Even though artificial intelligence has not become common enough to be implemented on a wide scale in the day to day life, research makes a lot of progress in this direction. The Artificial Intelligence and Computer Science Laboratory of the Massachusetts Institute of Technology created an AI system that can predict human reactions.
The team of researchers from MIT created an algorithm that analyzes audiovisual images, then translates those images into knowledge in order to predict human responses.
Over six hundred hours of television series, such as “Big Bang Theory” or “The Office”, were used by scientists to teach the AI system about the human behavior. The AI learned to identify handshakes, hugs, high-fives and kisses. After learning each type of behavior, the researchers tried to teach the AI system to anticipate them depending on the context.
However, it would be important to mention that this particular AI had a different learning mechanism than humans. Instead of understanding the context, the AI analyzed the movement and composition of pixels in order to create recognizable patterns.
To complete the experiment, after creating the algorithm and feeding it six hundred hours of TV series, the researchers showed the AI a single frame from a video he had not seen. Given the assignment to predict what would happen next, the algorithm replied correctly in almost 43 percent of the cases.
It may not seem much. However, all the previous technologies similar to this one could not do more than 36 percent.
Furthermore, even though 600 hours may sound like a lot, in fact, according to experts, even a toddler has more experience than this. Hamed Pirsiavash, one of the researchers who worked on the project (now an assistant professor at the University of Maryland), mentioned that he is very curious to see how the algorithm can evolve after getting more experience.
According to him, by the age of 10, every human being has more than 60,000 hours of direct observation of the environment or unmediated interaction with the surrounding world.
Compared to the 43 percent accuracy that the AI managed to score in predicting human reactions, humans average 71 percent. This difference can be easily explained if we take into account the life experience of a person, which can never be matched by a 600 hours TV series marathon.
Image Source: Wikipedia