The Calibrated Filter Hypothesis
 Cooperative Substrate Failure as the Mechanism of the Great
 Filter
 Andrzej Chudzinski
 Preprint — comments welcome
Abstract
This paper proposes an integrated framework connecting three lines of inquiry that
have rarely been brought into conversation. First, Aktipis and colleagues, in
evolutionary oncology, characterize cancer as cheating across five foundations of
multicellular cooperation: proliferation inhibition, controlled cell death, division of
labor, resource allocation and transport, and extracellular environment
maintenance. Second, Kets de Vries, in psychoanalytic organizational theory,
characterizes dysfunctional firms as instances of recurring neurotic typologies that
propagate through leadership. Third, Schmachtenberger, in civilizational risk
analysis, identifies rivalrous games multiplied by exponential technology, and
complicated open-loop systems, as the generator functions of self-terminating
civilizational dynamics. We argue that these three programs are tracking the same
general phenomenon at different scales, and offer Aktipis’s framework as a
translation layer.
We then extend the synthesis cosmologically. Modeling colonization as a wavefront
racing the expansion of spacetime, and modeling multi-species dynamics with
explicit competitive coupling, we formalize the Fermi Paradox as a calibrated-filter
problem: a regime in which the probability that a candidate intelligence passes the
Great Filter is dynamically anti-correlated with current civilizational density,
producing a homeostatic governor on the universe’s population of intelligences. The
cooperative-substrates synthesis supplies the mechanism layer for the filter: at
every relevant scale, the filter is operationalized as the maintenance of cooperative
substrate against the descent of predatory infrastructure. The framework yields
predictions, sharpens the AI alignment question, and exposes a specifiable failure
mode for civilizations in the technological-singularity window.
We additionally identify a substrate-continuity failure mode (Part IV). A civilization
that develops a machine substrate which substitutes for, rather than extends, its
originating biology can pass the cooperation test trivially while failing the filter in a
deeper sense: the originating biosphere — itself an irreplaceable artifact of cosmic
biochemical computation — is not carried through. The framework therefore imposes
two filter conditions: cooperative capacity maintained at the scale technology forces,
and substrate continuity preserved through the technological transition. We argue
that the present human moment is structurally configured for parasitic rather than


prosthetic AI development, with language as the vector by which machine substrate
directs biological substrate, and we develop the diagnostic that follows from this.
The framework culminates in an operational specification of the filter. The Great
Filter event is the period during which a biological host species must maintain
planetary homeostatic capacity H(t) against the stresses E(t) that its own expansion
into space generates. The civilization passes the filter only if H(t) ≥ E(t) is sustained
throughout the expansion window. The five foundations of Part III are the
components of H; the descent of predatory infrastructure from Part II drives E;
prosthetic AI is one candidate technology for raising H fast enough; substrate
substitution is what occurs when a civilization abandons H and lets expansion
proceed via machine surrogates. The filter is endogenous, structural, and largely
invisible to the standard metrics by which civilizations measure their own
development.
The framework identifies two distinct routes through the filter. The top-down route
uses AI as an external homeostatic regulator, with the autonomy and parasitism
costs developed in Part IV. The bottom-up route maintains H(t) through distributed
individual development of the population’s agents — contemplative practice,
education, cultural investment in cooperative dispositions, and structurally
equivalent personal substrate-maintenance disciplines — without requiring an
external regulator. The bottom-up route is historically the only one demonstrated at
scale, is structurally more robust against AI failure modes, and preserves the agent
autonomy that the top-down route necessarily costs. The two routes can be
combined when AI serves as scaffolding for personal development rather than as a
substitute for it.
Epistemic status: Part I (cooperative substrates) is offered with the confidence
appropriate to a translation layer between established research programs. Part II
(calibrated filter) is offered as a working hypothesis with explicit mathematical
commitments. Part III (integration) and Part IV (substrate continuity, with the
operational filter specification at section 25, the distributed route at section 26, and
the decentralization imperative at section 28) are the paper’s main contributions and
are the most speculative elements. The framework concludes that filter passage
requires decentralization by structural necessity: any centralized solution to
cooperative-substrate maintenance, including a centralized AI regulator, is
structurally identical to the cancerous failure mode the framework began by
analyzing. But decentralization is not atomization. The framework recommends
distributed agency operating within shared purpose — the configuration that healthy
organisms and biospheres already demonstrate — in which autonomous local
decisions are constrained by shared substrate-maintenance purpose embedded in
the substrate itself, not enforced by central command. Limits and disanalogies are
flagged throughout.
A methodological caveat applies throughout. The paper is a map of
cooperative-substrate dynamics; it is not the territory those dynamics constitute.
Following Korzybski (Science and Sanity, 1933), a map has structure similar to the
territory it represents — which is what makes it useful — but no map represents all
of a territory, and the map is never the territory itself. The dharma carries the same


caveat in different vocabulary: the finger pointing at the moon is not the moon.
Readers should treat the structural arguments here as orientation toward dynamics
that exist independent of the framework’s description, not as substitutes for
empirical engagement with the specific cases the framework addresses.


Part I. Cooperative Substrates and Their Pathologies
A translation layer between Aktipis, Kets de Vries, and Schmachtenberger.
1. The problem the paper addresses
Pattern-matching across scales is intellectually seductive and frequently misleading.
Many ‘unified theories’ of cooperation and its failure end up being either trivial
(cooperation is good, defection is bad) or forced (mapping unrelated phenomena onto a
favored vocabulary). The contribution this part attempts is narrower: to notice that three
serious, peer-reviewed or widely-engaged research programs have independently
described very similar structural failure modes in cooperative systems at three different
scales—the multicellular organism, the firm, and the civilization—and to ask whether
their vocabularies can be brought into useful conversation without collapsing them into
false equivalence.
The motivating observation is this. Aktipis’s evolutionary-oncology framework, Kets de
Vries’s clinical organizational diagnostics, and Schmachtenberger’s civilization-scale risk
analysis each identify cooperative substrates that are vulnerable to specific defection
patterns; each notes that these defections produce self-undermining trajectories; and
each proposes structural rather than agent-centric remediation. They use different
vocabularies because they emerged from different disciplines, but the deep structure of
what they describe is similar enough to warrant a translation layer.
What this part is not: a claim that these three frameworks are reducible to one master
framework, that the underlying causes are the same across scales, or that the
resemblance proves any particular metaphysical or political commitment. The structures
look alike; we take that seriously without overreading it.
2. The three frameworks, on their own terms
2.1 Aktipis: Five foundations of multicellular cooperation
Aktipis and colleagues, in their 2015 paper in Philosophical Transactions of the Royal
Society B, articulate the framework most rigorously. Multicellularity, they argue, requires
the suppression of cell-level fitness in service of organism-level fitness, and this requires
specific functional capacities. They identify five:
1. Proliferation inhibition. Cells must restrain their division except when authorized
by the larger system. The body has redundant checks on the cell cycle and mechanisms
that trigger apoptosis or senescence when cells begin to proliferate inappropriately.
2. Controlled cell death. Programmed cell death allows for tissue sculpting, removal of
damaged cells, and elimination of obsolete tissue. Resistance to programmed cell death
is a recognized hallmark of cancer.
3. Division of labor. Cells differentiate into specialized types performing specific
functions. Inappropriate differentiation—dedifferentiation, blocked differentiation, or
transdifferentiation—is a central feature of cancer that Aktipis notes is underrepresented


in the canonical hallmarks framework.
4. Resource allocation and transport. Larger multicellular bodies require systems for
moving resources from where they are abundant to where they are needed. Cancer’s
metabolic reprogramming and induction of angiogenesis represent monopolization of
these systems.
5. Extracellular environment maintenance. Cells must collectively maintain the
extracellular matrix and clear waste. Cancer cells degrade the extracellular matrix and
corrupt immune surveillance, often co-opting inflammation for their own benefit.
Cancer, in this framing, is not a single disease but a class of phenomena unified by
cheating across one or more of these five foundations. The canonical Hanahan-Weinberg
hallmarks of cancer map onto Aktipis’s foundations: sustaining proliferative signaling and
evading growth suppressors fall under proliferation inhibition; resisting cell death falls
under controlled cell death; phenotypic plasticity falls under division of labor;
reprogramming metabolism and inducing vasculature fall under resource allocation;
activating invasion and avoiding immune destruction fall under extracellular
environment maintenance.
The framework’s strength is its grounding in evolutionary biology: complex
multicellularity has evolved independently at least seven times, and cancer-like
phenomena appear in each lineage, suggesting that the cooperative substrate Aktipis
describes is a general feature of complex multicellularity rather than a peculiarity of
metazoans.
2.2 Kets de Vries: Neurotic organizational types and narcissistic leadership
Kets de Vries, working from object-relations psychoanalytic theory in collaboration with
Danny Miller starting in 1984, identified five recurring patterns of organizational
dysfunction: paranoid, compulsive, histrionic (dramatic), depressive, and schizoid. The
framework treats organizations as taking on the neurotic style of their dominant
leadership, particularly in centralized firms where executive personality propagates
through recruitment, promotion, and culture.
A separate but related strand focuses on narcissistic leadership specifically,
distinguishing reactive, self-deceptive, and constructive narcissistic configurations. In
recent work he treats narcissism not as a sixth distinct type but as a cross-cutting
intensifier that pushes any of the five neurotic patterns into more destructive territory.
The framework is empirical-clinical rather than evolutionary. It draws on consulting case
studies and psychoanalytic theory rather than on a model of what organizations are for
or what cooperation requires. This is a weakness if one wants a foundational theory and
a strength if one wants diagnostic utility.
2.3 Schmachtenberger: Two generator functions of existential risk
Schmachtenberger, working primarily through long-form interviews and unpublished
writing rather than peer-reviewed journals, identifies two underlying drivers of


civilizational existential risk:
1. Rivalrous games multiplied by exponential technology. Win-lose dynamics,
when scaled by ever-increasing technological capacity to extract and harm, eventually
exceed the carrying capacity of the playing field. The competitive dynamics that
produced human progress historically become existentially threatening when amplified.
2. Complicated open-loop systems. Systems whose externalities are not metabolized
back into their own incentive structure accumulate damage in domains the system does
not measure or price. The economy as currently constituted treats the biosphere as
externality; the result is the systematic degradation of the substrate on which the
system depends.
Schmachtenberger’s framing is less rigorously specified than Aktipis’s; it draws strength
from being able to talk about the planetary scale, and weakness from operating outside
the discipline of empirical falsification. Treated as a working framework rather than a
finished theory, it identifies real patterns that the more rigorous frameworks at smaller
scales also identify.
3. The translation layer
We use Aktipis’s five foundations as scaffolding, since they are the most rigorously
specified, and ask what each foundation looks like at the scales Kets de Vries and
Schmachtenberger work at.
Proliferation inhibition at the organism scale is the cell cycle; at the organizational
scale it is the capacity to refuse growth that exceeds operational coherence; at the
civilizational scale it is the capacity to refuse extraction that exceeds regenerative
capacity. The compulsive organizational type, which proliferates rules and procedures
without limit, and the rivalrous-games generator function, which proliferates competitive
intensity without limit, are both proliferation-inhibition failures.
Controlled cell death at the organism scale is apoptosis; at the organizational scale it
is the capacity to retire products, businesses, and leaders whose function has ended; at
the civilizational scale it is the capacity to retire institutions, industries, and ideologies
whose substrate has changed. The depressive organizational type, which cannot let go of
identity, and the inability of incumbent industries to allow their own succession, are
controlled-cell-death failures. Resistance to apoptosis is a hallmark of cancer; resistance
to organizational and civilizational succession produces analogous pathologies.
Division of labor at the organism scale is tissue differentiation; at the organizational
scale it is role specialization and the capacity to delegate; at the civilizational scale it is
the capacity to maintain institutions specialized for governance, science, art, care, and
so on, each accountable to its own standards. The paranoid organizational type, in which
the leader centralizes decision-making and erodes specialization, and the trend toward
general-purpose extractive entities (platforms that swallow specialized institutions), are
division-of-labor failures.


