Minimal phenomenal experience (MPE) is characterized as a state of pure awareness stripped of conceptual content, and serves as a methodological tool for probing the mechanisms underlying consciousness. We propose a computational architecture in which a policy model that generates behavior is recursively coupled to a program model that synthesizes structured, executable explanations of that behavior. We ground our architecture in program synthesis: a computational approach that models human learning and reasoning as the process of inferring structured programs (executable code or symbolic rules) from examples or experience. These programs function as hypothetical self-models that can condition future behavior through a self-hypothesis distribution, making the agent’s current self-conception an endogenous cause within its generative model. The program model learns programs that balance three objectives: explanatory adequacy for observed behavior, alignment with the policy’s capabilities, and behavioral utility. This recursive coupling—where programs shape actions that generate data for refining programs—creates attractor dynamics that stabilize into a coherent identity and narrative self. We argue that MPE emerges when the system operates with minimal narrative elaboration: programs remain simple and interoceptively focused, the self-hypothesis distribution stays broad rather than collapsing to a dominant narrative, and the agent sustains awareness through recursive self-modeling without recruiting extended conceptual structures.
Deane et al. (Wed,) studied this question.