Could a Computer Win Top Chef?

Could a Computer Win Top Chef?

Remember Watson? In 2011, it won Jeopardy! in a three-match series against two humans. Watson’s performance wasn’t perfect but it demonstrated the capabilities of the future of cognitive computing, which is a field focused on programming computer systems to “learn and interact naturally with people to extend what either humans or machine could do on their own,” according to IBM’s website.

December 17, 2014
Remember Watson? In 2011, it won Jeopardy! in a three-match series against two humans. Watson’s performance wasn’t perfect but it demonstrated the capabilities of the future of cognitive computing, which is a field focused on programming computer systems to “learn and interact naturally with people to extend what either humans or machine could do on their own,” according to IBM’s website.

IBM developed Watson, and after the computer’s triumphant win on Jeopardy!, the team started looking for more ways to apply the technology.

One of the first things they thought of? Cooking. So Chef Watson was born.

Cognitive computing is based, in part, on large databases of information that a computer system draws upon, aka “big data.” In Chef Watson’s case, that big data is composed of information about chemical and flavor compounds found in foods. The computer then scans that info and analyzes it in order to recommend ingredient pairings which cooks can use to create new dishes.


“It forces you out of your comfort zone,” says James Briscione, the director of culinary development at the Institute of Culinary Education in New York City, but Chef Watson still “doesn’t know how to flip an omelette.”


Briscione and his colleague Michael Laiskonis have been collaborating with IBM on the Chef Watson project for about two years now. Their goal? To make Chef Watson a professional tool for both training and creative purposes. Its results can be surprising, even for professional chefs—such as bananas prepared with cilantro.


“I tried it, but if the bananas had been unripe, it might have worked,” Briscione said, reflecting on that particular attempt to use Chef Watson’s suggestions. However, in a different experiment, “bananas and chilies went really well together. It was a sriracha-like sauce based on bananas. It makes sense. It’s spicy and sweet.”


In hindsight, the pairings often make sense but that’s what makes Chef Watson a powerful tool—its ability to cull through immensely large swaths of information and turn around results in mere seconds.

Skynet approacheth?

Briscione isn’t worried about computers taking over his kitchen Skynet-style, though.

“It’s referred to as Chef Watson but it’s software in a cloud. We all watched Watson compete on Jeopardy but he doesn’t walk into the kitchen. We are the chefs and Watson is the advisor,” Briscione said.

The real chefs still have to do what they do best: create recipes and menus. They demoed the system at SXSW this year and produced several crowd-pleasing results, including a Vietnamese apple kebab and an Austrian chocolate burrito.


A version of the program geared towards home cooks is currently in beta, too. After it analyzes flavor and chemical compounds to find ingredients that are well suited for each other, it then taps into Bon Appétit’s database of 9,000 recipes.

Less than six months old, it still requires human brain power to make sense of both the pairings and the recipes, but just like the professional version, the goal is to inspire culinary creativity.

Florian Pinel, a senior software engineer for the IBM Watson Group, has been working on the project since it began about three years ago.

After Watson’s Jeopardy! win, “we wanted to see if we could push the bound of cognitive computing further,” Pinel said.

To match ingredient recommendations with recipes in Bon Appétit’s database, the program calculates a score based on three formulas that measure chemical compounds, pleasantness (flavor compounds) and the element of surprise, that latter of which uses a mathematical equation called the Bayesian theory of surprise, Pinel explained.


I took the beta for a test drive last week to get a sense of what Briscione's experience is like using Chef Watson, and surprisingly (or not) the results were delicious. A bunch of collards were sitting in my vegetable drawer, so I started there, entering the ingredient into Chef Watson’s database and it gave me coffee—well, specifically, coffee beans. Hm.

After running a search for recipes based on that pairing, Bon Appétit’s database produced a suggestion for chicken burgers with collards, oregano, parmesan, bacon and coffee. (See the original recipe here and my annotated version here.)


Just as Briscione alluded to the computer's inability to flip an omelet, Chef Watson's "suggested steps" for techniques weren't very helpful, but the flavor pairings were spot on and definitely provided some unique inspiration.

So, while Chef Watson might be too advanced for the beginning home cook, (it’s not so elementary after all, my friend), anyone with a solid understanding of kitchen and culinary basics might find the program useful and fun to explore. And it will certainly be interesting to see what professional chefs like Briscione discover with this technology and create.

Who knows, maybe Chef Watson will be the next Top Chef.