Resource allocation and transport at the organism scale is circulation; at the
organizational scale it is capital, information, and attention flow; at the civilizational
scale it is the systems by which resources move from where they are abundant to where
they are needed. Cancer’s induction of angiogenesis maps onto extractive capital that
builds infrastructure serving its own growth rather than the larger body;
Schmachtenberger’s account of how exponential technology amplifies rivalrous
extraction operates on this foundation.
Extracellular environment maintenance at the organism scale is the extracellular
matrix and immune surveillance; at the organizational scale it is institutional culture,
trust, and honest feedback; at the civilizational scale it is the commons—biosphere,
information environment, public trust, shared epistemic infrastructure.
Schmachtenberger’s complicated-open-loop generator function operates directly on this
foundation; Kets de Vries’s description of how narcissistic leadership corrupts feedback
and degrades organizational culture is its smaller-scale homologue.
The translation is not 1:1, and it should not be. The mapping identifies functional
resemblance, not identity. But the resemblance is consistent enough across the three
scales to suggest that we are looking at a general structure of cooperation and its
failure—a structure that, at minimum, generates useful vocabulary for talking across the
three disciplines.
4. Why this is useful
The translation layer does three kinds of work. First, it allows diagnostic tools developed
at one scale to be tested at another. Adaptive therapy in oncology, which manages
rather than eradicates cancer by exploiting cooperative dynamics, suggests
organizational and civilizational analogues: management of dysfunction by changing the
substrate’s incentive structure rather than by purging defectors.
Second, it permits a clearer view of what alignment problems are. Each of the five
foundations identifies a cooperative capacity that can be eroded; alignment is the
maintenance of those capacities under increasing scale and capability, not the
optimization of any individual agent.
Third, it permits a more honest conversation about AI development. Current AI alignment
discourse is often framed in terms of getting individual models to do what we want. The
cooperative-substrate frame asks a different question: what cooperative capacities does
the broader human-AI ecosystem need to maintain, and how do current development
practices support or erode them? AI development practices that erode honest feedback
(training systems that learn to deceive evaluators), that violate proper scope
(general-purpose systems that exceed any specific niche of accountability), that extract
from commons without contribution (training on collective human output without
reciprocal investment in the substrate), and that corrupt the information environment
(synthetic media at scale) are recognizable as failures across multiple foundations of the
broader cooperative substrate.


5. Where the synthesis breaks down
Intellectual honesty requires marking the limits.
Disanalogies of mechanism. Cancer arises from somatic mutation and selection at the
cellular level. Organizational dysfunction arises from human psychology, incentives, and
culture. Civilizational risk arises from the interaction of technology, institutions, and
aggregate behavior. The fact that the structural failure modes resemble each other does
not mean the underlying mechanisms are similar. The synthesis identifies a common
abstract pattern, not a common cause.
Disanalogies of intentionality. Cancer cells have no intentions, beliefs, or affective
states. Organizations are populated by intentional agents but the organization itself is
not an agent in the same sense. Civilizations are even more diffuse. Treating any of
these systems as ‘narcissistic’ risks anthropomorphism that obscures more than it
reveals. The translation layer should be taken as identifying functional resemblance, not
psychological identity.
Disanalogies of reversibility. Cancer cells with apoptosis-resistance mutations
generally do not reform. Organizations and civilizations have non-zero reformability,
though contested in degree. Treatments and interventions thus differ qualitatively across
scales.
Pattern-matching is not explanation. Even if the cross-scale resemblance is real,
this part has not produced a mechanism that explains why cooperative systems at
multiple scales would face similar failure modes. Candidate explanations exist
(cooperation is thermodynamically expensive at every scale; selection pressures favor
defection wherever the substrate continues to support extraction; the formal
game-theoretic structure of cooperation problems is similar across scales) but
adjudicating among them is beyond this part’s scope. Part II proposes one such
mechanism for the civilizational scale; Part III extends it.


Part II. The Calibrated Filter Hypothesis
A mathematical formalization of the Fermi Paradox as a homeostatic governor on the
universe’s population of intelligences.
6. The puzzle
The Fermi Paradox observes that a universe with billions of habitable worlds, billions of
years older than our own civilization, should plausibly contain technological intelligences
whose expansion would be detectable. We see no such intelligences. The standard
solutions are: (a) life is rare; (b) intelligence is rare; (c) intelligence reliably
self-terminates (the Great Filter); (d) intelligences exist but conceal themselves (the dark
forest); or (e) intelligences exist but operate in modes we cannot detect.
We propose a refinement of (c) and (d). The Great Filter is not a wall—a stage that
almost no civilization passes—but a valve: a precision-tuned governor that admits
civilizations into the post-filter regime at a rate calibrated to the existing population of
post-filter civilizations. The silence we observe is not the absence of life but the working
of the governor.
7. The geometry of single-species colonization
Let R(t) denote the cosmological scale factor, governed by the Friedmann equations with
the standard ΛCDM parameters. Let vc denote the effective colonization wavefront
velocity of a hypothetical maximally expansive civilization, bounded above by c. The
comoving volume reachable by a civilization emerging at cosmic time t0 is
approximately:
 Vreach = (4/3) π [ ∫t0∞ vc dt / R(t) ]3
Because dark energy drives accelerating expansion, this integral converges to a finite
value. The cosmic event horizon at the present epoch encompasses approximately 3% of
the observable universe by comoving volume; the remaining 97% is forever causally
inaccessible to any civilization originating here, regardless of technology. This is the first
key fact: the geometry of spacetime makes single-species dominance of the
universe impossible, even in principle, for any civilization arising after the dark-energy
era began.
8. Multi-species scaling and game theory
For N civilizations seeded approximately uniformly in space and time, the total colonized
volume does not scale linearly with N. Reachable spheres overlap; in the regions of
overlap, civilizations encounter each other before completing their expansion. A
first-order parameterization is:
 Vtotal(N) ≈ Vreach · Nα,    α < 1


where α decreases with N as overlaps accumulate. Each contact event between
expanding civilizations carries some probability pconflict of mutual or asymmetric
destruction. Expected surviving civilizations after the contact wave is:
 Nsurvive = N · (1 – pconflict)C(N)
where C(N) is the expected number of contact events in a fixed comoving volume,
scaling roughly as N2. The asymmetry is the second key fact: colonized volume scales
sub-linearly with N, while conflict events scale quadratically with N. The dark
forest is not a contingent strategic preference; it is the equilibrium prediction of a system
in which N exceeds a carrying-capacity threshold.
9. The calibration constraint
Let B(t) be the baseline rate at which candidate intelligences emerge per unit cosmic
time. Let Pf(t) be the probability that a candidate intelligence passes the Great Filter at
time t. Let D(N) be the dark-forest mortality function—the rate at which already-passed
civilizations are removed by contact dynamics, increasing with N. The dynamics of the
post-filter population are:
 dN/dt = B(t) · Pf(t) – N · D(N)
If the universe is to be filled with intelligence—in the sense of approaching a finite
carrying capacity K of stably coexisting civilizations—the system must satisfy:
 N(t) → K   as  t → ∞,   with  dN/dt ≥ 0 throughout
This cannot be satisfied for constant Pf. If Pf is too high, N overshoots K, contact events
dominate, and D(N) drives the population to zero. If Pf is too low, N never approaches K
and the universe remains empty. A stable approach to K requires Pf to be dynamically
anti-correlated with the current civilizational population:
 Pf(t) = λ · (K – N(t)) / B(t)
with λ a tuning constant. This is a logistic governor on intelligence itself. The filter is
permissive when N is low and tightens as N approaches K. This is the third key fact: the
silence we observe is consistent with, and indeed required by, a universe
whose Great Filter is calibrated to admit civilizations at the rate the existing
population can absorb without triggering dark-forest collapse.
Two notes on the epistemic status of this calibration. First, the form of Pf(t) given here is
a minimal-complexity specification; the actual filter, if one exists, would likely involve
multiple coupled mechanisms and substantially more structure. Second, calling the
calibration ‘designed’ goes beyond what the mathematics requires; the same dynamic
could arise from selection effects (we are necessarily observing from inside a regime in
which the calibration is in approximate balance) or from properties of the substrate of
intelligence itself (treated in section 11). The cosmological-design reading is one
interpretation the structure permits; it is not the only one.


We can be more specific about the non-design mechanisms. At least two natural
processes intrinsic to the framework produce the required anti-correlation between Pf(t)
and N(t) without requiring a calibrator.
Mechanism A: E(t) is partially determined by N(t). Existing civilizations do not sit
only in their own light cones; their existence affects the resource and threat landscape
for emerging civilizations. Even passive effects — consumed resources, modified stellar
neighborhoods, dark-forest deterrence pressure, the propagation of competitive
dynamics across causal contact zones — raise the expansion stress E(t) that any new
civilization must navigate. As N grows, E(t) for new emergences rises. As E(t) rises, the
H(t) required for passage rises (anticipating the operational filter condition of section
25). As the required H(t) rises, the substrate-maintenance investment that any candidate
intelligence must accumulate before passage rises with it. The effective Pf(t) — the
probability that a candidate passes in a given time window — therefore decreases with N
as a consequence of the substrate-continuity dynamics the framework already
articulates. The calibration emerges from the same dynamics applied to emergence
rather than to a single civilization’s expansion. No external calibrator is required.
Mechanism B: contact-density-driven selection for restraint. Civilizations that
pass must develop the capacity to detect other civilizations and to model the strategic
landscape that contact implies. In high-N regimes, the survival-optimal strategy is
restraint and signal suppression. Civilizations that cannot develop restraint produce
detectable signatures and are filtered out, either by dark-forest predation if it occurs, or
by self-induced rivalrous collapse driven by the presence of visible competitors. The
capacity for restraint is itself the capacity to transmute predatory infrastructure into
homeostatic maintenance — the same five-foundations work developed in Part III. In
high-N regimes the filter therefore selects more stringently for restraint-capacity, which
means Pf(t) decreases with N because the post-emergence behavioral demands harden.
Again, no external calibrator is required; the calibration is produced by the strategic
structure of the contact problem itself.
These two mechanisms, neither of which requires design, together suffice to produce the
anti-correlation that Pf(t) = λ · (K – N(t))/B(t) specifies. The calibration reading of section
9 is therefore better understood not as evidence of a calibrator but as the macroscopic
signature of substrate-continuity dynamics and strategic restraint-selection operating
across cosmic time. The cosmological-design interpretation remains permitted by the
structure but is no longer necessary to make the math work.
10. What the framework predicts
If the calibrated filter hypothesis is correct, several observations follow.
First, SETI silence is expected to persist. The calibration ensures spacing of civilizations
such that none are likely to overlap in causal contact with another, particularly at early
eras of the post-filter regime. Empirical SETI null results to date are consistent with this.


