Google cooking up AI to count calories in your food photos
A Google research scientist spills the beans on an artificial-intelligence project that tries to calculate the calories in your food porn pictures -- and learns from its mistakes.
CNET freelancer Danny Gallagher has contributed to Cracked.com, Mental Floss, Maxim, Break.com, Mandatory, Jackbox Games, Geeks Who Drink and many, many other publications in his never-ending quest to bring the world's productivity to a screeching halt. He lives and works in Dallas. Email Danny.
For some reason, Instagram has become the social-media app people use when they want to show the world some massive meal they're about to shove into their stomachs. How did this become a thing? Is the entire world competing in some contest to rack up the highest calorie count before choking to death on a Bloomin' Onion?
It seems like we could do better things with Instagram than share images of our peanut butter and jelly sandwiches with everyone, but a new artificial-intelligence project reportedly being developed by Google may (and I never thought I would say this) actually make Instagram and its food porn collection useful.
The AI project, unveiled at the Rework Deep Learning Summit in Boston last week, is called Im2Calories and aims to be capable of counting the number of calories in every item on a person's plate simply by examining a photo, according to Popular Science May 29.
Kevin P. Murphy, a researcher who works with Google's research lab and specializes in artificial intelligence and algorithms and theory, said the new AI uses "the depth of each pixel in an image" and "sophisticated deep-learning algorithms" to identify food, judge its size and come up with a calorie count. Murphy also said the photo of the food doesn't need to be high-definition for it to work, according to the Popular Science article.
Murphy said during his presentation that the AI may not get the calorie content correct in the first few tries but it will improve the more people use it and share the results.
Obviously, this new technology could be used to simplify food tracking and improve food diary apps and other nutrition software. However, Murphy noted at the presentation that his big hope is that the technology could be applied to more than just keeping us honest with our Weight Watchers coaches or deterring us from supersizing. He suggested the same AI could be applied in other ways, such as examining traffic patterns to help drivers find a parking spot.
Google spokesman Jason Freidenfelds told CNET that Im2Calories and the algorithms that run it are still at the research stage and that there are "no actual product plans at this stage." So don't get your hopes up yet: "We don't have more to say at this point on whether or how this might make its way into some future product(s), because we'd really just be speculating," Freidenfelds said.
However, can we really be sure we actually talked to a Google spokesperson and not just some supersmart form of AI that can learn how to answer press requests through repetition? We might already be through the looking glass, people.