Science

A.I. Learns to Gaze into the Future By Watching Four Hours of Cooking Videos

by Danny Paez

Ripping the bong and binging cooking videos is nothing but a pass time for us meatbags. But to a sophisticated new artificial intelligence system developed in Germany, four hours of cooking videos is sufficient training for it to learn how to tell the future.

A team of computer scientists from the University of Bonn in Germany have developed the foundations of what could be the perfect android butler or robotic kitchen assistant. After showing the A.I. four hours of people making salads, the system was able to accurately predict the actions that would occur in a new situation several minutes in the future. The team will be presenting their results at the Conference on Computer Vision and Pattern Recognition in Utah, beginning on June 19.

“We want to predict the timing and duration of activities - minutes or even hours before they happen”, said Professor Jürgen Gall in a press release. The goal is a new generation of automated sooth-sayers who can use real world data to make accurate predictions about the future.

This is Prof. Jürgen Gall (right) and Yazan Abu Farha from the Institute of Computer Science at the University of Bonn.

Barbara Frommann/Uni Bonn

While gazing hours into the future is still a long way out, the A.I. was still capable of some pretty impressive foresight. When it was shown other salad-making videos it had never seen before, for example, the AI was able to anticipate what would happen next 40 percent of the time for short forecasts. It was even able to guess activities that were three minutes in the future 15 percent of the time.

Some other assorted hurdles remain. Gall and Farha’s system was only able to pull off this kind of accuracy when it is prompted with what happens in the beginning of the video. The human brain has an intrinsic skill to predict what’s next and fill in gaps when information isn’t present. But computers are still almost entirely new to the art of prediction.

Nonetheless, this a major step in the development of self-learning software that can estimate the timing and duration of what future activities. There is still work to be done, but with more time and research, breakthroughs like this could lead to a robot sous chef passing crème fraîche to the likes of Gordon Ramsey before he has a chance to yell at it.

Related Tags