We’ve seen a lot of impressive AI in recent times, from Microsoft’s achievement of human speech parity to chatbots and digital assistants. However, the latest Google AI project offers a glimpse into the how machine intelligence may eventually replace humans.
Taking screenshots from Google Street View, the company’s AI has been cropping and adjusting images to create professional looking content. Alphabet Inc. calls the deep learning system Creatism.
According to Google, the advancement was achieved by breaking aesthetics down into multiple aspects, letting Creatism learn each individually from a set of professional examples. Each can then be optimized efficiently, with a “dramatic mask” tool to make lighting more striking.
Google has published a full 1:1 comparison slideshow with all images created by Google Creatism and the original scenes from Google Street view on GitHub. In total there are 256 pictures with impressive landscape-scenes from all over the globe.
Turing Test-like Experiment
What’s more, the AI doesn’t even require before and after images for its training dataset. Google simply feeds in pictures and it picks out the elements itself. It goes further than that, however, also analyzing things like composition. So well in fact, that professional photographers are struggling to tell the difference. The paper explains:
“Using our system, we mimic the workflow of
a landscape photographer, from framing for the
best composition to carrying out various postprocessing
operations. The environment for our
virtual photographer is simulated by a collection of
panorama images from Google Street View. We design
a “Turing-test”-like experiment to objectively
measure quality of its creations, where professional
photographers rate a mixture of photographs from
different sources blindly. Experiments show that a
portion of our robot’s creation can be confused with
The incredible difference in these images really speak for themselves, but Google did test how exactly how passable its AI is. Showing these photos and a bunch of random ones to professional photographers, it found 40% were seen as semi-pro or higher photographers. For an AI in its infancy, that’s really impressive.
It’s also quite easy to see where this tech could go. An editing software that suggests all of the best edits for you, or a button in social media to automatically adjust your images. Solutions like that exist already, but rather than a simple filter or contrast adjustments, Google’s could intelligently edit different parts of the photo.
My guess is that we’ll see this in Google Photos and Snapseed as soon as it’s ready.