The week of March 8, 2015

A dinner date with a supercomputer

By Allen Weiner

Being a vegan may be healthy, but it sure can get boring. You can only eat beans and rice with a side of veggies so many times before you go screaming out the front door in the direction of the local burger joint. Enter Chef Watson.

IBM is using its supercomputer to ignite new ideas and inspire both professional chefs and DIY home cooks to push the boundaries of their current culinary capabilities. Chef Watson is an applications of IBM’s natural language processing capabilities in which a huge corpus of ingredients, recipes, and cooking information is examined, studied, parsed, and easily reassembled to afford what it calls “cognitive computing.”

Chef Watson studies recipes and breaks the ingredients down into individual components, understanding their chemical composition and determining what ingredients will work well together to afford a pleasant dining experience. Where the discovery and delight comes in is when Watson suggests pairing things together one would never figure would make sense. I mean, what food purist could imagine having lima beans and blood oranges together in a main dish? (More on that later.)

IBM, like so many other technology firms, often uses well-understood, accessible contextual applications to showcase its new products and their more abstract capabilities. By using the subject universe of food and cooking to demonstrate the power of Watson’s linguistic muscle, Steve Abrams, director of IBM Watson Life, told the Kernel that customers in other areas such as pharmaceuticals and financial investments can understand how the system’s ability to discover and analyze data can work in their disciplines.

IBM is using its Watson supercomputer to ignite new ideas and inspire both professional chefs and DIY home cooks.

 “Could we build a system that goes beyond [Watson’s ability to answer basic] questions to allow people to become more creative?” Abrams said of IBM’s product plan to extend the supercomputer’s capabilities. IBM’s mission was to find an area of content to showcase Watson’s ability to ingest complex information and produce easy-to-understand results. The topic of food and cooking, which married elements of science and recipe development, met that criterion. “Food was perfect because food is not just about subsistence; it’s about the cultural and social part of our fabric,” added Abrams.

In explaining the principles behind Chef Watson, Abrams broke the technology down to its fundamentals: “What Watson does is look at what ingredients are commonly used together and how they influence a specific style. For example, French cuisine implies a certain set of ingredients. It also looks at what makes up a certain dish—that is, create a mental image of what is a pizza or burrito. It then correlates ingredients and culinary styles.” 

To further illustrate its capabilities, IBM has two versions of Chef Watson in the field. One system is geared toward professional chefs and works with the New York-based Institute of Culinary Education (ICE). Its goal is to surprise even the most advanced food artist by suggesting unlikely pairings. A second version geared for home cooks was developed in conjunction with Bon Appétit.

(Photo Credit: Jon Simon/Feature Photo Service for IBM)

Engineer Steve Abrams (left) and Bon Appétit editor Dawn Perry welcome Chef Watson.

“Watson takes inputs from the chef, who specifies a few ingredients as well as those he wants to avoid,” Abrams explained. “He then presents a culinary style and dish type and is presented with thousands of possible options that are scored based on which ones will be most pleasant to the human palate.”

With the foundational elements of a recipe in place, it’s then up to the chef to assemble the pieces to create what might turn out to be an unexpected masterpiece.

Just how strange can things get? If you ask James Briscione, the director of culinary development at the Institute of Culinary Education who has been working with IBM for three years, interacting with Chef Watson can get downright funky.

“One day I said, ‘Let’s make a pork belly dish, and see what pops up,’” recalled Briscione to the Kernel. After entering a few basic ingredients, what came out was Czechoslovakian porkbelly moussaka. “It was definitely the dish in which I had the least amount of confidence,” he said with a laugh.

After a while, he said, you have faith the computer knows a thing or two about building offbeat yet tasty dishes.

Despite all reservations, apparently, Chef Watson knew what it was doing.

“What came out was fantastic,” Briscione said. “Students enjoyed it, and we offered it to staff people who walked by our kitchen. Everyone loved it.”

Briscione brings an impressive résumé to the table—a background in sous vide cooking techniques and a trophy as Food Network’s first Chopped champion—but he admits chefs can get locked into a certain way of doing things, and a good shove in a new directions can lead to some great outcomes.

For example, Watson taught him a thing or two about making breakfast pastries.


