Your eyes take the first bite of every meal. It’s the first judgment every diner makes. Which is why restaurants care so much for the presentation of a meal. Even the chain restaurants must present their food consistently and appetizingly. Naturally, we’re not far from restaurants using AI to assist in this department. One of the first, as always, to take this leap is Domino’s, using AI for pizza quality control.
Domino’s Quality Control
Back in February of 2019, Domino’s unveiled the latest in a long line of technological PR stunts: Points for Pies.
Simply download the Domino’s app and use it to take a picture of a pizza, any pizza. For every picture the app identifies as pizza—Domino’s or otherwise—you get ten points (with a limit of ten points per week). Once you get 60 points, you’ll be entitled to one free medium two-topping pizza from Domino’s.
Domino’s AI Marketing
Points for Pies is essentially building a database of what finished pizzas look like. Undoubtedly, there’s going to be massive variability in user-generated photos, which is a huge advantage for building out image recognition algorithms.
At the time, it seemed like a valuable marketing program while building a great database of pizza images. The application of this database, though, was unclear.
What I overlooked was how this specific image dataset could be used to train an algorithm for quality control purposes.
The new DOM Pizza Checker [only available in Australia and New Zealand] uses a ceiling-mounted camera and artificial intelligence to make sure that every pizza is up to standards. Domino’s claims that it takes a picture of the pizza, recognizes the type, analyzes the distribution of toppings and cheese spread, and grades it.
Jesus Diaz, tom’s guide
Dragontail Systems is the company behind DOM and they have ideas for an entire connected restaurant of the future. This initial vertical for AI quality control in pizza is well-complemented by the massive database of pizza images Domino’s collects through Points for Pies.
So what happens if the pizza doesn’t meet DOM’s quality standards?
For the time being, they’ll remake the pizza by hand. However, there is room for some collaboration with another algorithm out there.
AI Builds a Pizza
Researchers out of MIT created PizzaGAN, which recognizes the present toppings on a pizza and can generate an image of a new pizza with toppings added or deleted. It’s a generative adversarial network (the same type of algorithm used to create deepfakes), that operates by pitting two algorithms against one another until the desired pizza is a passable image creation.
(For more on how the algorithm works, head here).
If PizzaGAN can recognize toppings, then ideally, paired with the quality control AI, this AI could help guide a robot to put the finishing touches on a pizza – put it back in the oven if need be, add any toppings where they’re missing, etc…
In unison with the quality control AI, they would be a powerful duo. All they’d need is a precise robotic arm and the entire back end of the pizza line could be taken care of.
In theory, it’s a brilliant combo that could be applied to all types of chain restaurants. Not just pizza. In practice, we’re many years away from this future.
Nonetheless, AI in the culinary space is something I’m particularly interested in seeing grow.