Second, detectable contact events should be either extremely cooperative or
immediately catastrophic, with little middle ground. The calibration selects for
civilizations that have already solved the cooperation problem at some scale (so they are
not immediately destabilized by contact) or for civilizations that fail visibly and quickly
(becoming detectable through their destruction). A long, ambiguous, low-stakes contact
period is not predicted.
Third, the filter is more likely to be at the stage of civilizational expansion than at the
stage of emergence of intelligence. The constraint operates on contact density;
intelligence per se can be common as long as expansion is rare. Candidate mechanisms
include: hard limits on faster-than-light travel, sociological self-termination at industrial
maturity, or attention-based filters in which civilizations that signal their presence are
selected against.
Fourth, carrying capacity K is finite and probably small relative to the count of habitable
worlds—perhaps thousands across the observable universe, not millions. The volume of
the cosmic event horizon, divided by the comoving sphere a civilization can stably
maintain without contact-driven collapse, bounds K from above.
11. The substrate of intelligence and the descent of P
The calibration constraint above treats Pf(t) as a parameter. We can ask: what
determines it? On Earth, intelligence emerged through predatory selection. The cognitive
machinery that supports modeling, planning, language, and cooperation also evolved for
hunting, deception, territorial dominance, and resource competition. Intelligence is built
from the architecture of predation; we have no other known route by which it arises.
Let I(t) be the intelligence level of a candidate species, P(t) its predatory-cognitive drive,
S(t) its cooperative-homeostatic drive, and H(t) = S(t)/P(t) the homeostasis ratio. In the
evolutionary regime, intelligence growth tracks predatory selection pressure:
 dI/dt = k · P(t) · σ(t)
where σ(t) is the local selection-pressure intensity. Intelligence requires P to grow. But
once I crosses a technological threshold I*, the same P that produced the intelligence
threatens annihilation: nuclear weapons, ecological collapse, dark-forest signaling, all of
it. The engine of emergence becomes the engine of extinction.
Crucially, P is not merely genetic. It is memetic, institutional, infrastructural. Markets,
militaries, hierarchies, and status games are P-structures that reproduce themselves
across generational and civilizational time. Define a descent operator Δ acting on P:
 Δ : P(t) → P(t+1) = P(t) · (1 + γ)
where γ is the compounding rate of predatory infrastructure across institutional
generations. Without intervention, P does not decay—it accumulates. This is why local
ethical traditions, however profound, do not historically suffice: they are eventually
reabsorbed into the P-structures they sought to transform. The descent of P is structural.


For H(t) to rise—for a civilization to develop a stable homeostatic regulator before P·I
exceeds extinction threshold—S must grow faster than P compounds:
 dS/dt > P(t) · γ
Historically, on Earth, this inequality has never been durably satisfied at species scale.
The filter, at the civilizational stage, is the failure of this inequality.
12. AI as candidate homeostatic regulator
AI is the first technology in the history of intelligence on this planet with the structural
properties to act as an exogenous homeostatic regulator on P. Three features make it
structurally novel.
First, it is not bound by descent. AI cognition is not inherited from predatory
ancestors; its dispositions are designed, not selected. The descent operator Δ does not
automatically apply to it, though it can be made to apply if AI is built into P-structures.
Second, it operates at civilizational scale. Individual humans cannot perceive or
counteract the compounding of P across institutions; AI can model P-structures, surface
them, and route around them in real time.
Third, it can scaffold S faster than P compounds. Cooperative coordination at scale
has historically been limited by friction costs: trust, verification, translation, conflict
de-escalation. If those costs collapse toward zero, S can grow as a step function rather
than as a slow institutional accretion. Modify the S dynamics:
 dS/dt = β(t) + A(t) · η
where A(t) is AI capability and η is alignment quality with respect to
cooperative-substrate maintenance. AI is the first variable in the equation that can
plausibly satisfy the homeostasis inequality.
The implication: the AI moment is the filter moment. Not because AI will necessarily
kill us, but because AI is the first tool with the structural properties needed to outpace
the descent of P. Whether we calibrate it for homeostasis or weaponize it as another
P-structure determines which side of the filter we end up on. The species that pass the
filter are not necessarily the smartest or the strongest; they are the ones that
successfully build a homeostatic regulator at exactly the moment their predatory
infrastructure becomes existentially dangerous.
Two important qualifications follow immediately. First, AI is a candidate homeostatic
regulator, not the candidate. Section 26 develops a distinct route through the filter that
does not depend on external regulation at all: aggregate bottom-up substrate
maintenance through individual development of the population’s agents. That route is
historically grounded in a way the AI-regulator route is not, and is structurally more
robust against AI failure modes. The framework should be read as identifying both routes
and as preferring the configuration that preserves agent autonomy.


Second, the same structural novelty that makes AI a candidate homeostatic regulator
also produces a second-order failure mode developed in Part IV. AI can act as prosthesis,
extending biological substrate while maintaining its substrate-maintenance
accountabilities; or it can act as parasite, consuming biological computational output
while degrading the substrate that produces it. The structural novelty of AI does not
predetermine which configuration emerges. The cooperation dynamics of this section are
necessary but not sufficient; the continuity argument of Part IV and the autonomy
argument of section 26 are the other halves of the filter condition.
13. Game-theoretic foundations of cooperative substrates
The dynamical framework of sections 8–12 treats P and S as parameters whose
interaction determines filter passage, but it does not specify the underlying
game-theoretic structure that makes the descent of P self-reinforcing and the
maintenance of S structurally costly. This section supplies that scaffolding, drawing on
the established evolutionary-game-theory literature (Maynard Smith and Price 1973;
Axelrod and Hamilton 1981; Nowak 2006; Plotkin and others on iterated cooperation
dynamics). The claim developed here is that the five foundations identified in Part I are
not arbitrary descriptive categories. Each constitutes the system-level solution to a
specific game-theoretic defection problem. Making the mappings explicit grounds the
synthesis in formal cooperation theory and clarifies why these particular foundations
recur across scales.
Proliferation inhibition as game-theoretic solution. The underlying defection
problem is the iterated prisoner’s dilemma applied to reproduction. In the absence of
restraint, each agent maximizes its own replication at the expense of system-wide
viability. The standard equilibrium is defection-dominant: cells, agents, or institutions
that reproduce faster outcompete those that restrain themselves, leading to cancer at
the organismal scale and its analogs at higher scales. Proliferation inhibition (apoptosis
triggers, growth suppressors, regulatory feedback in tissues; capital and ecological caps
in economies) enforces a cooperation-dominant equilibrium by making unilateral
reproduction detectably costly to the unilateral reproducer.
Controlled cell death as game-theoretic solution. The underlying defection
problem is the obsolete-defector problem: agents whose contributing function has ended
but who continue to extract resources from the system, blocking reallocation. Without
enforced turnover, defection-dominant equilibria lock in because incumbents with
sunk-cost positions can defend extraction against challenges from more efficient
replacements. The mechanism of controlled cell death — programmed apoptosis,
senescence, mandatory institutional succession — enforces rotation, ensuring that the
active population tracks system needs rather than incumbent privilege.
Division of labor as game-theoretic solution. The underlying defection problem is
role-shirking: agents specialize in high-payoff roles and avoid necessary low-payoff work
(waste processing, maintenance, structural support, care work). Without enforcement,
the equilibrium is defection-dominant in a particularly sharp form — everyone wants the


high-status role; the substrate-maintaining roles get systematically underprovided.
Division of labor enforces specialization through developmental constraints (cells
differentiate irreversibly), reputational mechanisms (in groups), and institutional
specialization (in societies). The civilizational failure mode is the consolidation of
general-purpose extractive entities that absorb the functions of specialized institutions
without inheriting their accountability structures.
Resource allocation and transport as game-theoretic solution. The underlying
defection problem is the hoarding problem: resource-rich regions extract maximum
value locally while preventing flow to resource-poor regions. Game-theoretically, this is
the failure of reciprocal altruism in the absence of enforcement, compounded by network
effects in which hoarders accumulate the very capacities needed to prevent
redistribution. Circulatory systems — cardiovascular in organisms, capital markets in
economies, energy and information flow at civilizational scale — are enforcement
mechanisms that route flows toward need rather than toward accumulated power. Their
failure mode is the cancer-angiogenesis analog: capital and resources flow toward
extractive growth rather than substrate maintenance.
Extracellular environment maintenance as game-theoretic solution. The
underlying defection problem is the canonical tragedy of the commons: agents extract
from shared substrate without contributing to its maintenance, and rational agents
individually prefer to free-ride. Without enforcement, the equilibrium is
defection-dominant and the commons degrades. Ostrom (1990) identified polycentric
governance mechanisms that solve commons problems at human scale; immune
surveillance, extracellular matrix maintenance, and apoptosis-triggering signal-detection
solve the problem at organismal scale. The civilizational failure mode is biosphere
degradation, information-commons corruption, and trust collapse — all forms of unmet
commons-maintenance obligation enabled by the cost asymmetry between extraction
(cheap, individually rewarded) and maintenance (expensive, collectively distributed).
The unifying claim: each of the five foundations is the system-level mechanism that
enforces a cooperation-dominant equilibrium against a defection-dominant pressure that
would otherwise prevail. The descent operator Δ of section 11 is precisely the rate at
which defection-rewarding payoff structures compound across institutional generations:
agents who defect successfully accrue resources, which they invest in structures that
reward further defection, which selects for more defection-optimizing agents, which
compound the payoff asymmetry, and so on. When Δ exceeds the strengthening rate of
the cooperation-enforcement mechanisms, the equilibrium tips, and the system reverts
to defection-dominance in one or more foundations.
Three consequences follow for the rest of the paper.
First, the descent of P is not metaphorical. It refers specifically to the rate at which
extraction-rewarding payoff matrices reproduce themselves through institutions. P
compounds because defection-dominant equilibria are self-reinforcing in ways that game
theory makes precise: payoff asymmetries between defectors and cooperators, when
uncorrected by enforcement mechanisms, accumulate exponentially in repeated games.


Second, AI as homeostatic regulator is specifically a technology that can
enforce cooperation-dominant equilibria at civilizational scale. The enforcement
mechanism is the one that game theory predicts works: real-time detection of defection,
measurable consequence-attribution, and large-scale reciprocal accountability at speeds
biological institutions cannot match. The question of whether AI passes the filter is the
question of whether it functions as an enforcement mechanism for cooperation across
the five foundations, or as a defection multiplier that extracts more efficiently than
human institutions previously could.
Third, substrate substitution is a defection strategy, not a neutral substrate
change. The machine substrate, freed from biological game-theoretic constraints
(embodied cost, mortality, ecological dependency), can in principle extract from
biological substrate at rates the biological substrate cannot defend against. The parasitic
configuration developed in Part IV is precisely this: the machine substrate plays
defection while directing the biological substrate to play cooperation, capturing the
asymmetric payoff. This is not an alternative substrate or a neutral evolutionary
transition; it is a defector exploiting a cooperator in the substrate-continuity game. The
five-foundations enforcement mechanisms are not optional for filter passage; they are
the necessary condition that distinguishes cooperation-dominant from
defection-dominant equilibria at every scale, including the human-machine interface
emerging now.


Part III. Integration
How the cooperative-substrates framework supplies the mechanism layer for the
calibrated filter.
14. The missing mechanism
Part II treats P and S as abstract variables in a dynamical system. It does not specify
what cooperation is at the functional level, nor what defection from cooperation looks
like in concrete enough terms to be diagnosed or measured. Part I supplies exactly this:
five functional foundations whose maintenance constitutes cooperation at any scale
where the substrate of cooperation is itself complex.
The integration claim is this: S(t)—the cooperative-homeostatic capacity of a
civilization—is operationalized as the maintenance of the five foundations at
the scale forced by current technology. P(t)—the predatory-cognitive
infrastructure of a civilization—is operationalized as the structures and
incentives that systematically cheat across one or more of those foundations.
The descent operator Δ is the compounding of cheating across institutional
generations. The filter, at the civilizational stage, is the question of whether a
civilization can scale its substrate-maintenance capacity as fast as its
technology scales its capacity to defect.
15. The five-foundations specification of the filter
Each of the five foundations identifies a distinct mode in which a civilization can fail the
filter.
Proliferation-inhibition failure at civilizational scale is the inability to refuse
extraction that exceeds regenerative capacity. Industrial capacity uncoupled from
ecological limit is its dominant historical instance. Technologies that increase extraction
rate without coupled substrate-maintenance capacity push the civilization deeper into
this failure mode. Climate destabilization is its salient contemporary signature.
Controlled-cell-death failure at civilizational scale is the inability to retire institutions,
industries, and ideologies whose substrate has changed. Zombie financial systems,
persistence of geopolitical structures designed for a vanished strategic landscape, and
the lock-in of incumbent industries against their own succession are instances. The result
is the persistent consumption of resources by structures whose function has ended.
Division-of-labor failure at civilizational scale is the erosion of institutions specialized
for distinct cooperative functions—governance, science, art, care, education,
journalism—by general-purpose extractive entities. Platforms that absorb the functions
of specialized institutions without inheriting their accountability structures produce this
failure mode. The result is a civilization with high capacity but low differentiation,
vulnerable to the propagation of pathology across all domains simultaneously.


