Researchers at the Massachusetts Institute of Technology have developed a new computer vision algorithm that they say is capable of creating short animations simply by looking at a still picture and predicting what will happen next.

Computer vision is essentially a way for computers to understand images using artificial intelligence. But while computers are already becoming experts at recognizing objects in a scene, they are not very good at making assumptions based on that knowledge. For example, if you were to look at a picture of a wave at the beach, you could predict roughly what the water will look like when the wave crashes into the sand, but this sort of assumption is extremely difficult for a computer.

“Understanding object motions and scene dynamics is a core problem in computer vision,” the MIT research team wrote in an academic paper titled “Generating Videos with Scene Dynamics.”

The team, led by Ph.D. student Carl Vondrick, developed a deep learning algorithm that taught an AI to predict movements in a scene while looking only at a still image. To accomplish this, the team fed the AI data from millions of videos so that it could learn the types of movements things like trains or water frequently make.

The researchers then used a deep learning method called adversarial learning, where two different neural networks are pitted against one another. One of these neural nets, the “generator,” took the knowledge gained from the video data and created extremely short synthetic videos from a series of images.

Meanwhile, the other neural net, the “discriminator,” had to tell if these videos were genuine or not. Essentially, the goal was to get the generator to successfully dupe discriminator into thinking its videos were the real deal.

The researchers noted that most of the videos were easily identified as synthetic, but overall, they believe that the results of their methods are promising.

“These videos show us what computers think can happen in a scene,” Vondrick said in a statement (via PC Magazine). “If you can predict the future, you must have understood something about the present.”

You can watch a video outlining MIT’s new computer vision technique here.

Image courtesy of MIT