An Algorithmic Agent Model of Pure Awareness and Minimal Experiences

The phenomenon of "pure awareness", central to many contemplative traditions, has recently attracted scientific interest for its relevance to the study of consciousness. In this paper, we investigate pure awareness through the algorithmic agent model, a computational framework with roots in algorithmic information theory. This framework proposes that agents build compressive models of the world for evolutionary success. Structured experience arises from running such models, thus linking phenomenology with computation and offering insights into the emergence of pure awareness in both natural and artificial systems. We propose that pure awareness, as a stand-alone phenomenon, may correspond to minimally structured experiences achieved through meditation, psychedelics, or other deconstructive practices. Crucially, while such experiences are often perceived as contentless or unfabricated, both phenomenological and theoretical evidence suggest that some degree of fabrication or modeling remains involved. A key hypothesis is that the phenomenology of pure awareness arises from a specific model: the agent's model of its own modeling process. Importantly, the agent's recognition of the modeling process can occur alongside other phenomenal content (as in non-dual awareness) and is associated with stable changes in valence computation, potentially reducing suffering. Our exploration offers new insights into consciousness science by examining the minimal possible experiences for an agent. It also contributes to refining our moral stance towards both natural and artificial agents and helps us better understand the mechanisms and constraints involved in facilitating such transformative experiences.