Image credit: Consumer Goods/Mars Wrigley Maltesers® AI Cake
As the largest search engine in the world with over 92 percent of the market, Google has a lot of insight into global trends and data. Many worry that company’s data goes only toward targeted ads, but what if we told you that Google used this information to help create a nearly perfect dessert with all the right fillings?
Google did just that, and here’s how it unfolded. In 2021, Google Search Trends reported that public search for the term “baking” jumped up a whopping 44 percent, as opposed to the same time in 2020. Sweets company Mars, Inc., maker of 3 Musketeers, M&Ms, Snickers, and more, saw the opportunity to ride this popular trend by merging baking with AI and ML.
The Power of Search Engine Optimization
Mars Wrigley UK linked up with Google to get the sweet details on what types of baked treats and toppings their audiences were searching for. Google utilized its Cloud platform and Search Trends data to develop a machine learning model for Mars Wrigley, and enlisted Sara Robinson, Senior Developer at Google Cloud.
This model searched hundreds of recipes, ingredients, and amounts to give Mars ideas based on current trends around their base in the UK. The team settled on creating a Maltesers cookie/cake hybrid, or “cakie,” composed of a cookie on the bottom and a cake on top. Maltesers are a popular British candy made by Mars. They have an airy AI malted milk center with a chocolate coating. The resulting AI-cookie/cake hybrid has a light and fluffy center with a dripping, marmite-infused buttercream base and golden syrup.
Image credit: Maltesers® AI Cakes
Whipping Up Recipes in the Cloud with AI
Robinson used Google Cloud for the tooling to build the model, starting with Cloud AI Platform Notebooks for feature engineering and model development. After visualizing the data and generating statistics on it, she scaled the model inputs so that all ingredient amounts fell within a standard range.
With data preprocessing complete, Robinson then fed the data to a model, using TensorFlow’s Keras API. For optimal model architecture, she made use of AI Platform Hyperparameter Tuning, a service for running multiple training job trials to optimize a model’s hyperparameters. Once she found the ideal combination of hyperparameters, she deployed the model using AI Platform Prediction.
Google data influenced ingredient choices. For instance, the frosting is “marmite-infused buttercream,” which was selected because Google Search trends highlighted the terms “Is Marmite Sweet or Savory” as a top search around their desired higher-level keywords, “sweet and salty.”
“This idea opened our eyes to the endless possibilities of how AI can bring innovation to the kitchen by creating a foundation for recipe development,” says Sam Chang, the Global Head of Data Science and Advanced Analytics at Mars.
To help market the new dessert with a story around making “better moments,” Mars used its AI creation to put together a campaign-bake-off, dubbed “Bake Against the Machine.” Here, tasters voted on what tasted better: the AI-created cakie versus an opposing baked treat created by a human baker.
Voters weren’t sold on the cakie hybrid, however, as it lost to the 2020 Great British Bake-Off winner Peter Sawkins. Technology lost to tradition, as he used an old family recipe passed down from his mother to inspire the winning creation.
But look for more AI-bakes in the future. Developer Robinson states, “AI is a powerful tool, and I love thinking about the fun, new applications for it. In this case, maybe #BakeAgainstTheMachine will even inspire other amateur bakers out there to build their own creations.”
Sawkins describes his experience winning the competition, “Baking against the machine is a fun challenge. Not many people can say they’ve baked head-to-head against a Google Cloud AI.”