MIT Uses AI to Figure Out Recipes Based on Pictures of Food

As iPhone app developers already know, the researchers and engineers at MIT are responsible for some of the most exciting advancements in tech. Since its founding, the institute has had a reputation as one of the most elite schools in the country; it’s a gathering place for geniuses. We have written in recent months about some of the cutting-edge work the school is doing with drones, including a disaster response UAV that can stay in the air for five days at a time and a new autonomous camera drone system designed to pull off difficult shots. The school’s Computer Science and Artificial Intelligence Laboratory (CSAIL), the same team responsible for the aerial videography drones, has a new project its cooking up that uses AI for novel ends.

This new project is called Pic2Recipe, and it harnesses the power of AI to help identify recipes based on pictures of the food in question. Pic2Recipe uses AI to sort through Recipe 1M, a neural network the team compiled. Recipe 1M contains one million recipes with corresponding pictures. The neural network is able to make sophisticated distinctions between the pictures and recipes, enough so that the system can provide a ranked list of relatively accurate recipes based on the picture. Every iPhone app developer has Instagrammed the perfect meal; now there’s a tool that can take that photo and teach you how to recreate that meal.

Of course, HBO’s Silicon Valley showcased/mocked a similar app called SeeFood, and Pinterest has put together its own “dish recognition” app, which makes this use of machine learning something of a mini-trend. In an era where a guy like Elon Musk is warning the government about the apocalyptic potential of AI, it is useful to remember that at this point the technology is being used on a lot of practical (and sometimes goofy) projects like Pic2Recipe. That’s not to belittle the potential of AI. Boston iPhone app developers are well aware of both the exciting possibilities of machine learning and its scarier applications. But we can all agree that taking a picture of your lunch and having that picture translated into a recipe is pretty cool.