AI and Emotion

A team of researchers is working on a way to use artificial intelligence (AI) to decode human emotions. In a recently published paper, they reported that their neural network – dubbed EmoNet – accurately and consistently categorized eleven different emotions.

EmoNet was “trained” on an existing neural network called AlexNet. In its original form, AlexNet allows computers to recognize objects, but the network was further refined to also predict what emotions would be evoked by certain images. These images were then fed into EmoNet, which was instructed to categorize them by emotion. The results indicated that craving or sexual desire was accurately identified more than 95 percent of the time, but the system was less accurate when presented with more nuanced emotions such as confusion or surprise. Simple colors also registered emotions: black elicited anxiety and red invoked craving. And when tested on categorizing movie clips as romantic comedies, action or horror films, EmoNet was correct 75 percent of the time.

Eventually, rather than simply asking a person how he or she feels, AI may be able to more directly measure emotion-related brain processes and even eliminate some of the labels associated with mental health. Less subjective labels could even lead to better treatments and interventions.