Supercomputers Offer a Recipe for Success

Supercomputers Offer a Recipe for Success

What could you use a supercomputer for? Oh, studying particle physics, designing a rocket engine, beating a chess master or winning at Jeopardy. Even developing a recipe for Indian Turmeric Paella. Yes, developing a paella recipe. That’s what IBM Research has done as it works to develop an AI system that can create food recipes.

It might sound a little wacky, even to someone who just completed a online computer science degree, but it isn’t. Tackling kitchen creativity is the next logical step in developing artificial intelligence systems that can do far more than offer a good meal, including solve complex business problems, infer the hidden interests of crowds, or, like TV’s Dr. House, diagnose someone’s illness from a seemingly bizarre collection of symptoms, as IBM is already trying to do with the same computer that won at Jeopardy.

To understand why developing recipes is important, look first at the development of IBM’s high-profile attempts at AI in games. In beating former world chess champion Gary Kasparov, the company’s Deep Blue system had a task that was conceptually exhausting and simple at the same time. It had to recognize chess movements, see the patterns, look at the possible implications and outcomes of the moves, and choose the best one to get closer to the goal of checkmate. The number of potential patterns is enormous, but finite and it quickly narrows each step of the way. The goal was clear.

Watson, in playing Jeopardy, had to hear an answer, parse out the meaning of the sentence, and then move through connections and associations to find the right information that, formed as a question, would be right. The body of potential subjects and answers dwarfed that of a chess game. But there was only one right answer each time — although it wasn’t the ultimately the same, capture the king, in each case — and each piece of information narrowed down the choices.

Developing a recipe is similar in some ways. There is a large, albeit finite, list of potential ingredients. But after that, things change. Although there are principles and even rules of which ingredients might work together, the end goal is far more slippery: come up with something that people would like to eat. The system uses three types of information:

  • a recipe index, which demonstrates principles of how types of dishes are constructed;
  • a quantitative description of how much people like given flavors; and
  • a correlation between molecular flavors and ingredients and recipes.

For a human, it would be recipe books to teach structure, a wide set of experiences eating foods, and enough imagination and use of smell to tell what would work well together. Now the computer crunches the data and looks for novel combinations that people might like.

Such techniques and methods will become important as people and companies increasingly try to use computers to solve more fluid and subtle problems that may have no objective correct answer, but a series of choices that might be better for one company or worse for another. As a result, the way companies work, and the way they employ IT, will change drastically in the near to moderate future. To maintain and expand a career, that there’s no time like the present to learn a lot more about your field.

Tagged as