An MIT-designed artificial intelligence has made significant progress in perceiving body language and accurately anticipating the type of interactions individuals will have based on its initial perception.
The neural network was set to binge-watch over 600 hours of shows, such as “The Office” and “Desperate Housewives” in order to record and perceive the characters’ different types of body language. Researchers made sure that the neural network thoroughly analyzed the information given to it.
The artificial intelligence was then presented with an entirely new set of videos which contained several characters. The neural network was then tasked to anticipate if the characters in each video would kiss, hug, high-five, or shake hands.
In order to give the AI time to think, each video was paused one second in the scene before the characters would play out their pre-recorded action.
Researchers are proud to state that their artificial intelligence neural network was able to anticipate the proper interaction correctly roughly about 43 percent of the time.
It is still apparently uncertain how much of that percentage is an actual artificial logical deduction and how much of it is guesswork. The statistic does dictate that if the neural network were to have guessed in each setting, its rate of success would have been closer to 25 percent.
When viewing the same videos, adult humans have on average predicted the correct action 71 percent of the time.
When a person predicts future actions based on current body language, the process is very dependent on personal social experience. In the given case, the human experience was accumulated in a wider array of time from a larger amount of sources.
In order to continue the research, the artificial intelligence would have to observe natural human interactions for considerably more time. Researchers do believe that the neural network could surpass the current average prediction rate if given enough time.
MIT believes that AIs able to quickly read body language could be valuable in areas such as field medicine. The artificial intelligence could do a quick read and discern what is medically wrong with individuals without them having to express it.
The same artificial intelligence prediction technology could be used in detecting subterfuge or malicious intent, making it a considerably useful but intrusive tool.
Image Courtesy of YouTube.