Artificial Intelligence Learns How Predictable Humans Are By Binge Watching

Watching videos for over 600 hours straight hardly pays off for humans, particularly once the headache and general degeneration of the body sets in. However, for an Artificial Intelligence created by MIT engineers, it resulted in the ability to predict human interactions. Is this a sign that the world is one step closer towards being ruled by AI overlords? As it turns out, the answer is no.

According to GeekWire, the AI is nowhere near as good as humans are when it comes to predicting what will happen in a given scene where the choices for the resulting actions involve shaking hands, hugging, high-fiving, or kissing. The AI was able to predict the actions correctly by 43 percent, which is low compared to the human counterpart’s score of 72 percent.

On the other hand, context is important in this situation and the fact of the matter is that the researchers didn’t expect the AI to get such a high score. They placed the success rate at around 25 percent. Based on that, the results of the experiment are considered better than expected, even with the researchers saying that they still have a long way to go.

As Wired reports, the amount of time that the AI has had to study behavior via YouTube clips is nowhere near the level of experience that the human counterpart has amassed. The lead author of the project, Carl Vondrick is more interested in how much further the program can go with time.

 “Even a toddler has much more life experience than this,” he said. “I’m interested to see how much the algorithms improve if we train it on years of videos.”

The TV programs used to predict behavioral patterns include “The Office,” “The Big Bang Theory,” and “Desperate Housewives.” The experiment actually involved still frames of the programs as well.