Barandiaran, X. & Di Paolo, E. .
Department of Logic and Philosophy of Science, UPV-EHU University of the Basque Country, Donostia - San Sebastian, Spain
Recent trends in cognitive science have seriously undermined the notion of representation (symbolic or sub-symbolic) as a building block for theory construction and modeling in cognitive science (Stewart et al. 2011). The current lack of a clear theoretical building block for dynamical, embodied and situated cognitive approaches calls for a re-appraisal of the notion of habit as developed by early pragmatists (James, Pierce and Dewey) and continental psychologists and philosophers (Köhler, Goldstein, Merleau-Ponty). Whereas contemporary computational neuroscience and machine learning approaches (Daw et al. 2005) still sustain a behaviorist (S-R probabilistic association) conception of habit, we propose a richer notion by modeling habits as self-sustaining behavioural neurodynamic patterns where activity-dependent plasticity shows an extended temporal structure. We illustrate this point with some work on evolutionary robotics, implementing a combination of Hebbian and homeostatic plasticity (as recently supported by different neurobiological studies--Turrigiano 2000, 2007). In our robotic models, this mechanism is capable to generate sensorimotor development, reinforcement learning, spontaneous habit formation and re-habituation to sensorimotor disruptions (Di Paolo 2000, Barandiaran and Di Paolo 2010). We conclude that a richer notion of habit can significantly contribute to the foundations of cognitive science, opening up the possibility to model poorly understood psychological phenomena.