At SXSW, food trucks create prime opportunities for brands to influence the engineers, marketers, and executives who converge on Austin each March. Among the usual contenders at the 2014 SXSW was a strange food truck from an unlikely brand–IBM.
It was a huge success. For IBM, cooking has become a “killer app” scenario for Watson. So earlier this week, Bon Appetit magazine and IBM jointly released the Chef Watson portal, which applies Watson’s cognitive computing platform to the magazine’s massive recipe database to provide home cooks with “infinite recipes.” The invite-only portal lets users enter ingredients, the type of food they want to prepare (a sandwich? a stir-fry?), and a “style” to prepare food in such as Indian or Austrian, and then automatically generates 100 recipes based on those parameters. One of the big advantages for Watson’s data scientists is that Bon Appetit presented them with a recipe database that was pre-formatted and quality tested, making IBM’s job easier.
“We replatformed last year and switched to WordPress,” Bon Appetit digital director Stacey Rivera told Co.Labs. “At that time, we did a huge data cleanup, which meant the recipes themselves were well tagged and well organized. As media companies do, we have been organizing our tags and metadata over the past year, so it was seamless.”
But, of course, IBM isn’t working on food apps just because Peruvian poutine sounds delicious. Watson was designed to provide users with answers based around natural language queries, and cooking turns out to be an ideal way to sell potential clients on Watson’s gee-whiz factor.
Steve Abrams, director of IBM’s Watson Life division, recently wrote that by the system could someday even generate new recipes. “One great thing about having Bon Appetit’s repository is both the quality of recipes in the repository and that the tags show up in our user interface,” Abrams told me. “Watson learns about cooking by scanning recipes and seeing new ingredient combinations and how they are put together. The fact that recipes were previously put together is very important, because that means they were tested and validated.”
Because recipes are combinations of task lists and demonstrations of relationships between words and concepts–for instance, how cocoa and sugar show up in proximity and interchange far more than cocoa and avocado–they are a wonderful ground for attempting to understand instructional procedures and training software on a large corpus of work. One of the advantages of working with Bon Appetit, Abrams said, was that they offered IBM a higher-quality database of recipes than the public domain/Creative Commons recipes used to generate dishes for the SXSW food truck. According to Abrams, the public domain recipes they used had problems both in incorrect directions and outlandish flavor pairings or ingredients that skewed the final result. Bon Appetit’s database, with recipes written in a standard format and using techniques verified in the controlled environment of a test kitchen, helped IBM solve that problem.
And IBM isn’t the only company to see big value in the marriage of big data to food and recipes. Earlier this year, the Food Network and parent company Scripps Networks Interactive purchased a startup called Food On The Table, which combines users’ eating habits with sales information from local grocery chains to deliver personalized shopping lists and meal plans. Because the food industry and the logistics industry which supports it have a wealth of data available for import, expect Watson and its competitors to do more with dining in the future.