MIT is shooting for the stars by training an AI to predict our future. Their bold mission is even more surprising considering their starting point: TV shows #machinemagic
Researchers from MIT created an AI algorithm that analyzes video to tell what’s gonna happen next. They started with human interactions in TV shows, considering there is more likely to observe the whole range of emotions. After 600 hours of watching series like The Office and Big Bang Theory, AI was put to the test – what did he learn?
Its makers gave it several previously unseen pieces of footage and made it predict what would happen next. In 43% of the time, the AI got it right. By comparison, people made an accurate prediction 71% of the time. It’s a pretty big difference, but let’s consider the experience of the AI. As Carl Vondrick, the project’s lead author, put it “even a toddler has much more life experience than this. I’m interested to see how much the algorithms improve if we train it on years of videos.”
His hope is that, some day, this technology will be able to tell when a person is in danger and, if embedded in robots, even make them rescuers.