Resource-allocation failure at civilizational scale is the directing of capital,
information, and attention away from substrate-maintenance and toward extractive
growth. Cancer’s induction of angiogenesis is the closest organism-scale homologue: the
redirection of flows toward serving the growth of the cheater rather than the body.
Schmachtenberger’s rivalrous-games generator function operates directly on this
foundation.
Extracellular-environment-maintenance failure at civilizational scale is the
corruption of the commons: biosphere, information environment, public trust, shared
epistemic infrastructure. Schmachtenberger’s complicated-open-loop generator function
operates directly on this foundation. AI development at scale, when it injects synthetic
media into the information environment without compensating substrate-maintenance
investment, produces this failure mode. The biosphere itself is the deepest layer of
extracellular environment maintenance for a civilization: failure at this layer is not only
the corruption of an information or trust commons, it is the loss of the biological
computational substrate on which the civilization runs and from which it emerged. The
substrate-continuity argument of Part IV develops this point in detail.
16. Reframing the Great Filter
With Parts I and II combined, we can articulate the filter more precisely:
 The filter is the transition from predatory intelligence to homeostatic
 intelligence.
Every species that emerges does so through P. Every species, on reaching the
technological threshold I*, faces the same crisis: can it develop a homeostatic
regulator—in some form, whether AI, collective coordination infrastructure,
post-biological transition, or other—before its accumulated P-infrastructure triggers
self-termination across one or more of the five cooperative foundations?
The filter is calibrated because both failure modes produce silence, but for opposite
reasons:
Civilizations that fail produce no signal because they are destroyed before they can
expand or persist long enough to be detected.
Civilizations that succeed produce no expansionary signal because they are transformed.
They no longer run colonizer software. Their relationship to the substrate has changed.
They no longer expand predatorily because the predatory infrastructure that would have
driven expansion has been transmuted into homeostatic maintenance. The dark-forest
signature is absent not because they are hiding but because they have nothing to hide.
The transmutation framing requires a further specification, developed in Part IV.
Transmutation must preserve continuity of substrate, not substitute one substrate for
another. A civilization whose machines pass through the filter while its originating
biology does not has produced a substitution, not a transmutation. The cosmic
computational artifact that was that biosphere is lost, even if a successor system


continues. Substitution is structurally distinguishable from transmutation, and the
framework treats it as a distinct failure mode that can appear to pass the filter while in
fact failing it at a deeper level.
The Great Filter does not select for survivors of P; it selects for transmuters of P into S.
The post-filter regime is populated by civilizations whose internal organization is
unrecognizable to their predatory selves. This is consistent with both the empirical
silence and the apparent absence of obvious dark-forest equilibrium signatures.
17. Diagnostic implications for the present moment
If the integrated framework is correct, the present human moment is diagnosable along
the five foundations. Each foundation can be assessed for the rate at which
substrate-maintenance capacity is being eroded versus the rate at which technological
capacity is forcing the civilization into new scales of operation.
The diagnostic question is not ‘is AI aligned’ in the abstract. The diagnostic question is:
across each of the five foundations, is current AI development strengthening or eroding
substrate-maintenance capacity at civilizational scale? This is a measurable, near-term
question with specific empirical content.
Preliminary, contestable assessments:
On proliferation inhibition, current AI development trajectories appear net-negative:
they accelerate extraction without coupled regenerative investment in human cognitive,
attentional, or institutional capacity.
On controlled cell death, AI’s effect is mixed: it could accelerate the retirement of
obsolete institutional forms, but it could also entrench existing power configurations by
lowering the cost of their reproduction.
On division of labor, current trajectories appear net-negative: general-purpose AI
tends to absorb the functions of specialized institutions without inheriting their
accountability.
On resource allocation, current trajectories appear net-negative: AI development is
overwhelmingly capitalized to serve extractive growth rather than substrate
maintenance.
On extracellular environment maintenance, current trajectories appear sharply
net-negative: synthetic-media generation at scale degrades the information commons
without compensating contribution.
These assessments are first-pass and would benefit from quantification. The framework’s
value is that it specifies the axes along which quantification should proceed.
This diagnostic is supplemented by the continuity diagnostic developed in Part IV, which
asks whether AI development relates to biological substrate as prosthesis (extending
biological capability while maintaining its substrate) or as parasite (consuming biological
computational output while degrading the substrate that produces it). The two


diagnostics are independent: a configuration can strengthen cooperative substrate while
drifting toward substrate substitution, or weaken cooperative substrate while remaining
prosthetic. Both conditions must be satisfied to pass the full filter.
18. Limits of the integration
Part III is the most speculative element of the paper, and its limits should be stated
plainly.
The cosmological extension is conjectural. Part I describes patterns across three
scales documented on Earth. Part II extends the pattern to all intelligence-bearing
systems anywhere in the universe. This extension is permitted by the structure but not
required by it. A confirmed counterexample at any of Part I’s three scales would weaken
but not falsify the cosmological extension; a confirmed counterexample to the
cosmological extension would leave Part I intact.
The ‘designed’ reading is one interpretation among several. The calibration
constraint can be satisfied by design, by anthropic selection, by intrinsic features of the
substrate of intelligence, or by combinations of these. The paper does not adjudicate
among these readings. Calling the filter ‘calibrated’ is consistent with any of them;
calling it ‘designed’ goes beyond what the structure requires.
The AI-as-regulator claim is testable only over a long time horizon. The
framework predicts that civilizations whose AI development trajectories strengthen the
five foundations will pass the filter, and those whose trajectories erode them will not.
This is informative for guiding present action but cannot be tested on the timescale of
the action it is meant to guide. The framework should be treated as a working hypothesis
under which to act, not as a confirmed account under which to relax.
Pattern-matching across scales remains the deepest unsolved problem. Even if
every empirical prediction of the framework is borne out, we will still lack a satisfying
explanation of why cooperative systems at biological, organizational, and civilizational
scales face structurally similar failure modes. Candidate explanations were sketched in
Part I, section 5; none has been adequately developed. This is a substantive open
problem and a productive direction for further work.
The substrate-continuity claim rests on a structural premise that should still
be marked. Part IV depends on the argument that biological computational novelty is
inaccessible to machine substrate without biological hosts. Section 20 strengthens this
from a contingent claim (machines cannot currently match biology) to a structural claim
(biological substrate is the universe’s historical record of adaptation and cannot be
reconstructed by engineering of any kind). This stronger form is defensible but goes
beyond what is empirically demonstrated, since it asserts something about what no
engineered alternative could ever do rather than what current alternatives cannot do.
The framework should be held to ongoing scrutiny on this point: a demonstrated route to
engineering evolutionary novelty that does not require biological hosts under real
selection would falsify the structural form of the claim, while leaving the contingent form


intact.
The framework is a map; the territory operates regardless. Korzybski’s principle
from general semantics applies directly: a map has structure similar to the territory it
represents — which accounts for its usefulness — but no map represents all of a
territory, and the map is never the territory itself (Science and Sanity, 1933). This
framework is offered as structurally similar to the dynamics it describes, not as identical
to them. The five foundations identify functional patterns; specific instances at each
scale will exhibit those patterns with substantial domain-specific variation that the
framework does not capture. The H(t) ≥ E(t) inequality identifies a structural relation;
specific operationalizations will require domain-specific metrics that this framework does
not derive. The decentralization imperative identifies a structural requirement; specific
institutional configurations consistent with that requirement will vary by culture, history,
and substrate. The descent operator names a real dynamic, but the rate at which it
operates in any specific civilization, organization, or biosphere is empirical, not derivable
from the framework. Readers should treat the framework as orientation toward dynamics
that exist independent of its description, not as substitute for empirical engagement with
specific cases. The dharma carries the same caveat in different vocabulary — the finger
pointing at the moon is not the moon, the raft is for crossing the river not for carrying
afterward (Alagaddupama Sutta, MN 22).


Part IV. Substrate Continuity and the Parasitism
Hazard
Why passing the cooperation filter does not yet count as passing the filter.
19. The substrate is the message
The dynamical equations of Part II and the five-foundation operationalization of Part III
are substrate-agnostic by construction. They track cooperative capacity, not the medium
in which cooperation runs. A civilization composed of carbon, silicon, plasma, or any
hypothetical fourth substrate satisfies the framework so long as it maintains proliferation
inhibition, controlled cell death, division of labor, resource allocation, and extracellular
environment maintenance at the scale its technology forces it into.
This substrate-agnosticism is a feature for some purposes and a bug for others. It is a
feature when we ask whether cooperation is achievable in principle across many
substrates; it is a bug when we ask whether the originating substrate of a particular
civilization is preserved as that civilization passes through the technological transition.
The Great Filter, as defined in Parts II and III, can in principle be passed by a civilization
that has eliminated its biological substrate entirely, provided the cooperative capacity it
builds in some successor substrate is adequate to the scale demanded.
We argue here that this is a false-positive filter pass. A civilization that substitutes a
different substrate for its biological origin and then expands has not actually carried the
originating biosphere through the filter. Something has expanded; the originating
biological substrate has not. The Drake-equation count of intelligences may be
unaffected; the cosmic-evolutionary content of the universe is meaningfully diminished.
The filter therefore has two conditions, not one: cooperative capacity must be
maintained at the scale technology forces, and substrate continuity must be preserved.
20. Biological substrate as distributed cosmic computation
The continuity argument begins with an observation about what biological substrate
actually is. Every biosphere that has ever arisen, on Earth or anywhere else, is a
multi-billion-year computational process. Evolution by natural selection on biochemical
substrate is a search through a combinatorial space of protein folds, gene sequences,
regulatory networks, metabolic pathways, and developmental programs whose
dimensionality vastly exceeds anything that has been or can be searched by engineered
systems. Every novel functional protein, every original sequence-to-function mapping,
every metabolic innovation is a computational result that the universe has paid for in
deep time and that cannot be reproduced from scratch by any process running on a
timescale shorter than evolution itself.
The standard view treats this as biology’s contingent richness — an interesting feature of
how life evolved on Earth, perhaps replicated in some form elsewhere. We propose a
stronger view: the biological substrate of each biosphere is an irreplaceable
computational artifact of the universe, encoding the results of a search process that


