Latest News

# #

Posted on 20 September 2016

Thomas publishes article : Hierarchical Reinforcement Learning and Creativity

CogNovo research fellow Thomas R. Colin published in the special issue on creativity of the journal "Robotics and Autonomous Systems". The article, titled "Hierarchical reinforcement learning (HRL) as Creative Problem Solving", discusses creativity from the perspectives of cognitive psychology and machine learning, in the context of problem solving. Particular attention is given to the phenomenon of "insight" (also called the "Aha!"-moment). Thomas and his co-authors show that many of the characteristics of insight problem-solving behaviour are exhibited by HRL systems; moreover, the similarities are relevant with respect to creativity. However, important challenges remain to be met in order to call a robot "creative" or "insightful". The article goes on to present these issues, which continue to be the focus of Thomas’ work.

Thomas is working on the CogNovo project #11 "Developing creativity in cognitive robots" which aims at building robots capable of insight. In his approach the combination of learning and searching, and the automatic development of a hierarchy of policies through interaction with the world, are enabling robots to solve insight problems in a manner consistent with the bulk of the psychological literature on insight. As part of his PhD, Thomas is currently doing a secondment at the AI-Lab at Aldebaran Robotics (Paris, France).