h i g h l i g h t s • Reinforcement learning's option switches are analogous to psychological insight. • Insight and options reveal comparable capabilities for transformational creativity. • Open problems remain: lifelong learning, switching when exploring, option discovery.

The increasingly rich and diverse literature on creativity has its core in psychology, but spans the cognitive sciences from artificial intelligence to philosophy and borrows from the wider humanities. Perhaps because of this immense breadth, there remains considerable disagreement with respect to the identity of the object of research. How to define creativity?According to the "standard definition," creativity consists of "effectiveness and originality." This definition is (relatively) consensual and therefore appears to capture something common to academic concepts of creativity. I conduct a conceptual analysis of the definition; thereby, I isolate and describe two ambiguities. Firstly, the definition leaves open the choice of the context and norms against which to measure originality and effectiveness. Secondly, it does not discuss the possible role of a subjective judge.My goal is not to propose yet another model of creativity, but to clearly identify the possible meanings of the word creativity in academic research. The existence of different interpretations does not necessarily reflect a fundamental disagreement about reality, but rather a failure to achieve consensus on a shared technical language. Therefore, simply recognizing and acknowledging the competition between diverse interpretations can form the basis for successful communication and for a complementary division of labor; it could improve the viability of interdisciplinary collaborations and prevent unnecessary fragmentation of the field.

No abstract

Urban Search And Rescue (USAR) robots are used to find and save victims in the wake of disasters such as earthquakes or terrorist attacks. The operators of these robots are affected by high cognitive load; this hinders effective robot usage. This paper presents a cognitive task load model for real-time monitoring and, subsequently, balancing of workload on three factors that affect operator performance and mental effort: time occupied, level of information processing, and number of task switches. To test an implementation of the model, five participants drove a shape-shifting USAR robot, accumulating over 16 hours of driving time in the course of 485 USAR missions with varying objectives and difficulty. An accuracy of 69% was obtained for discrimination between low and high cognitive load; higher accuracy was measured for discrimination between extreme cognitive loads. This demonstrates that such a model can contribute, in a non-invasive manner, to estimating an operator's cognitive state. Several ways to further improve accuracy are discussed, based on additional experimental results.