A new artificial intelligence system can generate short videos and take still images which simulate what will happen next. This system is almost the same as how people can think about the scene that will happen next, based on the new study. People know how the world works intuitively, that makes it easier for human being because it is different to machine in order to envision how to play out the scene. But a couple of objects in a still image can interact and move in a couple of different ways which is making it very difficult for machines in order to reach this feat. But there is a deep-learning system that can be able to trick many people 20 per cent of the time if compared to the real footage.
Massachusetts Institute of Technology or MIT researchers come with two neural networks against each other with one trying to create a couple of videos which were realistic enough to trick the first system and the other trying to distinguish real videos from machine-generated ones.
This type of setup is called a generative adversarial network or GAN and the systems come with the competition that has the results in realistic videos increasingly. When the workers have been asked by the researchers on the Mechanical Turk of Amazon to choose which videos that were real, the machine-generated videos have been chosen over the genuine videos 20 percent.
For those of you who have a job as the film directors, you do not have to consider about machines that will take over your job because the videos were only 1 up to 1.5 seconds and were created at the 64 x 64 pixels resolution and the dynamic environments can be navigated by self-driving cars and humans can be interacted or you can tag your videos to your Facebook account automatically with labels that describe the condition and situation.
A realistic video can be generated with their algorithm of what will happen next in the future that shows that at some level it understands what the future is all about. They will encourage development in explaining that the computer scientists can get machines with a couple of advanced situational. This system can also learn unsupervised. It means that there are more than 1.5 millions videos. They train the system so that it will not be able to be labeled by human.