Science

IBM Chef Watson cognitive cooking app generates recipes on demand

A new "cognitive cooking" app based on IBM's Jeopardy-winning Watson computer is designed to come up with new recipes that have never been seen before, using the contents of your fridge.

Creations include Vietnamese apple kebabs, Belgian bacon pudding, Austrian chocolate burritos

IBM's Steve Abrams (left) and Dawn Perry, senior food editor for the magazine Bon Appétit, test out Chef Watson in the Bon Appétit kitchen in New York City. (Jon Simon/IBM)

You might not have a personal chef in your kitchen, inventing innovative recipes and techniques every night. But IBM may have the next best thing: An artificially intelligent kitchen helper called Chef Watson.

Chef Watson is based on the Jeopardy-winning computer that IBM showed off in 2011 and was programmed by Florian Pinel, a senior software engineer at IBM. It communicates with the app via the internet.

"We started about two and a half years ago," Pinel said. "Watson had just won on Jeopardy ... and we wanted to see if we could push the boundaries of cognitive computing even further to help humans be more creative."

Chef Watson has already been put to work at IBM events, in cafeterias and the company's food truck. The idea is pretty straightforward, when given a handful of ingredients, Watson is able to create new, original recipes. In other words, Watson acts like a chef who might be looking in your fridge or shopping at the farmers' market.

Earlier this year, at the SXSW Festival in Texas, Chef Watson helped the IBM food truck churn out Vietnamese apple kebabs, Belgian bacon pudding, Austrian chocolate burritos and Peruvian potato poutine.

Until now, cooking apps have helped you in the kitchen by essentially acting as complicated search engines looking through existing recipes.

This is the end result of a recipe for tamarind-cabbage slaw with crispy onions, courtesy of Chef Watson. (Jessup Deane/Bon Appétit)
"What makes it [Chef Watson] completely different from other cooking apps ... is that here we create new recipes that have never been seen before," Pinel said.

"This is not about search, this is about helping people be more creative and create recipes that are tailored to their dietary needs and preferences."

In order to be able to generate brand new recipes, Chef Watson was trained using 35,000 existing recipes. From those, it learned about different cuisines, what ingredients typically go together, and what it takes to make certain types of dishes, like soups or burritos.

Computer-assisted cooking since 1969

Cooking apps, or computers in kitchens, are nothing new. In fact, the very first home computer was meant for the kitchen. The Honeywell Kitchen Computer was released in 1969.

It's better at considering all the possibilities without bias.- Florian Pinel, IBM

But it didn't take off. It weighed 45 kilograms, cost $10,000 and required the user to learn binary code to use it. To search for broccoli recipes, the user had to enter the code for broccoli: 001101000.

As things became more user-friendly over the decades, the computer in the kitchen acted as a sort of virtual database, a giant binder full of existing recipes.

Chef Watson may be a step up from that old virtual database, but is he better than a real deal chef, with years of experience in the kitchen?

"It's better at considering all the possibilities without bias," Pinel explained. "I think you know that the way chefs work is that they have ingredients they like, the dishes they like, they have a certain style. Chef Watson doesn't have the bias, so it's going to help you diversify the dishes you create."

But bias can be a handy thing in the kitchen, and that's where Chef Watson might fall flat. A chef's bias might be based on previous experience, and a bias against combining certain ingredients might come from earlier disastrous results.

Chef Watson doesn't have years of experience, and so he might be able to produce some unexpected and unusual hits — and maybe the odd miss as well.