A metaphysics for predictive processing

For predictive processing, perception is tied to the upshot of probabilistic inference, which makes perception internal, affording only indirect access to the world external to the perceiver. The metaphysical implications of predictive processing however remain unresolved, which is a significant gap given the major influence of this framework across philosophy and other fields of research. Here, I present what I believe is a consistent metaphysical package of commitments for predictive processing. My starting point is a suitable challenge to predictive processing presented by Tobias Schlicht, who argues that the framework is committed to Kantian transcendental idealism, and marshals several lines of argument that this commitment undermines predictive processing’s aspirations to completeness, realism, and naturalism. I first trace Hermann von Helmholtz’s nuanced reaction to Kant, which sets out the preconditions for perception in a manner prescient of the notion of self-evidencing central to contemporary predictive processing. This position enables a fundamental structural realism, rather than idealism, which blocks Schlicht’s line of argument, allowing plausible versions of completeness, realism and naturalism. Schlicht’s challenge is nevertheless valuable because addressing it, in the specific context of Helmholtz’s response to Kant, helps bring to light the compelling structural realism at the heart of self-evidencing.