Can a Computer Cook?

Computers, it seems, have been edging up on us ever since the early 1800s, when Charles Babbage and Ada Lovelace collaborated on the Difference Engine, the first mechanical computing machine.

In the 1957 movie Desk Set, brainy Katharine Hepburn trounced a version of ENIAC, the world’s first electronic general-purpose computer—she was far better than it at recalling the names of Santa’s reindeer—but since then it seems to have been all downhill for the human race.

In 1997, I.B.M.’s mathematically ept Deep Blue defeated Chess Grand Master Garry Kasparov in a landmark game of nineteen moves. (Kasparov accused his electronic opponent of cheating; Deep Blue sullenly refused a re-match.) And in 2011, I.B.M.’s supercomputer Watson defeated all-time champ Ken Jennings at Jeopardy, successfully answering knotty questions about the Beatles, the Olympics, Harry Potter, and Dracula—and, incidentally, raking in a prize pot of $3.25 million.

Thinking Like a Human, but Better

Watson’s specialty is answering questions—but it’s far cleverer at it than are search engines such as Google or Bing, which simply spit out a long list of links and expects us users to dig out the solution for ourselves. Instead, Watson delves into its immense digital database, using multiple algorithms to analyze a question in hundreds of different ways, and generating hundreds of possible solutions. It then ranks these in order of probability to come up with—if not the one single correct answer—at least the best possible educated guess. Scientists refer to Watson’s modus operandi as cognitive computing. Watson, in other words, may think somewhat like a human brain.

And now Watson, with hands-on help from chefs at the Institute of Culinary Education (ICE) in New York, is tackling cognitive cooking.

Generating a new recipe, it turns out, is no simple feat. Given the number of available ingredients and flavors, there are easily a quintillion—that’s a one with eighteen zeroes after it—different ways in which to put foods together in a dish, and no human cook can possibly evaluate them all. Watson’s gargantuan memory bank and lightning speed, however, are more than equal to the task—and it turns out that Watson can be creative, too.

Shrimp and Licorice? Caviar and White Chocolate?

Watson’s cooking expertise begins with its backlog of some 35,000 recipes, which collectively provide basic information about food composition and flavor pairings. (What’s a quiche? What’s ratatouille?) It also knows the molecular chemistry of over a thousand different flavor ingredients—everything from black tea to Bantu beer—and has input from the racy-sounding field of hedonic psychophysics, which quantifies the tastes and flavor sensations that people tend to like. (Shrimp and licorice? Caviar and white chocolate? Blue cheese and rum?) Watson’s mission, based on these data, is to invent recipes that are both yummy and unconventional. And it looks like it’s succeeding.

Given a theme and a description—say “Spanish” and “breakfast bun” or “Thai” and “sweet potato”—Watson can come up with any number of suggestions with lists of novel ingredients. Its culinary mix-and-matches have produced such unexpected combos as bearmeat with saffron and sandalwood, avocado Napoleons, and an off-the-wall kebab featuring pork, chicken, strawberries, shitake mushrooms, pineapple, apples, curry, green onions, carrots, lemon, lime, and mint. Other Watson inventions include Creole Shrimp-Lamb Dumpling, Baltic Apple Pie, Austrian Chocolate Burrito, and Bengali Butternut BBQ Sauce, this last a scrumptious-sounding blend of white wine, butternut squash, rice vinegar, dates, cilantro, tamarind, cardamom, and turmeric, plus such old-time BBQ standbys as molasses, garlic, and mustard.

Are Machines Making us Obsolete?

Watson’s promoters also see cognitive cooking as an opportunity to promote healthy eating—by coming up with inventive recipes that are low in fats and sugars, or specifically targeted at the diabetic, the lactose-intolerant, or others with special dietary needs. Perhaps, with Watson in their corner, school lunch providers will be inspired to move away from chicken fingers and Sloppy Joes.

Proposed possibilities for cognitive computing are now legion, extending to any field that requires rapid analyses of large and tangly amounts of data. Watson may find a niche, for example, in the healthcare system as a medical diagnostician, in biopharmaceuticals as a predictor of new and effective drugs, as an analyst in the sales and travel industries, as a consultant in the home-buying market. It may even eventually be able to help you find a job.

In our computer-dominated world, it can be nervous-making to contemplate our own clumsy capabilities. Is there anything we’re good at that computers aren’t? Are machines making us obsolete? Are our hard drives secretly laughing at us up their electronic sleeves?

According to researchers, there still are a few skills in which humans have it all up on computers. We’re better at pattern recognition. We’re more emotive. We’re more innovative. We tell better stories. And we’re better at non-routine physical tasks such as gardening, portrait-painting, cabinet-making, and fly-tying—and, of course, cooking, which is why Watson leaves the real in-the-kitchen work to a battery of trained chefs.

In the realm of food, however, people have one magnificent advantage over even the biggest and brightest of electronic brains. Computers may be brilliant when it comes to designing recipes. But people can eat.

Photo of a chocolate burrito.

Watson’s Austrian Chocolate Burrito

Ingredients:
Lean ground beef: 1 lb. browned and drained of excess fat
Zest of one orange
Pinch of ground cinnamon
Dark chocolate: 2 ounces, (70% cocoa solids or higher), very finely chopped
Apricot puree: 1/2 cup
Vanilla bean: 1/2 bean, split and scraped
Edamame : 1 1/2 cups, shelled
Fine sea salt: as needed
Flour tortillas : 6
Vegetable shortening : melted, as needed
Edam cheese: 1 cup, grated
Queso fresco : 1/2 cup, crumbled

Brown the beef and drain. While still warm, stir in the orange zest, cinnamon, and 1oz chocolate. Season with salt, and reserve for assembly.

In a saucepan, combine the apricot purée and vanilla and slowly reduce over medium heat to roughly ¼ cup. Remove from heat and stir in 1/2 oz chocolate. Cool and reserve for assembly.

Next, blanch the edamame in boiling salted water for about 1 minute, then drain and shock in ice water.

Transfer the edamame to a food processor and pulse to achieve a rough textured paste. Season with salt and reserve for assembly.

This story is part of National Geographic’s special eight-month Future of Food series.

For more of Watson’s recipes and information on cognitive cooking technology, see I.B.M.’s Cognitive Cooking.

Thompson, Clive. What Is I.B.M.’s Watson? New York Times, 16 June 2010.