10 May 2019
TITLE: Informational Routes to Intelligent Cognition
DATE/TIME: Friday, 10th May 2019
LOCATION: Bayes Centre
SPEAKER: Daniel Polani, Professor of Artificial Intelligence at the Department of Computer Science, University of Hertfordshire
The recent success of learning methods to successfully address hitherto unsolved problems has obscured several fundamental issues concerning the nature of intelligence. First of all, it is not always clear what properties of the world or the learning methodologies the success is owed to. Second, it does not answer the question how intelligent managed to emerge in evolution. And third, despite displaying formidable fine-tuned expert performance in selected, well-defined tasks, this does not provide satisfactory insight into where the abilities of generalization, robustness and context-switching derive from that organisms regularly display, at the price of being less specifically optimized. In short, typical AI solutions are often specialists, whereas organisms are typically generalists.
Under the assumption that the ultimate evolutionary goal of cognition is decision-making to ensure viability, and that decision-making requires information, one can argue that any cognitive processing needs to concern itself with the appropriate processing of the "right" information. If so, any intelligent decision-making implies the presence particular preferred information-processing architectures, which underlie any viable intelligent behaviour. This gives us insight on constraints on intelligent architectures which, at the same time, may turn out to be an asset in understanding strategies to tune out irrelevant details which is essential for generalism.
Daniel Polani is Professor of Artificial Intelligence at the Department of Computer Science. His main interests concentrate on understanding how information forms the basis for the emergence and evolution of cognitive capabilities. Founder of the SEPIA (Sensor Evolution, Processing, Information and Actuation) lab at the University of Hertfordshire, his activities in this direction includes methods for the cognitive methods for adaptive and self-organizing complex systems, intrinsic motivation, and the control of collective robot/agent systems. He is co-founder of the 18-monthly GSO (Guided Self-Organization) workshop series, and has organized of the FP7 TRUCE funded IDeM (Information in Decision-Making) workshop in 2015. Associate Editor of journals such as the Journal of Autonomous Agents and Multi-Agent Systems, Frontiers in Computational Intelligence, and Advances in Complex Systems. Currently, he is president of the RoboCup Federation (until July 2019).