If you’re wondering how good the camera will be on the next phone you buy (and the one you’ve got now is probably already pretty good), you’d be wise to pay attention to what the manufacturer has to say about artificial intelligence. Beyond the hype and bravado, technology has enabled amazing advances in photography in the last couple of years, and there’s no reason to think progress will slow down.
There is still a lot to innovate, without a doubt. But the most impressive recent advances in photography have taken place at the software level rather than at the camera sensor or lens itself, and that’s largely because artificial intelligence gives cameras a better understanding of what they are “looking at”.
Smarter cameras
Google Photos provided a clear demonstration of how powerful a combination of artificial intelligence and photography would be when the app launched in 2015. Prior to that, the search engine had been using machine learning to categorize images on Google+ for years, but the launch of its Photos app included consumer-facing AI features that would have been unimaginable to most users. Users’ disorganized libraries of thousands of untagged photos morphed into huge searchable databases overnight.
Suddenly, Google knew what your cat looked like
Google built on previous work from an acquisition in 2013, when it bought DNNresearch, and set up a deep neural network trained on data that had been labeled by humans. This is called supervised learning; the process involves training the artificial intelligence based on millions of images so that it can look for visual clues at the pixel level to help identify the category. Over time, the algorithm gets better and better at recognizing, say, a panda, because it contains the patterns used to correctly identify pandas.
What happens is that the camera learns where the black hair and the white hair tend to be in relation to each other, and how it differs from that of a cow, for example. With further training, it is possible to look up more abstract terms like “animal” or “breakfast,” which may not have common visual cues but are immediately obvious to humans, and the AI also recognizes because it has “learned” it.
teach the algorithm
It takes a lot of time and processing power to train an algorithm like this, but once the data centers have done their thing, it can be run on low-power mobile devices without much trouble. The heavy lifting has already been done, so once your photos are uploaded to the cloud, Google can use its model to analyze and tag your entire phone library. About a year after the launch of Google Photos, Apple announced something similar.
