Machine Empathy: Digitizing Human Emotions

Abstract

The primary objective of this work is to emulate machine empathy through digitizing human emotions. A simple proofof-concept experiment is conducted, where a brain-computer interface (BCI) captures the brain's electroencephalogram (EEG) signals using an Emotiv Epoc headset. A two dimensional (2D) intensity (heat) map of the brain's EEG is obtained for a pre-defined set of an emotional stimulus, namely excitement and stress. An artificial neural network (ANN) is subsequently used for classifying the 2D image. The key contribution of this work is to leverage the already powerful and mature tools for image recognition developed in ANN systems for emotion recognition through adapting the 2D intensity map of the EEG brain activity. The resulting BCI system was set-up to control a surrogate humanoid robot, allowing the robot to emulate empathy and interact with the subject according to pre-defined behavioural models. The ANN classifier exhibited an accuracy of 87.5% for recognizing two of the emotional states targeted in this study.