Our (with Christopher Buckley and Jelle Bruineberg) manuscript was recently accepted for the Artificial Life conference (Alife 2020) in Montreal. The organisers let us know that this event will be 100% online, due to Covid-19.
- Predictions in the eye of the beholder: an active inference implementation of the Watt governor, Baltieri M., Buckley C. L. and Bruineberg, J., Accepted at the conference on Artificial Life, Montreal, Canada, 2020
(Preprint to be uploaded soon)
The Watt governor is a paradigmatic system presented in the literature of “anti-representational” models of cognitive architectures. Historically, the Watt governor was proposed as a system constituing an alternative to the computer metaphor of traditional cognitive science. Active inference is a modern process theory of cognitive agents inspired by principles of approximate Bayesian inference that attempts to describe different aspects of perception, agency, learning and other functions under the mandate of free energy (variational and expected) minimisation. Often, active inference has been identified as a “representational” theory and in doing so, many of its propoments have implicitly assumed a metaphysical role for its main constructs, i.e., precisions, prediction errors, predictions, etc.. Here we studied the prototypical case of the Watt governor using this theory, discussinng its possible uses and advantages while highlighting, crucially, an epistemic role for active inference that goes against the common idea of automatically granting ontological status to statistical (i.e., observer-dependent) properties of a system.