Sony’s FlavorGraph uses AI to predict which ingredients will pair together

AI has gone into games, self-driving and other areas with mixed success, and now it’s trying its hand at cooking. After Google’s AI went up against a Great British Bake Off winner, Sony has developed a deep learning system called FlavorGraph designed to pair up ingredients like garlic, olives and milk.

Sony and Korean University (KU) researchers noted that chefs figured out how to combine ingredients through intuition, resulting in a gradual evolution for pairings like cheese and tomato, pork and apple and garlic and ginger. Many of those classic combinations were later explained by science, as researchers realized that ingredients sharing dominant flavor molecules often worked well together. At the same time, other ingredients that combine well may have vastly different chemical makeups. 

To figure out why, the team examined both molecular information about ingredients and how they’ve been used historically in recipes. They then created the FlavorGraph database with flavor profiles like bitter, fruity and sweet based on 1,561 flavor molecules. At the same time, they examined nearly a million recipes to see how ingredients have been combined in the past. 

The resulting data shows the chemical compounds shared by foods like wines and citrus groups and how they affect their overall taste, showing which foods might go well with specific wines or fruits. Some of the sample food pairings are obvious (cookies and ice cream) and others less so (white wine and Campbell’s condensed golden mushroom soup). The researchers didn’t yet discover anything extraordinary (citing caviar and white chocolate as an example of that), but the FlavorGraph is just a starting point. 

“As the science develops and we get ever better representations of food, we should discover more and more intriguing pairings of ingredients, as well as new substitutes for ingredients that are either unhealthy or unsustainable,” the team wrote.