New AI From MIT Can Tell You How To Make A Pizza Just By Looking At It

Cooking needs skill, practice, time, and patience. So can a machine do what takes years for professional human chefs to master? Well, research from MIT suggests that it might be possible.

MIT researchers have trained a new artificial intelligence on reversing engineer pizza. Basically, they have taught an AI to observe a pizza photo and learn the toppings as well as the order where they are placed.

Although it may sound whimsical at first, eventually it will result in a system that would be capable of processing any food photo and produce a recipe based on the photo. This is an exciting example proving the potential of using AI in the food sector.

The pizza AI, which is called PizzaGAN, requires the use of about 5,500 images of pizzas in clipart-style. Each of the pizza needs to have a maximum of 10 toppings. After that, the researchers fed the AI with 9,213 real pizza images which they took from Instagram.

With this huge dataset, they could train the system to take a look at a pizza, determine the kind of toppings and how they were placed on the pizza. The pizza AI could generate an image which demonstrates how the pizza would look if a specific topping is added in or taken out. Moreover, the AI can even produce an image of how the pizza would look like before or after it is cooked.

By looking at its name, it is expected that PizzaGAN can only work with one type of food at the moment.

But according to the paper from the MIT team, the AI system’s capability in identifying and manipulating layers proves that it could eventually be used with other types of foods like salads or burgers. Or perhaps, the AI could take a turn to the fashion industry, where it can combine different layers of clothing to create a complete outfit by enabling a digital shopping assistant.