Mars Inc has tied up with Google Cloud AI in the UK to whip up a kitchen collaboration with its British candy Maltesers. According to Google Search Trends, baking was searched 44% more this year compared to the same time last year and both parties decided to jump on it and build a relationship between AI and baking.
To kick things off, Google's developer advocate Sara Robinson trained a new machine learning model to generate recipes for cookies, cakes, scones, traybakes, or any hybrids of those options. Several hours later, the model successfully combined chopped and whole Maltesers with an AI-optimised cake and cookie recipes to create a new dessert. The team, however, wanted to add a creative twist to it and decided to also create Marmite-infused buttercream.
According to Google, the project's aim was to create a model that could offer the foundation for it to build a new recipe featuring Maltesers and Marmite. Robinson got to work by using a dataset of British recipes to create the model. The data set comprised of four categories of popular British baked goods: biscuits, cakes, scones, and traybakes. To create a cake recipe, the model inputs and outputs would resemble the following:
Robinson then used Cloud AI Platform Notebooks to visualise the data and generate statistics. Thereafter, she leveraged TensorFlow's Keras API to find the ideal combination of hyperparameters and deployed the model using AI Platform Prediction.
According to Google, the deployed model generated a list of ingredient amounts and also predicted recipes for each of the distinct baked goods. The model was also able to generate hybrid recipes for biscuits and cakes. This led to the team creating a machine learning-generated cake batter on a machine learning-generated cookie.
Since the model outputs only included basic baking ingredients, the team had control over how to add Maltesers to the cake and biscuit recipes. It decided to chop and incorporate them into the batter, and three whole Maltesers were hidden between the cake and biscuit:
To top it off, the team also paired Marmite with a buttercream base and golden syrup for a frosting combination. Sam Chang, Mars Wrigley's global head of data science and advanced analytics, said the ease and speed of bringing this idea to life has already sparked multiple ideas around the endless possibilities of how AI can bring innovation to the kitchen by creating a foundation for recipe development.
This is not the first time Google has attempted to create baking recipes using AI. Last December, Robinson used a dataset of recipes she collected to build a TensorFlow model that took in lists of ingredients and churned out predictions such as "97% bread, 2% cake, 1% cookie". Her model came up with a new recipe back then, which was 50% cookie and 50% cake, also known as "cakie".
Separately last May, Grab also tapped on TensorFlow to turn any child's doodles of their favourite dishes into food orders delivered via GrabFood. This was part of its Grab Delivery Doodles campaign which rolled out in Malaysia, Vietnam and Indonesia. Grab Delivery Doodles was built using a machine learning model that can recognise drawings of local dishes such as "martabak", "bánh mì" and "nasi lemak". Volunteers from Grab, Google and their families submitted over 10,000 doodles to help teach the AI to recognise kids’ drawings.
MARKETING-INTERACTIVE's Content 360 Week is back from 6 to 8 April this year! Super charge your content production, distribution and monetisation strategies by learning from brands such as NBA Asia, P&G, Malaysia Airlines, and Marriott International, among others. Sign up today!