By Florian Pinel
Raise your hand if you have interests outside of your day job. Probably most everyone, I imagine. Now, how often do they serendipitously collide? Probably not that often, right? But that’s what happened for me two years ago when I applied my computer science skills to my love of the culinary arts as part of IBM’s Cognitive Cooking project.
If you’re attending South by Southwest in Austin from March 7-11, come meet me at the IBM Food Truck. I’ll be showing how this recipe-generating technology works, while chefs will be preparing the dishes you can vote for on Twitter.
The idea started when my team was brainstorming on “Watson-like projects” – nothing so specific as “cooking,” yet. We wanted to know: could the cognitive computing that Watson uses in healthcare and other industries, also be creative?
That angle then spun into “could a machine come up with a recipe we could make into a dish that we would actually want to eat?” You can read about its early iterations, and its potential societal impacts, in the 2012 IBM 5 in 5: Taste prediction.
This idea of building a virtual chef was especially appealing for me because in 2005 I earned a Culinary Arts degree. After working at IBM during the week, I would go to the Institute of Culinary Education (ICE) in New York on weekends to hone my hobby into a craft. Now, I am being asked to write algorithms that come up with novel preparation methods for everything from stews to cakes into a computer system.
First, we built a database of online recipes. Then, we applied algorithms to the system in three ways:
- The combinations between geographic regions, ingredients and dish type had to be creative, so not only be of quality, but novel.
- We must be able to evaluate the combinations for their hedonic psychophysics. Or, would a quiche that incorporated traditional Swiss and Thai flavors taste good?
- And the combinations must also be in correct proportions for a recipe. (Get the recipe for Turkish bruschetta and other machine-generated recipes, here.)
The system comes up with a new recipe that has a good flavor profile score after we manually input region, ingredient, and dish type. Keep in mind that the new recipe also adheres to fundamental cooking knowledge. It doesn’t, for example, try to reinvent how ice cream is made The actual cooking, of course, is then left to us. You can get an idea of the system output – from ingredient pairings to “surprise” ratings – in my break down of Baltic apple pie on my personal blog Food Perestroika.
Today, we’re working with ICE to further develop and refine these recipes. We already served several never-before-seen dishes out of the IBM Food Truck at IBM’s Pulse Conference last month. Next, the truck will be in Austin on the corner of Red River and 4th Streets for SXSW Interactive. Come check out the Chef Watson tech and try the food, too.
I think any chef or home cook would like something that helps them put a new twist on food they like – and never have to eat the same thing twice. But this technology has many other applications. For example, it could account for food allergies, or ingredient substitutions in places where certain foods are not grown, or easily obtained.
And while it’s great to be working on something that I’m also personally invested in, we’re already thinking about how to apply these “creative” algorithms to domains outside the culinary arts and into areas like fashion, and fragrance. What about a new perfume?
About the IBM Food Truck
- For more about this research and how it can transform your business, contact Research Staff Member and Manager of Consumer Modeling Anshul Sheopuri.
- To learn more about Watson and the Food Truck project, read IBM Fellow Rob High’s perspective.
- Get a chef’s perspective from ICE Creative Director Michael Laiskonis.
- Download images of the Watson Food Truck.
- Have more questions? Check out the Watson Food Truck fact sheet.