The Center for Data Innovation spoke with Florian Pinel, senior software engineer at Watson Life, a division of IBM’s Watson Research Center. Pinel discussed Chef Watson, a project that explores what IBM’s Watson supercomputer can discover about food and generate new and unexpected recipes.
Joshua New: You have both a master’s degree in computer science and engineering and a culinary arts diploma. People don’t typically associate artificial intelligence and cooking, so how did the Cognitive Cooking program come to exist?
Florian Pinel: Watson is all about cognitive computing, and we thought that one thing that was missing, shortly after Watson won on Jeopardy, was creativity. As in, “how could Watson help people be more creative, and help them discover new associations they never would have thought of?” That’s what inspired us to start this project. We picked food because it’s something everybody cares about and there’s also a lot of information available—recipes, information about ingredients, food chemical compounds, nutrition facts, research on people’s likes and dislikes, called hedonic psychophysics, and so on. We thought that by putting all this information together, we would have a chance to create new recipes and also predict how good they would taste.
New: How exactly does “Chef Watson” work? How can it know what tastes good?
Pinel: Chef Watson relies on a number of theories about taste. For example, the food pairing theory says that the more flavor compounds that ingredients have in common, the more likely they are to taste good together and appear in the same dish in western cuisine. Whereas in eastern cuisine, ingredients are more likely to be paired with ingredients that have less in common with each other, rather than more. So that’s one way Watson can evaluate whether or not something might taste good.
Another theory is hedonic psychophysics, as I mentioned. In an experiment, researchers had people smell flavor compounds individually and rate them for pleasantness. They found out that there was a consensus, but also that you could predict the pleasantness of a mixture by adding the individual scores of each compound. So, if we know the flavor compounds in an ingredient, we could assign it a universal pleasantness score. Then, a whole recipe could be evaluated by the sum of these scores.
Surprise is also very important, because we’re trying to develop novel recipes, not repeat existing ones.
New: Surely there is value to be had in other fields from this kind of pattern recognition. Can processes or insights from the Cognitive Cooking program be applied elsewhere?
Pinel: The importance of surprise is relevant to many other domains, beyond food. When we use Watson for drug discovery or cancer research, we want to show researchers associations that they don’t know already. For example, we’ve partnered with Baylor College of Medicine and they’ve been using Watson to help identify proteins that might be able to help fight cancer. Basically, what Watson discovery does is helps researchers identify things that are worth pursuing. It gives you more confidence in the chance of success for a particular research direction and it helps you get faster results.
I think creativity can be useful in any domain. Take financial portfolios, for example. What if you could use creative technologies to create new combinations of assets that your competitors haven’t thought about, that could maximize return on investment or minimize risk? We’ve also been approached by fragrance companies to help develop new perfumes. There are also other kinds of health applications. People with severe dietary restrictions, diabetics, people with heart disease, and others that need to closely monitor their diet are probably not happy about these kind of constraints. This kind of technology could develop recipes that could help satisfy cravings in ways these people didn’t think was possible.
New: Speaking of alternative uses for Watson, is there a goal for the Cognitive Cooking program beyond just tastier recipes?
Pinel: Helping people with dietary constraints, as I mentioned, is a big one. Additionally, many of the main causes of death in the United States—cancer, heart disease, diabetes, and so on—might not be entirely preventable, but people can greatly mitigate their risk by watching their diet.
Food waste is another one. About one third of the food produced globally is wasted. If you could have an app that could help you create enjoyable recipes just with ingredients you have in your kitchen, you would be less likely to waste food. We’ve had ideas about a futuristic kitchen with sensors in the refrigerator and trash can that could measure your consumption pattern and suggest recipes ahead of time so ingredients aren’t wasted. A retailer or supermarket might the same problem—they have ingredients that are going to spoil so they look for new recipes to repurpose that food.
New: What’s next for Cognitive Cooking? Or do you have your sights set on a different project with Watson?
Pinel: I think there’s always more to be done. Overall, we’re still learning how people want to interact with these systems. The cooking application helps us to understand what people expect from cognitive computing programs in general. People don’t just want a single result for a recipe, they want to interact with the information and get the right level of recommendations and customization that best suits their needs. We’re working on releasing new versions of the user interface that try to make that experience more valuable to our users.
