Minimal phenomenal experience and the synthetic data hypothesis

Minimal Phenomenal Experience (MPE) refers to the simplest form of conscious experience, characterized by the absence of time, self, and sensory content, where only pure awareness or consciousness remains. In this paper, we present a computational neurophenomenological account of minimal phenomenal experience within the active inference framework. We propose a generative mechanism that allows active inference agents to leverage capacities akin to those in generative adversarial networks, which use a dual framework of data generation and discrimination to improve perceptual accuracy. Our hypothesis posits that the brain continuously generates synthetic data to fill perceptual and cognitive gaps, and we provide a computational mechanism within the active inference framework to support this. We explore how synthetic data generation, when amplified by meditation, psychedelics, or other altered states, can the brain to ‘defabricate’ entrenched conceptual structures and approach pure consciousness at the limit. We provide simulations of an active inference agent embedded in a virtual environment, we demonstrate that recursive loops of synthetic data generation can lead to a breakdown in conceptualization — 'seeing as' — while enhancing the system’s ability to predict its own observation stream. We argue that this model can characterize the essential features of MPE, including contentlessness, epistemic openness, and the unique valence profile associated with minimal phenomenal experiences.