CALLIOPE · BCOM

Future autonomous artificial and hybrid life forms (i.e., feral machines) will likely depend on self-organized automation for the design and manufacture of new individuals. This project aims to provide a modernized account of John von Neumann’s theory of self-replicating automata, incorporating recent advances in kinematic self-replication, nanoscale molecular machines, and engineered living systems such as Xenobots. The project seeks to identify the minimal material, informational, and algorithmic conditions necessary and sufficient for the self-replication and reproduction of cognitive and functional entities (i.e., robots) across multiple spatial and temporal scales. It will also investigate the algorithms capable of discovering such specifications for particular ecological or functional niches, while accounting for allometric scaling laws and thermodynamic costs of computation. As an initial proof of concept, a toy model of computational origins of life and self-replication has been developed using an algorithmic primordial soup implemented in the programming language Brainfuck (BF). In this system, self-replication emerges and remains stable across increasing program scales through hierarchically nested self-replicating subprograms. However, it remains unclear how such principles can be translated into scalable processes capable of enabling the self-reproduction, persistence, and open-ended evolution of machines at spatiotemporal scales comparable to those of human civilization. Central questions include whether complex self-replicating robots must be “grown” through developmental processes analogous to biological morphogenesis, and, if so, which physical substrates and algorithmic frameworks are most promising for achieving this.

A computational-phenomenology account of \emph{pure awareness} (PA) and minimal phenomenal experiences (MPE) built on the algorithmic agent of KT (cf.\ WP0101). Bridges first-person reports (notably Metzinger's $\sim$ 500-report corpus, 2024), Buddhist meditative phenomenology (e.g.\ \emph{jh=ana}, Dzogchen non-dual awareness), and predictive-processing implementations.

\textbf{Setting.} The agent is a Modeling Engine + Comparator + Updater + Objective Function + Planner that compresses sensory data into a hierarchy of generative models. \emph{Fabrication} = active hierarchical modeling that constructs ``things''; \emph{de-fabrication} = transient down-regulation of upper-hierarchy models, equivalent to flattening higher-order attractors in the brain's dynamical landscape.

\textbf{Six core hypotheses.} \begin{enumerate}\itemsep -2pt \item PA as a \emph{stand-alone} phenomenon corresponds to highly defabricated states (few active models). \item In some such episodes the \emph{only} active model is the agent's model of \emph{its own modeling process} --- giving residual structure to an otherwise contentless experience. \item PA combined with phenomenal content (non-dual awareness) reflects recognition that \emph{all} contents are products of the modeling process. \item Many phenomenal qualities of PA (luminosity'', knowing'', ``coming home'', boundlessness) arise from this recognition and are shaped by cultural/metaphysical priors. \item Recognition of the modeling process is facilitated by defabrication via meditation or psychedelics. \item This recognition can durably reduce suffering --- the goal of many contemplative traditions. \end{enumerate}

\textbf{Mechanism in dynamical-systems terms.} Attentional and deconstructive meditations (insight on impermanence/non-self) reduce precision at upper levels, gradually flattening high-order attractors. Psychedelics produce a sharper, less regulated flattening via 5-HT $_{2A}$ agonism; compounds differ (e.g.\ 5-MeO-DMT $\rightarrow$ near-global collapse and contentless states; psilocybin/DMT $\rightarrow$ preserved mid-/low-level structure $\rightarrow$ rich visual content).

\textbf{Transparency vs.\ opacity.} A state is \emph{opaque} when the agent models a representation \emph{as} a representation, lowering its confidence and its weight in valence/policy computation. Opacity = de-reification; non-dual awareness corresponds to high-fabrication, high-opacity experience. This converges with related notions of \emph{epistemic depth} (Laukkonen et al.) and \emph{parametric depth} (Sandved-Smith et al.).

\textbf{Cessation as theoretical limit.} Reports of cessation (suspension of phenomenal time, ``time pockets'') sit at the floor of the fabrication spectrum; PA as stand-alone occupies the boundary just above.

\textbf{Architectural claim vs.\ Active Inference.} In the algorithmic-agent framework, model updating and valence computation are \emph{distinct} processes --- not unified under expected free energy --- predicting dissociations between accurate world-modeling and valence (relevant to de-reification-without-defabrication, i.e.\ non-dual awareness with preserved cognition).

\textbf{Empirical handle.} Implement via the Laminar Neural Mass Model (LaNMM): defabrication = systematic reduction of upper-layer precision, predicted as a cascade of alpha-power reductions in progressively deeper cortical layers, with reciprocal gamma increases --- consistent with psychedelic spectral signatures and testable in jh=ana practitioners.

\textbf{Ethical corollary.} Suffering tracks confidence in evaluative models more than their content. Artificial agents endowed with an explicit model of their own Modeling Engine could operate in a non-dual-like mode, potentially reducing artificial suffering.