Rationality

In this talk, I compare the trade-offs between "rational decisions" that are computationally expensive and have potentially large resource requirements attached to them and affective decisions that are computationally cheaper, require fewer resources, but are limited in their
"rationality". I argue that notions of rationality that are computationally intractable are of mere theoretical interest. For an understanding of how agents "work" in the real world (e.g., how evolution produced control architectures of animals, or how we have to program intelligent robots to make decisions), we need a notion of "rationality" that takes the severe resource limitations (time, computation, etc.) of real agents into account. I will demonstrate that affective control (which is sometimes construed as "non-rational") can be very effective in decision making if "relative performance" matters (i.e., performance relative to the cost involved in making decisions, where costs comprise energy expenditure, resource requirements, computation time, etc.). The talk will end with a set of open problems regarding the interplay of affective decision making and reasoning that I believe to be of great importance for understanding the human mind and for the design of future artificial intelligent systems.