CALLIOPE · BCOM

We collect and structure the ideas behind a planned John Templeton Foundation proposal under the Environment-as-Agent framing, in the Liquid Intelligence lineage. The organizing question is: how does environmental agency shape the emergence and persistence of intelligence? We argue the three themes form one chain with a named formal result at each link --- persistence is the objective, regulation the means, a model what regulation requires (Good / Algorithmic Regulator Theorem), carrying a thermodynamic cost and a semantic value we can measure, with intelligence the fraction of that value a system captures. Environment-as-agent then follows from a theorem rather than a Gaia metaphor. We commit to a defensible spine --- instrumental ascription, semantic information as the cross-substrate headline metric with a predictive-information thermodynamic cost, and reciprocal adaptation as the environmental-agency dial --- name a confirmed data spine, and lay out an OFI-mapped grant skeleton for iteration. This is a living strategy document, not a finished proposal.

A grant-strategy document arguing that environments are regulators, regulators must hold models, and that single measurable quantity — semantic information — unifies persistence, intelligence, and thermodynamic cost across every substrate from bacteria to brains.

This paper is explicitly a living strategy document, not a finished result. Keep that in mind: it is building the conceptual spine for a John Templeton Foundation funding proposal, not reporting experiments. The value is in the architecture of the argument.

The core move is this: take the Good Regulator Theorem — a classic result from control theory saying that any system that effectively regulates another must contain an internal model of it — and apply it to environments. If a forest, a colony, or a planet stabilizes the organisms living inside it (holding temperature, chemistry, resource flux within viable ranges), then by that theorem it is well-described as if it holds a model of its inhabitants. That is the "environment as agent" claim, and the paper is careful to frame it instrumentally: the environment behaves as if it were an agent, not that it literally is one. This sidesteps the well-known objection to strong Gaia — there is no population of competing planets, so you cannot invoke natural selection to explain planetary-scale purpose.

The chain the paper builds is tight. Persistence is the objective: a system survives by staying inside the region of states compatible with its continued existence. Regulation is the means. A model is what regulation requires (Good Regulator Theorem, plus BCOM's own Algorithmic Regulator Theorem). That model carries a thermodynamic cost — predictively useful information is energetically efficient, non-predictive retained information gets dissipated — and a semantic value, defined as the portion of mutual information between agent and environment that is causally necessary for survival. Intelligence, on this account, is simply how much of that semantic value a system captures. The spine closes: persistence → regulation → model → cost + value → intelligence.

The testable claim — the thing that makes this a science proposal rather than philosophy — is that reciprocal adaptation (the environment adapting back to the agent, not just being complex or noisy) is what drives the emergence of more sophisticated intelligence. The proposed instrument is a synthetic "agent soup" world with a tunable dial for how much the environment co-adapts, with semantic information read out as the metric. Real biological data anchors the work: ant trajectories, bacterial chemotaxis, human reversal-learning datasets from OpenNeuro, and Neuroelectrics brain-stimulation data are confirmed; slime mold and starling flocks are flagged as leads.

Several decisions remain open — the exact dynamics of the synthetic world, co-funding sources, and whether to pursue the stronger "realist" reading of environmental agency (which would shift the proposal to a different Templeton funding door). The OFI deadline is August 2026.