by Angela Guess
Matt Burgess reports in Wired, “One difficulty faced by artificial intelligence is predicting what humans are going to do next. To help solve that problem, researched have trained an algorithm by making it binge-watch TV. Computer vision experts from MIT’s Computer Science and Artificial Intelligence Lab (CSAIL) made an algorithm watch 600 hours of TV shows including Ugly Betty, Scrubs, The Big Bang Theory, The Office (US) and more. In each of the clips, taken from YouTube, humans were performing tasks and interacting with each other. After analysing the videos, the AI was then made to watch a clip it hadn’t seen before and predict what would happen. The system was allowed to make multiple predictions about what might happen between one and five seconds in the future. ‘In some cases our model correctly predicts that a man and woman are about to kiss or hug or that men in a bar will high five,’ the researchers wrote.”
Burgess continues, “The computer model was able to predict what would happen next 43 per cent of the time. ‘We believe abundantly available unlabelled videos are an effective resource we can use to acquire knowledge about the world, which we can use to learn to anticipate future,’ the researchers wrote in their paper. And the ‘unlabelled’ nature of the videos is key: they didn’t have captions or descriptions explaining what happened so the AI had to work everything out by itself. To validate the results the researchers used labelled videos in the reported tests.”
Photo credit: Carl Vondrick, Hamed Pirsiavash, and Antonio Torralba