Reversal Learning Based on Somatic Markers

One of the main aspects in the field of Artificial Intelligence is the creation of agents with the ability to learn like human beings do. Based on made experiences humans are able to adapt their behaviour in order to solve tasks. Another important aspect of human decision making is the ability to discard learned behaviour when the usual decisions, concerning a stimulus, lead to a bad outcome. For robots intended to be embedded in a social environment, the adaptability of behaviour is an important factor. Research of human decision behaviour shows, that emotions play a decisive role, even for learning and reversal learning. In this paper, improvements and further results of a previously presented framework for decision making based on an emotional memory are presented. The improvements include the reduction of the amount of previous knowledge that has to be implemented and an evaluation concerning reversal learning. For evaluation purposes, a typical reversal learning task, performed by real subjects, has been used. The results show that this framework allows the adaption of behaviour comparable to human subjects and offers decisive improvements, which lead to better results in reversal learning tasks without the need to directly declare a task as such one.