Last night I ate a computer-generated meal. It was fantastic.

A bourbon cocktail with turmeric and banana juice. Two different preparations of duck with ingredients that probably have never before been paired with a duck. An olive and cherry compote that delivered a flavor profile that was completely delicious and utterly different. All ingredient pairings suggested by IBM’s artificial intelligence software and then executed and adapted by chefs from the Institute of Culinary Education.

This idea of Chef Watson, using the same software as the AI powering IBM’s Jeopardy-playing computer that has since transitioned into a financial advisor and a medical assistant, has been written about for the last year. But at South by Southwest taking place over the next few days in Austin, Texas, attendees of the show will get a chance to taste the recipes inspired by IBM’s cognitive cooking program via a food trailer cooking up a crowd selection each day. Combining food and data to generate new recipes is not an IBM specialty, several startups such as Yummly, Food Genius and Foodpairing are all working on this problem.


The idea — repeated by countless videos and several times last night — is that humans can only handle about two or maybe three ingredient pairings before hitting a creative wall. However, IBM’s computer can handle many more suggestions — up to a quintillion flavor pairings — helping augment chefs’ creativity with computer-generated ideas. The chefs at last night’s event seemed genuinely excited by the possibilities discussing their original doubt over the proffered list of ingredients and then their eventual surprise as they managed to create something worth eating.

IBM’s taking information from three different databases to turn Watson into a chef, from a database that includes the molecular compounds in food to one that measures how people react to it, and then adding the ability to match the databases against novelty and cultural preferences around taste. In some ways, IBM’s cognitive cooking experience can accelerate the new combinations of ingredients we’ve seen as chefs travel the world tasting new flavors and bringing new ingredients to their native lands. For the details, check out this video:


But amid the technical details about how the cognitive cooking process works, I noticed that IBM staff never called the machine Watson, which seemed odd given that Watson seems to be ready to compete with Barbie for the number of careers it can have. When I asked, a representative from the IBM camp told me that it was because the “official” Watson runs on IBM’s Power 7 servers while the cognitive cooking program runs on SoftLayer’s cloud. So while the process of probabilistic reasoning is the same, the data sets and hardware are different.

Which brings up an interesting question, what is Watson? Given that the databases are different for cooking, medicine, Jeopardy! and financial planning, does the hardware really matter for an AI system? When pressed, an IBM spokeswoman said: “The cognitive cooking technology you saw last night is a Watson system. It’s not the same Watson technology that played on Jeopardy! or is being used in healthcare.”

Since the goal of last night’s event and Watson’s whole chef career is an effort to show that machines can succeed in a creative endeavor (although I would argue that it does require human interpretation), it appears that perhaps Watson’s next creative effort might be a journey of self discovery. It may sound silly, but as IBM prepares to deliver cognitive computing as a service, it’s actually a topic worth pondering.

We’ll actually discuss Watson’s future as a cloud service in less than two weeks at our Structure Data event March 19 and 20 in New York. A main theme of the conference is using data to build entirely new products and capabilities rather than just using “big data” as a euphemism for “better business intelligence.” IBM seems to be doing just that.