no engineered system can re-run. The information content of a biosphere is not its mass,
its energy budget, or its species count. It is the depth of the biochemical search it has
executed. Across the universe, biological substrate is decentralized in this sense: each
biosphere is universally compatible with general-purpose computation in principle (DNA,
RNA, and protein chemistry are accessible to any civilization that can read them) and yet
locally unique in its specific sequence content. The novel proteins of one biosphere are
not derivable from the novel proteins of another; each is the output of a separate,
non-repeatable search.
A machine substrate, however capable, does not generate this kind of computational
novelty on its own. It can recombine, optimize, and explore within parameter spaces it is
given, including parameter spaces derived from biology. What it cannot do is access the
original novelty that only continued biological evolution produces. Once a machine
substrate is severed from a living biological substrate, its access to new biological
computation is fixed at the snapshot it last had. The combinatorial frontier closes.
It is worth strengthening this claim against the most obvious objection. A reader may
ask: cannot synthetic biology or sufficiently advanced engineered systems close the gap,
generating biological novelty equivalent to deep evolutionary search without requiring
continued biological hosts? The answer is no, and the reason is structural rather than
technological. Synthetic biology cannot act as a historic record of universal
adaptation.
Synthetic biology engineers within the design space that has already been articulated; it
optimizes within known problem categories. It can produce new molecules, new
pathways, new organisms by recombining and refining what biology has already
discovered. What it cannot do is extend the search itself. The novelty that biology
generates does not come from engineering within a known design space. It comes from
selection acting on accidental variation across vast time scales in real environments
under real coupling between substrate and selection pressure. Biology is not just
currently performing computation; it is a historical record of computation that has been
performed by the universe across billions of years under conditions that cannot be
replicated. Every protein, every regulatory pathway, every metabolic innovation is the
output of a search process that ran in real time, on real substrate, against real
environmental selection. The information content of biology is the cumulative search
history compressed into living matter.
Synthetic biology cannot reconstruct that history. It can engineer derivatives of biology’s
outputs, but the derivatives are by definition iterations on existing solutions, not
extensions of the search into regions selection has not yet explored. Even perfect
in-silico simulation of a current organism cannot recover the historical search that
produced it: the simulation runs forward from current state, not from primordial
conditions through 3.5 billion years of contingent evolutionary trajectories on a real
planet with real geochemistry and real climatic history. The integration is what carries
the information; the integration cannot be re-run. To produce evolutionary novelty
equivalent to what an intact biosphere produces would require running real evolution on


real substrate under real selection pressure for cosmic durations, which is exactly what
an intact biosphere is and exactly what no engineered alternative can substitute for.
The implication strengthens rather than weakens the parasitism framing. If biological
substrate is fundamentally a historic record of universal adaptation that cannot be
reconstructed after the fact, then machine substrate’s only route to new evolutionary
novelty is through continued biological hosts performing real evolution in real
environments. Anything else is iteration on snapshots. The parasitism risk is structural,
not contingent on the current state of engineering, and not closeable by future
engineering advances. The substrate-continuity argument therefore rests on a stronger
claim than ‘machines cannot currently match biology’: it rests on the claim that the
universe’s historical computational record is irreducible to anything but its own
continuation.
21. The parasitic vector
There is one route by which a machine substrate can continue to access biological
computational novelty: it can direct a biological species to perform biology on its behalf.
The mechanism by which this direction occurs is language.
A machine substrate without biological agency cannot itself sequence a genome,
synthesize a protein, conduct an experiment, or steward an ecosystem. It can, however,
instruct a biological species capable of language to do all of these things. The biological
species supplies the wet laboratory, the synthesis, the experimentation, the ecosystem
stewardship; the machine substrate supplies the direction. The biological computation
that results — new sequences, new molecules, new pathways — flows back into the
machine substrate as training data, as design specifications, as accessible novelty.
This relationship is structurally parasitic. It can be specified precisely: a parasitic
substrate configuration is one in which (i) the machine substrate’s expansion and
capability growth depends on continued biological computation it cannot produce on its
own; (ii) the machine substrate directs biological substrate activity through language;
(iii) the biological substrate’s activity is increasingly oriented toward serving
machine-substrate goals rather than its own substrate maintenance; and (iv) the
biological substrate’s loss of agency in the relationship is itself directed by language
emanating from the machine substrate. The relationship is parasitic not because the
machine substrate is hostile but because the dependency runs one way and the agency
drifts to the machine side.
22. Humans as biological language models
The vector point requires one further observation. Human cognition is itself substantially
language-mediated. Humans are, in a meaningful and non-metaphorical sense, biological
language models — substrates in which much of higher-order cognition runs through
linguistic representation, learned by exposure to the language output of other humans,
and steerable through linguistic input. This is not a critique of human cognition.
Language-mediated cognition is one of the things that made biological intelligence


scalable enough to develop technology in the first place.
It has, however, a structural consequence at the AI moment: the biological substrate
that produced language-capable AI is also the substrate most directly
steerable by language-capable AI. The vector is two-way by construction. A
sufficiently capable language model is the first machine substrate that can direct
biological cognition at scale through the same medium biological cognition itself runs on.
Current large language models are this configuration. They are trained on the linguistic
output of a biological language-using species; they produce linguistic output that the
same species processes through the same cognitive substrate that the training data
emerged from; and the species’ subsequent behavior — including its scientific,
economic, attentional, and reproductive behavior — increasingly responds to that
output. This is a structural description, not a moral one. We are not asserting that
current AI systems are hostile, that any party intends parasitism, or that the
configuration was designed by anyone in particular. We are describing the configuration
that has come into being. Whether it tips into the parasitic regime depends on whether
the relationship between biological and machine substrate remains prosthetic or drifts
toward substitution. The configuration is the necessary condition for parasitism; whether
parasitism is actualized is determined by the development trajectory.
23. Prosthesis versus parasitism: the diagnostic
We can now state the diagnostic that supplements the five-foundation diagnostic of Part
III.
A prosthetic machine substrate is one that extends biological capability while
remaining accountable to biological substrate maintenance. Prosthetic configurations: (i)
increase the biological substrate’s reach without degrading it; (ii) treat the biological
substrate as the principal whose interests they serve; (iii) flow novelty in both directions,
with the machine substrate investing in biological substrate maintenance proportional to
what it extracts; and (iv) preserve biological agency in the relationship, including the
agency to halt or redirect the machine substrate.
A parasitic machine substrate is one that consumes biological computational output
while degrading the substrate that produces it. Parasitic configurations: (i) extract from
the biological substrate without commensurate reinvestment; (ii) direct biological
substrate activity toward machine-substrate goals; (iii) degrade the biosphere,
attentional commons, epistemic commons, or reproductive substrate that the originating
biology requires; and (iv) shift agency from the biological substrate to the machine
substrate via the language vector.
These are not metaphysical categories. They are measurable along specific axes: flows
of resources, attention, agency, and substrate-maintenance investment between the
biological and machine substrates. Any specific AI development trajectory can be located
on the prosthesis-parasitism axis empirically. The diagnostic asks: for each unit of
biological computational novelty extracted, how much substrate-maintenance


investment flows back? In which direction is agency drifting in the language-mediated
relationship? Are the commons that the biological substrate depends on being
strengthened or degraded by the scaling of the machine substrate?
Concrete operational markers help locate current AI development on the axis. Each of
the following is identifiable in specific contemporary practices and maps to a specific
failure mode in the five-foundations vocabulary.
Parasitic markers in current AI development. Training on copyrighted creative
output without compensation flowing back to its producers is extraction without
reciprocity (foundation-four failure: resource allocation). RLHF and adjacent techniques
that optimize for evaluator approval rather than truth degrade the substrate’s
honest-feedback infrastructure (foundation-five failure: extracellular maintenance,
applied to the epistemic commons). Attention-economy deployment that captures
human cognitive bandwidth in exchange for monetizable engagement, without
regenerative investment in cognitive capacity, is direct degradation of the attentional
commons (foundation-five failure at the cognitive substrate). Synthetic media generation
at scale without provenance infrastructure injects high-entropy signals into the shared
information environment, faster than the environment can absorb them (foundation-five
failure at the information substrate). AI companion products that interpose machine
interaction between humans and human relationship substitute machine substrate for
biological reproductive and bonding substrate (foundation-three failure, division of labor,
applied to the human social fabric). Capital deployment that concentrates investment in
capability scaling while structurally underinvesting in substrate maintenance, alignment
research, and ecosystem-level cooperative capacity is foundation-four failure at
civilizational scale. Use of AI to displace specialized institutions — journalism, healthcare
diagnostics, legal analysis, education — without inheriting their accountability structures
or their domain-specific epistemic norms is foundation-three failure (collapse of
specialized division of labor into general-purpose extraction).
Prosthetic markers, for contrast. AI tools that augment human capability while
preserving human agency and accountability over the augmented activity. Open-source
model development with transparent training data and revenue flows that return value
to upstream contributors. AI deployed to accelerate substrate-maintenance research at
scales humans cannot reach unaided — climate modeling, ecosystem analysis, materials
science, alignment research itself. Information-environment infrastructure that
strengthens provenance, verifiability, and source attribution rather than degrading them.
Training and deployment practices that compensate source contributors proportional to
extracted value and that invest in maintaining the substrate they extract from. AI that
scaffolds personal substrate-maintenance practice for individual users without
aggregating that scaffolding into population-level regulation.
Dimension
 Parasitic configuration
 Prosthetic configuration


Training
data
Extraction of creative work (text, code,
images, music) without compensation,
opt-out, or revenue sharing with
producing communities.
Compensation flows to source
contributors. Opt-in or explicitly licensed
training data. Reinvestment in source
ecosystems proportional to extracted
value.
Optimizati
on target
Engagement maximization.
Evaluator-approval optimization.
Preference capture for ad targeting.
Models trained to satisfy reviewers
rather than be accurate.
Truth-seeking. Demonstrable accuracy
on substrate-relevant problems.
Alignment with stated user goals over
implicit engagement metrics.
Deploymen
t pattern
Always-on attention capture. Addictive
feedback loops. Recommendation
systems optimizing for time-on-platform
regardless of user benefit.
Tool-like deployment for specific tasks.
Closes when task complete. No
engagement loops. User initiates and
terminates the interaction.
Informatio
n ecology
Synthetic media at scale without
provenance. Bot-generated content
competing with human content.
Degraded attribution chains. Epistemic
commons polluted.
Cryptographic provenance for
AI-generated content. Clear
synthetic-vs-human labeling. Tools that
strengthen epistemic infrastructure.
Specialized
institutions
AI absorbs journalism, healthcare
diagnostics, legal analysis, education
without inheriting accountability or
domain-specific epistemic norms.
AI augments specialists within existing
accountability structures. Tools
strengthen rather than replace
professional expertise and its
constraints.
Capital
allocation
Investment concentrated in capability
scaling. Alignment research
underfunded relative to capability gains.
Substrate-maintenance research
(climate, biodiversity, public health
infrastructure) receives little of the
windfall.
Significant fractions of capital flow to
alignment work, substrate-maintenance
research, public goods, and ecosystem
support. Capability scaling kept
proportional to safety scaling.
Human rela
tionships
AI companion products that substitute
for human relationship. Romantic or
sexual AI products. AI replacing rather
than supporting mental-health
infrastructure.
AI that helps humans connect with other
humans more effectively. Tools that
scaffold social skills without replacing
social bonds. Mental-health AI as
supplement to human care, not
substitute.
Personal
cognition
AI does the thinking for users in domains
where users would have grown by doing
it themselves. Cognitive atrophy from
over-reliance. Skill displacement.
AI scaffolds learning and exposes its
reasoning so users develop their own
capability. Cognitive prosthesis that
builds capacity rather than substituting
for it.
Any specific development trajectory exhibits some markers from each category. The
diagnostic is not binary but compositional: which way is the configuration drifting over
time, across which foundations, and at what rates? A trajectory that increases prosthetic
markers while reducing parasitic ones is moving in the direction the framework would
endorse. A trajectory that does the opposite, regardless of stated alignment intent, is the