Fennel-spiced ribs with tangy apple-mustard barbecue sauce | Photo via IBM Research/Flickr

When Watson dialed up the ingredients for spanish almond crescents, Briscione received some unexpected incidents. “When you think of breakfast pastries, you don’t normally think of black pepper and saffron. Also, crescents don’t normally have yeast and never call for oil and not butter.” After a while, he said, you have faith the computer knows a thing or two about building offbeat yet tasty dishes.

“It’s often challenging to be locked into a set of ingredients not even knowing where to start,” the chef said of Watson’s sometimes-curious output. “But the system can predict all the things that will taste good together. It forces you to become more creative.

“You are starting from zero and trying to discover new combination, and challenge the way you used to do things.”

Elementary, my dear Watson

Moving out of the professional kitchen and into mayhem of my own cooking venue, I put Chef Watson through its paces. For my trial, I limited Watson to my own vegan diet, eliminating dairy, meat, fish, and other such no-nos from the list of possibilities.

The more consumer version of Chef Watson was created using 9,000 recipes from Bon Appétit as a baseline. That data set gives the supercomputer a starting point upon which it can curate new ideas based on parts of existing themes and its ability to understand the science of food—that is, what items work well together and which ones will leave you out of the clean plate club. 

After entering ingredients and the type of dish I was interested in, I began receiving suggestions for what would pair nicely, as well as the theme of the cuisine—holiday, barbecue, Fourth of July, Mediterranean. As I began to further refine my ideas, a column called “Inspiration station” began to populate with recipes.

My mission was to create a four-course meal (dessert included) and let the chips fall where they may.

Did I mention Chef Watson allowed me to create a shopping list for those things not always in my pantry or fridge?

I asked my partner to consider an appetizer using celery root and mushrooms with sweet paprika. Starting off slow, as not to blow out any circuitry, I allowed it to go “classic.” The result: “American Sweet Paprika Celeriac Carrot Shiitake Mushroom Pickles.” Not a dish that rolls easily off the tongue, but it sounds appealing—at least to me. The ingredients were spelled out in great detail, as were the suggested steps for preparation. Did I mention Chef Watson allowed me to create a shopping list for those things not always in my pantry or fridge?

Going a bit more exotic for my soup course, I offered escarole, chard, and green olives as a deliberately obtuse base. The net result was the tongue-twisting “Galician Escarole Chervil White Corn Green Olive Rice Soup.” This Spanish-influenced vegan soup brings together such seemingly mismatched items as orange juice and white corn for the recipe. (Also, I neglected to tell Watson I don’t eat soy, so it had soy cheese as one of its toppings. No matter: There are plenty of replacements to be found in nut-based cheeses.)


Photo via IBM Research/Flickr

For the main course, I dared Watson to put lima beans in the same dish as blood oranges and figs. Up to the task, the result was “Tailgating Mission Figs Blood Range Baby Spinach Cider Vinegar Lima Bean Main.” (While Watson appears undaunted by even the most oddball request, it could use some work naming its crazy concoctions.) 


For dessert, I went way off the reservation and wondered if IBM’s Jeopardy!-beating pile of transistors would have difficulty blending cucumbers, fennel, blueberries, and garlic. Without hesitation, we have “French Wild Blueberry Vegan Dairy Gazpacho.” I did not ask what made it French, but I am sure Watson has its reasons. 

Looking over my four-course, computer-generated feast, I see nothing that could stump even first-time chefs. For the at-home cook, Watson offers a precise list of ingredients and a step-by-step guide with each action clearly described. As long as you know a pot from a pan and an onion from a blueberry, you should be in good shape. There’s also plenty of room to improvise, and Watson boasts a “surprise me” mode to add layers of complexity.

I don’t want to hurt Watson’s feelings, but I think it might be nice if it offered me some nutritional information, as well as telling me in advance the overall cooking time. Picky, I know, but Chef Watson seems to have the chops to do everything but scrub the pots and fold the napkins.

The next time my wife and I battle over what’s for dinner, I will take out my tablet, ask her if she is in the mood for a bean dish, and suggest we try “Family Reunion Epazote Black Bean Cassoulet.” I’ll even offer to print it out, tie her apron, pour the wine, and set the table. Wake me when it’s ready.

Illustration by J. Longo