ALGOCOSMOS: Project Proposal Materials

Agency is a universal organizing principle of life and mind — and this project wants to make that claim mathematically precise and falsifiable.

The core idea is simple but ambitious: whether you're a single cell, a brain, or a society, you count as an "agent" if you do three things — build a compressed model of your world and self, maintain goals (homeostatic or longer-range), and act by simulating possible futures. ALGOCOSMOS calls this the Algorithmic Agent Triad. The formalism comes from Algorithmic Information Theory (AIT), which treats intelligence as compression: a good model is one that compactly encodes regularities in the environment. Because true Kolmogorov complexity is uncomputable, the project proposes using practical compression proxies (Lempel-Ziv variants, neural compressors) as stand-ins.

What makes this more than philosophy is the commitment to operational tests. The proposal describes an ON/OFF intervention protocol: perturb a putative regulator and check whether the system's behavioral complexity drops relative to a null baseline — and only when that regulator embodies a non-trivial internal model. This is the "Algorithmic Regulator Theorem" in action, tracing back to Conant and Ashby's 1970 result that every good regulator must be a model of the system it controls. The project also wants to map a "cognition morphospace" — a quantitative coordinate space where cells, brains, and societies can all be plotted — and then study the empty regions ("voids") to understand what constraints or evolutionary accidents keep certain kinds of agency unexplored.

The three flagship pathologies are the sharpest part of the proposal. Cancer is framed as multicellular anarchy: cells defecting from the shared tele-homeostatic objective of the tissue. Depression is framed as attractor rigidity in a fixed-connectome brain: the modeling and planning loops get stuck, unable to update maladaptive world models. Polarization in social networks is framed as objective-function drift in a rewiring ("liquid") collective, where feedback control degrades. These aren't loose analogies — the proposal explicitly treats them as homologous failure modes of the same triad machinery, which forces the mathematical framework to generalize or break.

This is a grant proposal document, not a completed research paper, so the source is necessarily programmatic rather than results-bearing. The scientific content is substantive and internally consistent, but the decisive experiments, agency atlases, and biomarkers described are targets, not findings. The proposal's self-aware risk section is worth noting: the authors anticipate the "theory of everything" criticism and respond with a deliberately small exemplar set, a single metric family, and explicit go/no-go milestones designed to kill the framework early if it's wrong.