Biography

Daniel Polani is Professor of Artificial Intelligence at the School of Computer Science at the University of Hertfordshire, UK.  He is Associate Editor, amongst other, of the Journal of Autonomous Agents and Multi-Agent Systems and Advances in Complex Systems, and currently president-elect of the International RoboCup Federation, PI of the Horizon 2020 projects WiMUST and socSMCs and host of the Marie Curie Fellowship InterCoGaM. His research interests include the use of information-theoretic language to approach questions concerning the origin of life and intelligence, principled approaches for cognition, and for autonomous robotics.

Abstract

What is worth doing?

When we ask the question what an autonomous intelligence should or would do, too often this question is coloured by external requirements; explicit goals or performance measures. It is clear that this is not sufficient for truly autonomous intelligence. It is also clear that some internal goals of an agent may be acquired over the long run via evolutionary processes, but they cannot address all possible variations that an agent may be exposed to in its operative time. Therefore, for the question where goals or preferred behaviours come from if they are not explicitly imposed or derive from internalized behaviours, we need to look at the wider picture. It turns out that the interplay between agent and environment per se induces a significant preferment structure which, for instance, becomes visible through the eyes of Shannon information, the “universal currency” of decision making. The talk will highlight pathways for the emergence of intrinsic motivations and goals, based on principles including empowerment and goal-relevant information and discuss lessons learnt for the understanding of autonomy and, if time permits, will highlight the duality between the role of environment and that of the agent.