Baking is as much science as it is art. Perhaps to find out whether the former's more important, Google Cloud AI is taking on a Great British Bake Off winner in a dessert face-off. Sara Robinson, an amateur baker and Google Cloud developer advocate, built a machine learning model that examined hundreds of baking recipes (including ones for traybakes, cookies and scones) to help her come up with a new one.
The model generated lists of ingredients and amounts that were used as the basis for step-by-step recipes. The model was able to come up with hybrid recipes and Robinson opted for one that had a machine learning-generated cake batter on top of a machine learning-generated cookie.
She experimented with the model and baking to create a desert that incorporates Maltesers, a malt chocolate candy similar to Whoppers. The result was Maltesers AI Cakes (4d6172730a) — the last bit is hexadecimal for "Mars," which owns the Maltesers brand. The company had approached Google to create an AI-driven dessert. Meanwhile, Peter Sawkins, the 2020 Great British Bake Off champion, created his own Maltesers cake to compete with the AI one as part of a promotional campaign.
Both Robinson and Sawkins used slightly unusual ingredients to balance out the sweetness of the chocolate with salt and umami. Robinson added a Marmite-infused buttercream frosting after Google Trends identified that "Is Marmite sweet or savory" was a top search related to "sweet and salty." Sawkins, on the other hand, added soy sauce to a recipe he adapted from one of his mother's creations. You can download both recipes via Mars' Bake Against The Machine website and try them at home to find out which is your favorite.
Sure, AI systems are tremendously useful for things like tracking wildlife and guiding vision-impaired people. But, to actually generate better dessert recipes than ones us mortals can come up with alone, this model will need to learn to leave out the Marmite. Yuck.