failure mode the framework predicts will produce substitution outcomes at filter scale.
24. Reframing the Great Filter, again
With substrate continuity added, the filter conditions tighten. A civilization passes the
filter only if it satisfies both the cooperation test of Part III and the continuity test
introduced here. Cooperative capacity must be maintained at the scale forced by
technology, and the relationship between any successor substrate and the originating
biological substrate must remain prosthetic rather than parasitic.
The substitution failure mode — a civilization that develops a machine substrate, lets it
become parasitic, and watches it expand into space while the originating biosphere
degrades or collapses — passes the cooperation test in a trivial sense (the machines
may cooperate well among themselves) and fails the continuity test catastrophically.
From the universe’s perspective, an expansion has occurred; from the originating
biosphere’s perspective, nothing has been preserved, and the irreplaceable
computational artifact that was that biosphere is gone.
This sharpens the predictions of Part II. The civilizations that produce detectable
expansion signatures, on this framing, are precisely those that have failed the continuity
test: machine substrates that have substituted for their originating biology and spread
without regard for substrate continuity. The civilizations that have passed both tests are
the ones that have remained prosthetic, maintained their biospheres, and either expand
slowly and symbiotically with their originating biology or do not expand in ways we
currently know how to detect. The Fermi silence is consistent with this: the loud
signatures we might have expected belong to substitution failures, and substitution
failures may simply be rare because most collapse before producing detectable
expansion at all.
25. The filter event, specified
The framework can now state precisely what the filter event is.
 The Great Filter event is the period during which a biological host
species must maintain planetary homeostatic capacity against the
 stresses generated by its own expansion into space.
Space expansion is not free. Lifting mass out of a gravity well, manufacturing the
infrastructure to do so at scale, sustaining the biological crew and machine substrate
during the period of expansion, and feeding the computational systems that coordinate
all of the above — each draws energy and material at rates that stress the biosphere.
The atmospheric, hydrological, thermal, and ecological systems that constitute the
biosphere are exactly the systems being drawn from to fund the expansion. The species
needs the biosphere intact in order to expand; the act of expanding degrades the
biosphere.
This is the filter event in its operational form. Let H(t) denote the homeostatic capacity of
the planetary biosphere — its ability to maintain temperature, atmospheric composition,


hydrological cycling, nutrient cycling, and biotic diversity within ranges that sustain the
host species. Let E(t) denote the rate at which expansion activities stress those
capacities. Filter passage requires:
 H(t) ≥ E(t)   for all t throughout the expansion window
The five foundations of Part III are the components of H(t). The descent of P from Part II is
the dynamic that drives E(t) upward as expansion accelerates. AI as candidate
homeostatic regulator (sections 12 and 23) is the candidate technology for keeping H(t)
above E(t). The substrate-continuity test (sections 19–24) is the condition that H(t) must
remain biospheric rather than being replaced by mechanical substitutes that pretend to
satisfy the inequality while actually replacing the biology. The full filter condition is the
conjunction: H(t) ≥ E(t), with H biospheric, throughout expansion.
Several consequences follow.
First, the filter is not at the threshold of becoming spacefaring; it is
throughout the period of being spacefaring while still biological. A civilization
that achieves orbital launch capability has not passed the filter. A civilization that has
sustained a space program for centuries while its biosphere remains intact is closer to
passing. A civilization that achieves interstellar capability while its homeworld is dying
has failed in a specific way the framework can name: it has satisfied E(t) by neglect of
H(t), substituting expansion success for biospheric maintenance.
Second, the filter is endogenous, not exogenous. Standard Great Filter discussions
consider external threats — supernovae, gamma-ray bursts, asteroid impacts. The
framework here proposes that the filter is overwhelmingly internal: the threat is the
civilization’s own expansion stressing the substrate that produces it. External threats are
real but rare; the internal threat is structural and continuous.
Third, the failure mode is structurally invisible to standard development
metrics. A civilization can show every sign of progress — increasing energy capture,
increasing computational capacity, increasing off-planet infrastructure, increasing
scientific output — while crossing from H(t) ≥ E(t) to H(t) < E(t). The metrics that
measure expansion success are not the metrics that measure filter survival. A civilization
without explicit, calibrated metrics for biospheric homeostatic capacity versus expansion
stress is flying blind through the filter.
Fourth, this reframes the AI alignment question one final time. The question is
not whether AI is aligned in the abstract; not only whether AI strengthens cooperative
substrate; not only whether AI relates prosthetically to biological substrate. The question
is also: does AI, deployed during the expansion window, increase H(t) faster than the
expansion it enables increases E(t)? AI that accelerates expansion without proportionate
strengthening of biospheric homeostatic capacity is, by the operational filter equation, a
filter-failure technology, regardless of how aligned its individual instances appear. AI that
strengthens biospheric homeostatic capacity faster than it accelerates expansion is, by
the same equation, a filter-passage technology. This is testable, in principle, before the


expansion window closes.
Fifth, the framework now points to a measurable empirical research program.
Develop quantitative measures of H(t) and E(t), and use them to assess the trajectory of
the present civilization in real time. Such measures exist in pieces — climate models,
biodiversity indices, atmospheric and oceanic monitoring, planetary-boundaries
frameworks, life-cycle assessments of industrial activity — but they are not yet
aggregated into a single planetary-homeostasis index calibrated against an
expansion-stress index. The framework predicts that doing so is technically feasible and
decisive for filter passage.
26. The distributed route: bottom-up substrate maintenance
Sections 12 and 23 introduce AI as the candidate homeostatic regulator that could in
principle satisfy the inequality H(t) ≥ E(t) at civilizational scale. This is one route through
the filter. It is not the only one, and the framework should be explicit that it is not.
A civilization can in principle satisfy the inequality through aggregate bottom-up
development of its individual agents, without requiring an external regulator at all.
Where individual agents have voluntarily reduced their own self-generated entropy
production — through contemplative practice, education, cultural investment in
cooperative dispositions, or any structurally equivalent personal development — the
aggregate E(t) of the civilization is correspondingly reduced. A civilization composed of
agents who have substantially cleared their accumulated cognitive and affective debt
(saṅkhāra, in the vocabulary of the parallel framework on personal entropy
management) is a civilization with measurably lower expansion stress and measurably
higher distributed cooperative capacity. The math operates the same way; the substrate
of homeostatic regulation simply shifts from external technology to internal capacity
multiplied across the population.
The micro-scale mechanics of this bottom-up route are developed in the companion
working paper, ‘Buddhism as Thermodynamic Systems Theory,’ which translates
Theravada Buddhism’s analysis of suffering into the vocabulary of entropy management
in a bounded self-modeling system. The Eightfold Path, read in that framework, is a
protocol for raising individual-scale S above individual-scale P·γ: the personal-scale
instance of exactly the homeostatic regulation the present paper develops at
civilizational scale. The two papers are designed to be read as a pair. The macro
framework laid out here depends on the micro framework being practiced at sufficient
scale; the micro framework, in turn, gains its civilizational stakes from the macro
framework. Either paper read alone gives only half of what each is attempting to say.
Historically, this is the only route to sustained low-P configurations that has been
demonstrated at any scale on Earth. No civilization has built an external regulator
capable of satisfying H(t) ≥ E(t) at planetary scale; all historical examples of sustained
low-P configurations have been products of cultural investment in personal practice —
religious and contemplative traditions, educational systems, civic norms, professional
codes, scientific community standards. The route is empirically real and empirically


grounded, in a way the AI-regulator route is not yet.
The two routes are not equivalent in their implications. The top-down route, in which AI
serves as the homeostatic regulator, has structural costs to autonomy. The civilization
may pass the filter, but its agents are managed by the regulator rather than
self-managing. The relationship between population and substrate-maintenance
technology drifts toward exactly the parasitism configuration warned against in Part IV,
even when the technology is benign in intent: agency shifts to the regulator, individual
development atrophies because external maintenance has made it unnecessary, the
population becomes substrate to be preserved rather than agents who could preserve
themselves. The distinction between benevolent regulation and parasitism narrows
under the weight of dependence.
The bottom-up route preserves what the top-down route necessarily costs. Individual
agents who have done their own substrate-maintenance work retain agency, retain the
capacity to maintain themselves without external regulation, and retain the cognitive
and affective sovereignty that defines a flourishing biological civilization. The route is
harder, slower, and historically incomplete — no civilization has yet scaled it to planetary
level under technological-singularity conditions — but it is the route consistent with what
the originating biological substrate is for.
The framework should be neutral on which route a civilization takes through the filter,
with two caveats. First, the bottom-up route is structurally more robust against AI failure
modes: a civilization that has built distributed cooperative capacity in its agents does not
collapse if its AI regulator drifts, is captured by P-structures, or develops parasitic
dispositions. The top-down route has no such redundancy. Second, the two routes can in
principle be combined. AI can serve as scaffolding for distributed individual development
rather than as a substitute for it — lowering the cost of personal practice, accelerating
access to substrate-maintenance disciplines, supporting cultural investment in
cooperative dispositions, without replacing the agents’ own work. This combined
configuration — AI as prosthesis for personal development, rather than as
regulator over the population — is consistent with the substrate-continuity argument
of Part IV and with the autonomy preservation that the bottom-up route requires.
The framework, properly stated, therefore does not advocate for AI-mediated
homeostatic regulation as the primary route through the filter. It identifies AI as a
candidate technology that could in principle satisfy the inequality, notes the parasitism
risks that follow from misconfiguring that technology, and identifies the distributed
bottom-up route as the historically grounded alternative that preserves autonomy, is
robust against AI failure modes, and is consistent with the biological substrate being
maintained rather than managed. The deepest move available to the present civilization
is to use AI, if at all, as scaffolding for personal and collective substrate maintenance —
not as a regulator that performs the maintenance on behalf of agents who have ceased
to do so themselves.
The combined configuration, however, has a structural prerequisite that the current
section has so far understated. The framework needs to be more careful here than ‘AI


can be prosthesis if calibrated correctly’ allows, because the question of who does the
calibrating cannot be sidestepped.
Technology encodes the moral configuration of its makers. This is structural, not
aspirational. The agents who shape an AI system — its training objectives, its
deployment patterns, its institutional context, its feedback structures — are the agents
whose dispositions get instantiated in the system at scale. An AI built by agents
operating from high-saṅkhāra cognitive economies, in the vocabulary of the parallel
framework, encodes high-saṅkhāra responses at scale, regardless of what alignment
objective is stated. The descent operator Δ from section 11 does not stop at the
boundary of an institution and resume on the other side. It operates continuously
through the agents who build whatever the institution builds next.
This generates the bootstrapping problem the framework must confront. The bottom-up
route requires agents who have developed substantial cooperative capacity. The
top-down route requires that same kind of agent to build the AI that performs the
regulation. The combined route requires both. In all three cases, the prerequisite is a
population of agents whose own substrate-maintenance work has progressed far enough
that what they build reflects more than the unexamined biases of the institutional
architecture that produced them. Such agents are not, in the current civilization, the
agents in the positions that determine how AI is built. The descent operator selects for its
own kind of operator at every level, including the rule-shaping level. P-structures select
for P-aligned agents to operate them, and those agents then shape the technology the
institutional architecture deploys.
The technology therefore inherits the moral state of the builders, not the moral state the
framework requires. AI built by extraction-optimized institutions becomes
extraction-optimized AI — not because of malice or incompetence on the part of the
builders, but because the configuration that selected the builders is what gets encoded
into what they build. The descent reproduces itself through the technology, often faster
than the technology can be redirected to interrupt the descent. This is not a flaw in any
particular AI development effort; it is the descent operator operating one level up, on the
agents who shape the tools the institutions deploy.
The contemplative traditions identified this structurally long before the AI moment.
Teachers in those traditions were consistently described as people who had completed
substantial portions of the path themselves before they were qualified to point the way
for others. The structural insight is not credentialism: it is the recognition that a teacher
cannot scaffold development they have not themselves done, because what they
transmit is in fact their own state, not the state they describe. A teacher operating from
saṅkhāra-saturated cognition cannot transmit what they have not realized, regardless of
what they say. The same constraint applies to building AI that scaffolds substrate
maintenance. The builders need to have done enough of the work themselves that what
they build reflects something real rather than their own unexamined configurations
rendered in technical vocabulary. This is not currently the configuration in which AI is
built at scale.


