Why should any body have a self?
We use a general computational framework for brain function to develop a theory of the self. The theory is that the self is an inferred model of endogenous, deeply hidden causes of behavior. The general framework for brain function on which we base this theory is that the brain is fundamentally an organ for prediction error minimization.There are three related parts to this project. In the first part (Sections 2-3), we explain how prediction error minimization must lead to the inference of a network of deeply hidden endogenous causes. The key concept here is that prediction error minimization in the long term approximates hierarchical Bayesian inference, where the hierarchy is critical to understand the place of the self, and the body, in the world.In the second part (Sections 4-5), we discuss why such a set of hidden endogenous causes should qualify as a self. We show how a comprehensive prediction error minimization account can accommodate key characteristics of the self. It turns out that, though the modelled endogenous causes are just some among other inferred causes of sensory input, the model is special in being, in a special sense, a model of itself.The third part (Sections 6-7) identifies a threat from such self-modelling: how can a self-model be accurate if it represents itself? We propose that we learn to be who we are through a positive feedback loop: from infancy onward, humans apply agent-models to understand what other agents are up to in their environment, and actively align themselves with those models. Accurate self-models arise and are sustained as a natural consequence of humans’ skill in modeling and interacting with each other. The concluding section situates this inferentialist yet realist theory of the self with respect to narrative conceptions of the self.