Warning: The robots that power ever-popular review platform Yelp are becoming sentient. Okay, not really, but as Fast Company reports, the company is using artificial intelligence to better serve its millions of users.
For an entire generation of diners, taking photos of their food has become second-nature — and thanks to them, Yelp has a gigantic database of photos. While previously the company has relied on users to caption their own photos with "search-friendly metadata," it's now armed with software intelligence that can identify information about a restaurant based on photos alone.
As FastCo explains, this is thanks to deep learning, "a form of machine learning that involves training neural networks to solve problems using large sets of data." (Basically, machines are getting smarter by mimicking the structure of the human brain.) In Yelp's case, its software is using image analysis techniques to identify color, texture, and shape, meaning it can recognize the presence of say, burritos, or whether a restaurant has outdoor seating.
At this point, "the company is now able to predict attributes like 'good for kids' and 'ambiance is classy' with 83% accuracy" based solely on photos (arguably a more reliable source of information than user-submitted reviews, which can often be terribly biased or just factually incorrect). Soon, Yelp will be able to use this information "to auto-caption images, improve search recommendations, and better select an assortment of images to feature on businesses’ listing pages."
Yelp is just one of countless companies currently caught up in the AI arms race. One of Google's many AI projects involves using deep learning to calculate the caloric content of food from photos, while a number of companies — including Facebook — have utilized AI to make restaurant recommendations. In London, a company called IntelligentX is currently using Facebook chat bots, user-submitted feedback, and artificial intelligence to brew better beer.
• Your Photo of a Burrito Is Now Worth a Thousand Words [FastCo]