Let's say you see a dish that looks tasty – on your dinner plate, on a restaurant's website, wherever – and you want to know how to make it.
A new tool from an artificial intelligence lab might be able to help. But we have to emphasize might.
It's a neural network developed in the Computer Science and Artificial Intelligence Lab at MIT.
The idea is simple: you upload a picture of some food and the Pic2Recipe! page gives you the recipe.
Maybe. There's a good chance it'll give you several recipes – in order of confidence. Or it might tell you the picture turned up no matches.
There are some sample photos of food you can plug in If you want to try it out here.
When we tested a picture with lots of peas, for example, the network had 80 percent confidence it was "roasted edamame with garlic and olives." It had the same confidence it was fresh English pea salad, which is the correct answer. It had recipes for both of those dishes and other possible candidates.
MIT explains the science behind the neural network in this paper. Basically, they say, they're trying to train machines to understand food preparation by analyzing ingredient lists, cooking instructions, and food images that are already on the web.
The technology may not be sufficiently cooked, though. The Verge says it uploaded pics of ramen noodles, bags of potato chips, and rice with beans, all of which turned up "no matches."