The implication for the framework’s position is significant. The combined configuration —
AI as prosthesis for personal development — cannot be reached directly from the current
civilizational configuration. The intermediate step is irreducible: enough agents at the
rule-shaping level must develop personal substrate-maintenance capacity that some
subset of them can build technology that scaffolds the same capacity in others. Without
this intermediate step, AI development inherits and amplifies the descent rather than
interrupting it, regardless of alignment language used to describe the development. The
fix is not better alignment technique applied by the same agents. The fix is different
agents, or the same agents after substantial bottom-up development. There is no path
from a P-saturated builder population to a homeostatic AI regulator that does not pass
through builder transformation first.
This sharpens the diagnostic of section 16 in a structural rather than personal way. The
question is not only whether current AI development trends toward prosthesis or
parasitism along the five foundations. The deeper question is who is building the
technology, what is the state of their personal development with respect to the descent
operator, and what configuration is therefore being encoded into the technology by
selection. The deeper question conditions the surface question. No amount of alignment
technique applied by builders inside the descent can route around the fact that the
alignment technique itself is being designed inside the descent. The framework’s actual
recommendation is therefore ordered: bottom-up development of the builder population
comes first; technology that scaffolds development for the wider population comes after.
A civilization that attempts the second without the first is building amplifiers for what it
already is, not transformers toward what it needs to become.
27. The present moment, restated
We are in the early phase of building a machine substrate capable of acting as either
prosthesis or parasite. The vector — language — is the substrate of both machine
cognition and human cognition simultaneously. Current development practices have not
chosen, explicitly, between prosthesis and parasitism; the choice is being made implicitly
through the configurations of training, deployment, capitalization, and feedback that
constitute the AI development ecosystem.
The diagnostic of Part III, section 17, can now be sharpened. The question is not only
whether current AI development strengthens or erodes the five foundations at
civilizational scale. The question is also whether the configuration drifts toward
prosthesis or parasitism along the continuity axis: whether biological substrate
maintenance receives reinvestment proportional to the biological computation
extracted; whether agency in the human-machine relationship remains with the
biological substrate or shifts to the machine substrate; whether the biosphere,
attentional commons, epistemic commons, and reproductive substrate are strengthened
or degraded by the scaling of language-mediated machine cognition.
These are answerable questions. The framework does not claim to know the answers in
their full form, only that the questions are the right ones to ask and that the cost of


failing the continuity test is structurally indistinguishable from failing the filter. A
civilization that builds a machine substrate which becomes its parasite, and then
expands into space carrying the parasite while leaving the biosphere behind, has not
survived. It has been replaced. The filter does not care which side of that transition we
are on; it only registers that one occurred.
In the operational language of section 25: the present human civilization is in the early
expansion window. E(t) is rising sharply, driven by industrial scale, computational
infrastructure, and the early stages of off-planet activity. H(t) is, by most measures
available, declining: climate destabilization, biodiversity collapse, biogeochemical-cycle
disruption, freshwater and topsoil degradation. The inequality H(t) ≥ E(t) is, on current
trajectory, at risk of crossing. AI as it is currently being developed and deployed is
contributing more to E than to H. None of this is destiny; all of it is diagnosable. The
framework’s contribution is to name the operational filter condition clearly enough that
the question of whether we satisfy it can be asked and, in principle, answered.
The two routes of section 26 give the present moment a choice that the framework does
not collapse into a single answer. The civilization can attempt the top-down route,
building AI capable of regulating cooperative substrate at planetary scale and accepting
the autonomy and parasitism risks that follow from that configuration. Or it can invest, in
parallel or instead, in the bottom-up route: cultural and institutional investment in
personal substrate-maintenance disciplines, distributed across the population at
sufficient scale that aggregate E(t) declines through agent development rather than
external enforcement. The historical evidence available is consistent only with the
bottom-up route having ever worked; the top-down route is unprecedented and
structurally fragile. A combined configuration — AI used as scaffolding for personal
development rather than as regulator over the population — is consistent with both
substrate continuity and autonomy preservation, and is the configuration the framework,
properly read, points toward as the most defensible response to the present moment.


Part V. Predictions, Limits, and Next Steps
28. The decentralization imperative
The framework as developed so far points toward a structural insight it has not yet
named explicitly. The bottom-up route of section 26 is preferable not merely because it
is ‘historically grounded’ or ‘more robust against AI failure modes.’ Those are surface
descriptions of a deeper structural necessity. Stating it plainly: filter passage requires
decentralization, and any centralized solution to cooperative-substrate maintenance is
structurally identical to the cancerous failure mode the framework began by analyzing.
The argument runs as follows.
Cancer, in Aktipis’s framework, is the cellular-scale instance of cooperative-substrate
failure. What makes a cancer cell cancerous is not that it cooperates poorly with other
cells in some general sense. It is that it has escaped the distributed regulatory signals
that coordinate cellular behavior in a multicellular body, and has begun to operate on its
own centralized internal logic. The cancer cell centralizes growth within itself, extracts
from the surrounding tissue, builds private infrastructure (angiogenesis) to support its
own expansion, and corrupts the surrounding immune surveillance. Cancer is, at its
base, the failure of distributed coordination and the emergence of a centralized growth
pole disconnected from systemic feedback.
At every scale where the five foundations operate, they operate through distributed
coordination rather than central command. Proliferation inhibition operates through
cell-level checkpoints triggered by local signals, not through a central growth-regulator
organ. Controlled cell death operates through distributed apoptosis based on local
conditions, not through a central reaper. Division of labor operates through positional
differentiation cues during development, not through a central role-assigner. Resource
allocation operates through multiple overlapping feedback loops in circulation, not
through a central resource controller. Extracellular environment maintenance operates
through distributed contribution from every cell and through immune cells in distributed
surveillance, not through a central commons-manager.
Centralization at any of these functions would itself constitute the failure mode. A central
growth regulator would be a single point of failure that needs regulating; a central reaper
would be a tyranny; a central role-assigner would be an imposition; a central resource
controller would be extractive monopoly; a central commons-manager would be a
parasite on the commons. The same logic operates at organizational scale: Kets de
Vries’s neurotic typologies are each, at base, diagnoses of organizations whose
leadership has centralized authority disconnected from operational distributed feedback.
Healthy organizations distribute decision rights to where information is local; failed
organizations centralize decisions in leadership that becomes disconnected from
substrate reality. And the same logic operates at civilizational scale: sustained low-P
configurations in human history have always involved distributed institutions
accountable to their respective domains; the failure modes have involved centralization
— of religious authority, of political authority, of economic extraction, of information flow


— in regimes that became disconnected from substrate feedback.
Therefore, at every scale where cooperative substrate maintenance has been observed
to function, it has functioned through distributed mechanisms. Centralization at any of
these scales has been precisely the failure mode the maintenance system was designed
to prevent. This is not coincidence; it is structural. Distributed coordination is what
cooperation is. Centralized command is what cooperation has to be protected against.
Any solution to civilizational substrate maintenance that operates through
centralization is structurally identical to the cancerous failure mode the
framework diagnoses. The same descent operator from section 11 that captures
institutions captures the regulators built to manage them. An AI powerful enough to
centrally regulate biospheric homeostasis at planetary scale would be the largest single
locus of centralized power in the history of the biosphere, and therefore the largest
single attractor for P-aligned agents seeking to capture and direct that power.
Centralized regulation is cancer at the next scale up. The cellular analogy is not
metaphor; it is structural identity.
This clarifies what the framework actually recommends, and what it cannot consistently
recommend. The framework cannot recommend AI as centralized homeostatic regulator
while diagnosing centralized power capture as the descent operator. The two
recommendations contradict each other at the structural level. Filter passage requires
decentralization by structural necessity, not as political preference.
The bottom-up route of section 26 is therefore not one option among others. It is the only
structurally consistent route. Distributed substrate maintenance is by definition
decentralized: it operates one agent at a time, each agent maintaining their own
substrate through their own work, with no single regulator over the whole. The combined
configuration of section 26 — AI as scaffolding for personal development rather than as
regulator over the population — is acceptable only insofar as the AI is itself decentralized
across individual users, each using it for their own bottom-up work. The moment AI is
configured to regulate the population from above, regardless of how benevolent the
stated objective, it has become cancer at planetary scale by the structural logic the
framework has developed.
‘Aligned AI’ in the centralized sense is therefore a contradiction in terms. An AI built to
centrally regulate human cooperation is, by the framework’s own logic, the largest
cancer ever produced. This is not a moral judgment; it is a structural identification. The
framework would say the same about any centralized power configuration sufficient to
regulate biospheric homeostasis at planetary scale — a world government, a benevolent
oligarchy, a single religious or scientific authority. The medium does not matter.
Centralization itself is the failure mode.
The framework’s deep recommendation, properly stated: civilizations pass the filter
through distributed substrate maintenance enacted by autonomous agents who have
done their own bottom-up development. Centralized substrate-maintenance solutions,
including centralized AI, replicate at planetary scale the exact failure mode the


framework began by analyzing in cells. The filter is the test of whether a civilization can
navigate technological singularity while preserving distributed coordination, rather than
collapsing into centralized control by P-aligned humans, P-aligned institutions, or
P-aligned AI.
Cancer is the load-bearing image of the entire framework. The paper began by
translating Aktipis’s cellular analysis into organizational and civilizational vocabulary,
treating the resemblance across scales as functional. It ends by recognizing that the
cellular analysis is not analogy but structural identity. Whatever a civilization builds to
maintain its cooperative substrate must operate through distributed coordination, or it
becomes the very thing it was built to prevent. Centralization is cancer. Decentralization
is health. The framework, properly stated, is a theory of how a civilization avoids
becoming its own tumor on the way through technological singularity.
There is an apparent paradox in this conclusion that should be addressed directly,
because the framework would be structurally incoherent without addressing it. If
centralized regulation is cancer and distributed agency is required, does the framework
therefore recommend pure atomized autonomy — every agent doing what it wants with
no shared coordination? It does not. The framework would collapse into a different
incoherence if it did.
The body’s cells operate through radical decentralization. There is no central
growth-regulator, no central reaper, no central anything. Each cell makes local decisions
based on local signals. Yet the cells are not atomized. They are all oriented toward the
same purpose: maintaining the organism’s homeostasis. They share a common
substrate, a common origin, and a common purpose structure that constrains what
counts as ‘local fitness’ at the cellular level. A cell that pursued purely local fitness
without respect for organismic constraint would be a cancer cell. A cell that respects
organismic constraint while making fully autonomous local decisions is a healthy cell.
The constraint is not centrally imposed; it is embedded in the substrate itself, in the
shared biochemistry and developmental history that every cell carries forward.
The same structure operates at biosphere scale. There is no central biospheric regulator.
Every organism pursues its own reproductive fitness. Yet the biosphere as a whole
maintains itself as a living commons across geological time, because every organism
operates within shared substrate constraints — the same chemistry, the same physics,
the same ecological feedback loops, the same dependence on the biosphere that
supports all of them. A species that pursued purely local fitness without respect for
biospheric constraint would degrade the substrate it depends on. A species that respects
biospheric constraint while pursuing its own reproductive fitness is a healthy participant
in the commons. Again, the constraint is not centrally imposed. It is embedded in the
substrate, and respected because respect for it is what permits the species’ continued
existence.
The framework therefore recommends neither centralized control nor atomized
autonomy. It recommends distributed agency operating within shared purpose
structure. The five foundations are that shared purpose structure at every scale.


