Cooking Watson: super computers become recipes.


Cooking Watson: super computers become recipes.

IBM’s Watson supercomputer is now not content with the winning quiz show and is now injecting a bit of creativity into your kitchen.

I like to think I’m a pretty good cook. I took the risk of having a decent understanding of what I needed, and I haven’t poisoned anyone yet. But I still find myself cooking the same experiment and trust, using the same old ingredients.

To change this, I asked to test the prototype of a brain-cooking application using IBM’s Watson supercomputer to create new dishes.

The key to the ChefWatson application is the ability of a supercomputer to consume large amounts of information and build links between it. It has proved its courage and won the quiz show! It was used to help doctors diagnose cancer at memorial sloan-kettering hospital in New York. Now, it’s trying to do something that computers often find hard to find: using creativity to invent recipes, people actually want to eat.

To provide these data, IBM partnered with Bon Appetit, a us cookbook website. There is a database of more than 9000 recipes, with labels based on their ingredients, types of dishes and cooking styles – such as Indian or Thai. Watson creates a statistical correlation between the composition, style, and recipe steps in the database, and USES it to determine what ingredients are usually together and what ingredients are needed for each food. Steve Abrams, of IBM, said: “this is how it knows that rolls, burgers and soups require different elements. “It knows that the rolls always need some kind of packaging, and the soup always needs liquid. That’s what you don’t have. ”

To use the application, the chef first inputs the ingredients they want to use. Next, they decided that they wanted watsons to take the number of experiments, from “keep classic” to “surprise me.” Watson then provides the chef with further ingredients, styles and dishes, which he thinks are usually very good with the original ingredients. They can promote or remove ingredients by clicking “love it” or “hate it” buttons. Finally, the chef searches, Watson analyzes its database to present a series of basic recipes, which can then be further tweaked to make them more or less experimental.

If cook wanted to increase the recipes, you can query all kinds of food flavor compounds contained in the database, and use it to suitable composition in theory, such as vodka and Brussels sprouts, cauliflower and chrysanthemum. (PLoS computational biology,, psychological research on the “surprise” fraction of composition, the higher the score, the lower the frequency of the recipe.

“When you get to the end of the experiment, you see what kind of ingredients are fused together, and more often in the common characteristics of these components,” abrams said. “It solves something that might be together, and you never think of it.”

Well, at least that’s the theory. From my cooking, it’s easy to see why the application hasn’t been seen as ready to launch. In a creamy pasta (I chose the “elegant” style), caramel fraiche has been replaced by a glass of milk. Another time, in a very high experimental environment, Watson told me that my tuna would need half a kilo of goose. I refused.

This is an occasional inspiration. When I was pleasantly surprised to find the element of surprise, Watson suggested that I replace the clam juice in the classic soup plate with a Japanese soup base. Its work – to give this dish a deep taste, I will never think of trying.

Michael Laiskonis, creative director at education college in New York, is one of the first professional chefs to help Watson. “It allows you to organize your creativity in a more organized way, rather than experiment in the kitchen,” he says.

The plan is to make ChefWatson more complex and to absorb data from more sources, said Florien Pinel, chief software engineer for the project. It has been searching wikipedia’s world food web site and looking for nutrients in the usda database to help determine the proportion of ingredients.

In addition to the occasional odd mix of ingredients, ChefWatson has other flaws that make ChefWatson conjure up ideas from pure data. One tester, Matt McSherry, said it seemed to be about the size of the score. Another found a recipe that required 554 gin, and she was told “skin and bones” her tofu.

All in all, he enjoyed the experience very much. “I got dirty in the kitchen with a thinking supercomputer. “That’s really exciting,” he said. “This is the future we promise as children.”

Thai Turkey stuffing – la ChefWatson.

6 doses


450 grams of Turkey

Frozen pastry

Half the seeds, chopped Thai chili.

1 quarter teaspoon of rice flour.

Dash lemon lemongrass.

Green curry paste

1 delighted many customers a head of lettuce

500 grams of potatoes, chopped.

13 scallions, minced

1/2 teaspoon vegetable oil.

Olive oil spray

Gruyere 115 grams, chedin.

100 grams of prostrate cheese.

Recommended steps

Cook lettuce in boiling water.

Drain and squeeze.

3. Heating vegetable oil

Add spring onion and Thai chili and stir-fry for about 7 minutes.

5. Fine Turkey, cheese, lemongrass and rice noodles.

Transfer to a bowl and stir in scallions, lettuce and potatoes.

Season with salt and pepper.

8. Preheat the oven to 180 ° C

Spray a large baking pan with oil.

Layer and stack pastries and spray with olive oil.

Apply the Turkey mixture to the center of the pastry.

Fold the short side of the pastry on top and roll it into a log.

13. Bake for about 40 minutes.

Set aside the green curry sauce and serve.


Please enter your comment!
Please enter your name here