A group of researchers from the Leverhulme Center for the Future of Intelligence and the Center for the Study of Existential Risk at the University of Cambridge, led by researcher Alexa Hagerty, has developed a small online game that, through a website, try to “read” the emotions of the participants.
Artificial intelligence faces the challenge of interpreting human emotions through facial recognition
In addition to facial recognition to identify individuals, many technology companies want to go one step further and be able to achieve the identification of emotions through facial recognition. For that, you also have to train artificial intelligence (AI) and that is the reason why this simple online game has been developed.
The group of researchers from the University of Cambridge seeks to reinforce the ability of their AI in recognizing the facial patterns that make up the expressions associated with different emotions. For this they have created the web emojify.info, where they pose the challenge of show six different emotions (happiness, sadness, fear, surprise, disgust and anger) through facial gestures that the developed AI will try to detect, identify and associate with said emotions.
When accessing the web, permission is requested to use the user’s webcam and thus be able to analyze the participant’s face. It is ensured, that yes, that no information will be stored on the servers of the research group. Once this is done, different challenges are posed to the participant, who to achieve the facial recognition of these emotions sometimes you will have to exaggerate the expression.
All with the aim of training artificial intelligence, which tries, for example, to learn when a smile or a gesture of surprise is clearly faked, or the difference between a wink and a blink, something that is not easy for an AI to identify, as it is difficult to discriminate a mere reflex action of a voluntary act and endowed with some specific meaning (complicity, seduction …).
In other words, the AI is not yet performing perfectly … it must continue to learn. This video explains how the experiment works: