Chow-obsessed AI suggests recipes to match food photos

It's a good thing artificial intelligence systems can't get hangry. Researchers at MIT's Computer Science and Artificial Intelligence Laboratory have developed an AI system with a million recipes at its fingertips. The neural network, dubbed Pic2Recipe, suggests recipes based on looking at and analyzing photos of food. 

The recipes and images database it pulls from is called Recipe1M. The data comes from a variety of popular cooking websites, including Food.com and Kraft Recipes. The researchers, who have published a study on the AI's abilities, call it "the largest publicly available collection of recipe data."

"In computer vision, food is mostly neglected because we don't have the large-scale datasets needed to make predictions," co-author Yusuf Aytar said in a news release Thursday. "But seemingly useless photos on social media can actually provide valuable insight into health habits and dietary preferences."  That means your endless Instagram snaps of your meals could come in handy for scientists.

You can try out the AI system through an online demo. I uploaded a generic photo of a pizza covered in toppings to Pic2Recipe. It analyzed the photo and then suggested five possible recipes ranging from California bacon-ranch pizza to roasted-vegetable pizza. If you click on a recipe, it gives you the ingredients and instructions and a link to the source. 

Pic2Recipe is still in learning mode. The research team says the system found the correct recipe match about 65 percent of the time. It returns a "no matches" response when it doesn't recognize the dish in a photo. But the AI doesn't need to be perfect. When it gets it right, it's pretty accurate about finding appropriate recipes. Sometimes it's downright impressive, like when it returns a group of sugar cookie recipes after seeing a photo of a tray of fairly plain-looking cookies.

The researchers also shared some fun statistics from the recipe database. The average recipe contains nine ingredients and the top three most common ingredients are salt, butter and sugar.

"This could potentially help people figure out what's in their food when they don't have explicit nutritional information," says the lead author, MIT student Nick Hynes. It could also help you figure out how to make your favorite restaurant dishes at home. So keep snapping those smartphone photos of your meals. For science.

Virtual reality 101: CNET tells you everything you need to know about VR.

CNET en Español: Get all your tech news and reviews in Spanish.