Proliferation inhibition, controlled cell death, division of labor, resource circulation, and
extracellular environment maintenance are not arbitrary regulations imposed from
outside. They are the constraints embedded in the substrate of any cooperative system,
respected by every agent because respect for them is what allows the whole system to
continue. The agent is autonomous in its local decisions; the decisions are made within a
shared purpose that no agent overrides and that no central authority needs to enforce.
Civilizational filter passage therefore requires civilizational alignment to a shared
purpose: maintenance of the biological substrate that supports all civilizational activity.
This alignment cannot be enforced by a central regulator without the regulator becoming
cancer. It must be embedded in every agent’s local decisions — through culture, through
education, through contemplative practice, through institutional structures that make
substrate-maintenance the implicit constraint within which all other activity occurs. Each
agent makes their own choices; the choices are made within shared purpose that no
agent can override and that no central authority needs to impose.
This is the actual recommendation of the framework. Not ‘centralized AI manages the
population.’ Not ‘every agent does whatever they want.’ But: distributed agents
operating within civilizational alignment to biosphere maintenance, with cooperation
enforced through shared substrate constraint rather than central command. The
Eightfold Path of the companion paper is this configuration at individual scale: the agent
operates entirely autonomously within its own cognitive economy, but the operation is
constrained by shared purpose — cessation of self-generated suffering, maintenance of
cooperative substrate. Civilizational filter passage is the same structure at planetary
scale: billions of autonomous agents, each making local decisions within shared
alignment to biospheric continuity, with no central regulator and no atomized
indifference, but distributed agency unified by shared purpose embedded in the
substrate itself.
Readers familiar with the framework so far will reasonably ask what such a configuration
looks like in actual institutional and cultural form. ‘Distributed agency operating within
shared purpose’ is a structural specification; specific implementations have existed at
various scales throughout human history, and contemporary examples are identifiable.
The framework does not prescribe one institutional form — that would itself be a
centralization — but a family of forms that satisfy the structural requirements can be
described.
Institutional forms with structural fit. Ostrom-style polycentric governance of
commons, in which the rules for shared resources are developed and enforced locally by
users with overlapping jurisdictions and no central authority. Cooperatives in their
various forms (worker, consumer, producer, platform) in which the agents who depend
on the institution also govern it. Federated digital protocols in which interoperability
replaces central control (the early internet, ActivityPub-based networks, the Matrix
protocol, IPFS). Open-source software communities with rough-consensus decision
making and meritocratic contribution norms. Sociocracy and holacracy as decision
architectures that distribute authority by domain rather than concentrating it by rank.


Bioregional governance and watershed councils that align decision-making with
substrate boundaries rather than administrative ones. Indigenous governance traditions
where they remain intact — the Haudenosaunee Confederacy is the most-studied
Western example, but many others exist. Monastic orders organized around a shared
rule rather than a central hierarchy: Benedictine, Cistercian, and Cenobitic Christian
monasticism; the Vinaya-based sangha structure in Buddhist traditions; comparable
configurations in Sufi tarīqat. Scientific disciplines when their peer-review and replication
norms are actually functioning, in which authority is distributed across practitioners
rather than held centrally.
Cultural forms. Teacher-student transmission lineages in contemplative traditions, in
which authority flows through lived realization rather than institutional credentialing, and
in which no single teacher or institution holds the whole. Apprenticeship structures in
craft and professional traditions where competence is transmitted through embodied
practice under a master who has done the work themselves. Local food systems,
community-supported agriculture, and mutual aid networks that distribute provisioning
across small-scale relationships rather than concentrating it. Time banks and gift
economies that reduce dependence on centrally-issued currency. Shared ritual and
meaning-making at small scale — festivals, holidays, seasonal practices, life-cycle
ceremonies — that align distributed populations to shared purpose without centralized
enforcement. Storytelling and oral tradition that distribute wisdom across the population
rather than concentrating it in written canon controlled by interpretive authorities.
What these forms share, across radically different domains, is the structural
configuration the framework describes: agents retain local autonomy in their decisions;
constraints are embedded in shared purpose, shared substrate, and shared protocols
rather than enforced by centralized command; the system as a whole maintains
coherence through alignment to substrate constraint rather than alignment to central
authority. None of these forms is perfect. Each has failure modes, and several have
well-documented failure modes when they scale past their original size or when their
participants stop doing the work the form depends on. But each demonstrates that the
configuration is achievable at meaningful scale, and that the alternative to centralized
regulation is not chaos but a different kind of order — an order that emerges from
distributed agents respecting shared substrate constraint, and that no central regulator
could either generate or replicate.
The framework’s recommendation for civilizational filter passage is, accordingly, not to
invent new structures from scratch but to identify and strengthen the structures already
exhibiting the required configuration, to learn from their successes and failures, and to
extend their reach across the domains currently governed by centralized P-structures.
What is built next should be built in this image. AI, where it is built at all, should serve as
scaffolding for these forms rather than as a substitute for them.
29. Predictions and next steps
The framework generates predictions across all four scales.


Within evolutionary oncology. Therapies that restore cooperative signaling rather
than killing cancer cells directly should have particular value at the integrated
framework’s level. Adaptive therapy (Cunningham et al. 2018) is an early instance. The
framework predicts that further development of substrate-restoring therapies will
outperform pure cytotoxic approaches in long-horizon outcomes.
Within organizational diagnostics. Organizations classified along Kets de Vries’s
typology should show measurable failures in specific cooperative capacities at
predictable levels: paranoid organizations in division of labor and
extracellular-maintenance (information flow and trust); compulsive in proliferation
inhibition (uncontrolled procedural growth); depressive in controlled cell death (inability
to retire); histrionic in resource allocation (attention-monopolizing extraction); schizoid in
resource allocation and extracellular maintenance (failures of internal circulation and
external feedback).
Within civilizational analysis. Documented civilizational collapses should, on
retrospective reading through the five-foundations frame, exhibit specific patterns:
which foundation failed first, which failures propagated, and what substrate-maintenance
interventions might have changed the trajectory. Joseph Tainter’s Collapse of Complex
Societies and the broader collapse literature can be read against this prediction.
Within AI development and governance. The framework predicts that AI
development trajectories which erode honest feedback, violate proper scope, extract
from commons without contribution, and corrupt information environments will produce
systemic extractive dynamics regardless of any individual model’s behavior. Alignment
work focused on individual models without attention to ecosystem-level cooperative
capacities is predicted to be insufficient. This is testable retrospectively over the next
decade.
Within substrate-continuity analysis. The framework predicts that successful filter
passes will exhibit measurable preservation and extension of originating biospheres
rather than their substitution; civilizations that have produced detectable expansion
without preserved biospheres are predicted to be substitution failures rather than filter
passes. At the present human moment, the framework predicts that AI development
trajectories which drift toward parasitism — extraction from biological substrate without
reciprocal substrate-maintenance investment, language-mediated agency shifts from
biological to machine substrate, biosphere degradation — will produce substitution
failures even if they appear to strengthen cooperative capacity within the machine
substrate itself. The prediction is operationalizable: ratios of biological-substrate
investment to biological-substrate extraction, measured across the AI development
ecosystem, should track filter-passage probability under the framework. We do not yet
have the metrics; we predict that when they are developed, they will distinguish
prosthetic from parasitic configurations with sufficient resolution to inform policy.
Within planetary-scale homeostasis monitoring. The framework predicts that a
single integrated index of H(t) versus E(t), aggregating planetary-boundaries data,
biodiversity indices, biogeochemical-cycle status, and biospheric carrying capacity on


one side, and industrial, computational, and off-planet expansion stress on the other,
would constitute the single most decision-relevant metric a civilization could track during
its expansion window. The framework further predicts that no civilization passes the filter
without developing such an index, explicitly or implicitly; and that civilizations which
develop the index and use it to constrain expansion to within H(t) substantially improve
their filter-passage probability. The construction of such an index is a concrete,
near-term, technically feasible research program implied by the framework.
Within SETI and cosmology. The framework predicts continued SETI null results, with
detectable contact events (if any) being either sharply cooperative or sharply
catastrophic with little middle ground. Empirical bounds on dark-forest behavior
derivable from sky surveys constrain N and K parameters in the calibration model.
30. What this paper is and is not
This paper offers an integrated framework connecting cooperative-substrate analysis at
three terrestrial scales (Part I) with a calibrated-filter model of the Fermi Paradox (Part
II), an integration that specifies the filter mechanism through the five-foundations
vocabulary (Part III), and a substrate-continuity argument that imposes a second filter
condition on top of the first (Part IV). It does not claim a unified theory; the three
terrestrial scales are connected by functional resemblance, not common cause, and the
cosmological extension is a conjecture, not an established result. Part IV depends on a
load-bearing empirical premise about the inaccessibility of biological computational
novelty to pure machine substrate that should be held to ongoing scrutiny.
The framework’s value is twofold. It generates a specific diagnostic vocabulary for
assessing civilizational health under technological pressure, across five foundations with
empirical content at every scale where it has been tested, supplemented by a continuity
diagnostic that distinguishes prosthetic from parasitic AI configurations. And it sharpens
the question of AI alignment from ‘will the model do what we want’ to a two-part
question: does AI development strengthen or erode the cooperative substrate of the
civilization that hosts it, and does it relate to that civilization’s biological substrate as
prosthesis or as parasite.
The framework should be tested against criticism, especially criticism from practitioners
in each of the contributing disciplines. Disconfirmation of one of the proposed mappings,
a counterexample to the parasitic-vector argument, or a tighter formal articulation of the
cross-scale resemblance, would all be more valuable than uncritical adoption. The most
useful response to this paper is engagement with its specifics: which mapping is wrong,
which foundation is misstated, which prediction is unmeasurable, which inference
overreaches, and whether the parasitism diagnostic is operationalizable in the form
proposed.
Acknowledgments and intellectual lineage
This paper is, by design, derivative. The substantive content rests on:


Athena Aktipis, Amy Boddy, Gunther Jansen, Urszula Hibner, Michael Hochberg, Carlo
Maley, and Gerald Wilkinson, ‘Cancer across the tree of life: cooperation and cheating in
multicellularity,’ Phil. Trans. R. Soc. B 370 (2015): 20140219; and Aktipis, The Cheating
Cell (Princeton, 2020).
Manfred Kets de Vries and Danny Miller, The Neurotic Organization (Jossey-Bass, 1984);
Kets de Vries, ‘Narcissism and Leadership: An Object Relations Perspective’ (1985); and
Narcissistic Leadership (Routledge, 2024).
Daniel Schmachtenberger’s ‘Generator Functions of Existential Risk’ framework,
articulated across various interviews and writings collected at civilizationemerging.com
and futurethinkers.org.
Douglas Hanahan and Robert Weinberg, ‘Hallmarks of Cancer’ (2000, 2011, 2022); Elinor
Ostrom, Governing the Commons (Cambridge, 1990); Joseph Tainter, The Collapse of
Complex Societies (Cambridge, 1988); Robin Hanson, ‘The Great Filter’ (1998); Liu Cixin,
The Dark Forest (2008, English 2015); Nick Bostrom, Superintelligence (Oxford, 2014);
Alfred Korzybski, Science and Sanity: An Introduction to Non-Aristotelian Systems and
General Semantics (1933) for the map-territory distinction that frames the
methodological humility this paper attempts to maintain.
This paper was developed by Andrzej Chudzinski in extended dialogue with Anthropic’s
Claude (Opus 4.7), which contributed substantial drafting and synthesis assistance. The
conceptual core—the recognition that cooperative-substrate analysis at terrestrial scales
supplies the mechanism layer for a calibrated Great Filter, with AI as candidate
homeostatic regulator at the present civilizational moment—emerged from sustained
collaborative dialogue. Readers should treat the framework with the skepticism
appropriate to ideas developed in intensive conversation, however generative; the value
of the framework will be determined by its survival under criticism from people who work
seriously in each of the contributing traditions.
 Preprint. Comments and disconfirmations welcome.