of Culinary Education, proves my assumptions wrong.

The book contains more than 65 original recipes that were created partially by a cognitive computer named Watson, which IBM created in 2011 to answer questions on Jeopardy, and partially by chefs at ICE, an award-winning immersive culinary school in New York. And of those 65 dishes, the vast majority of them will be daunting to the average home cook. With the exception of a few cocktails, none of them are simple enough for weeknight meals.

First, how does Watson work? Here is very simplified version: Data scientists at IBM loaded it up with thousands of recipes and a database of food chemical compositions. They then “taught” Watson known food pairings, like rosemary and potatoes, so it could learn how the chemicals of foods interact, to create those tasty pairings. Then they gave him a list of known favorites like roasted chicken to keep him from reinventing the wheel.

Watson then created “recipes” – or rather, groups of ingredients that he felt should be paired together in a dish. Then professional chefs from ICE created actual recipes, including ingredient amounts and cooking instructions, to create a dish.

Think of it as “Chopped,” plus more ingredients, minus the ticking clock. IBM calls it “cognitive cooking.”

Each recipe in the book is rated by three metrics. “Surprise” measures how rare of the flavor combination is. Recipes are also measured by “pleasantness,” which is based on flavors known to give people pleasure on a molecular level (also called “hedonic psychophysics.” And finally, recipes are rated on their “synergy,” which factors in how common compounds known to taste better together are used.

What was the result? A dizzying collection of totally off the wall concepts, such as the Austrian Chocolate Burrito, which has a beef filling flavored with dark chocolate and ground cinnamon, a cocoa-apricot purée, and mashed edamame (100 percent surprise, 75 percent pleasant, and 45 percent synergy).

Or there’s the Vietnamese Apple Kebab. Don’t be fooled by its simple name – it’s not grilled apples on a stick. The recipe calls for pork meatballs, which contain apples, Vietnamese curry power, and vanilla bean. That’s paired with a pressure-infused curry chicken. Combine those with a pineapple broth, and finally, there are pickled carrots and shitake mushrooms. This one is 65 percent surprise, 60 percent pleasant, and 75 percent synergy.

So what’s the point of the book? It could be seen as a fun challenge to a professional cook. At the very least, it’s some good entertaining, and educational, reading for anyone who is curious about the intersection of food and science.

Here are some of the more interesting recipes from Chef Watson. If you’re feeling bold, try whipping one up! Otherwise, enjoy the sheer genius of what it’s created.

Vietnamese Apple Kabob

Portuguese Lobster Roll

Belgian Bacon Pudding

Curious about how ICE and IBM partners? Here’s more info: