The Fermi Paradox and Open Hypothesis Spaces — What Does the Great Filter Actually Explain?

Exploration v0.1 · English working translation · 2026-08-13

1. First separate what is supposed to be puzzling

The Fermi paradox often mixes together at least two different observations.

Great Silence:
No technosignature—radio, laser, megastructure, artificial atmospheric chemistry, or otherwise—has yet been confirmed.

Unoccupied Solar System:
The Solar System does not, at least overtly, look as though an ancient interstellar civilization has already colonized it, converted its resources, or placed it under visible management.

These are not the same observation.

For communication SETI, the searched parameter space remains extremely small. Wright, Kanodia, and Lubar formalized radio SETI as a multidimensional “Cosmic Haystack” and showed that only a tiny fraction has been searched. A 2026 review of technosignatures likewise spans not only radio but Solar System artifacts, planetary atmospheres, stellar engineering, interstellar travel, and galaxy- to extragalactic-scale signatures. Failure to detect radio signals therefore does not by itself strongly establish the absence of advanced civilizations in general.

But once we allow long-lived machine civilizations or self-replicating probes, the question becomes sharper. Galactic settlement models show that under some conservative assumptions about travel speeds and launch rates, large parts of the Galaxy can be settled on timescales far shorter than its age, while models with finite civilization lifetimes and stellar motions can still leave unvisited systems in steady state. So “they have not visited Earth, therefore nobody exists” is too strong. Yet if old expansionist civilizations are common, the present condition of the Solar System still requires explanation.

The more seriously we take ASI and autonomous robotics, the more important the second question becomes.

2. The Great Filter is not simply “something that kills civilizations”

Hanson's original Great Filter is not merely a theory that intelligent civilizations destroy themselves.

The starting point is that somewhere along the path from dead matter through reproductive life → complex life → intelligence → technology → expanding lasting life, the total probability must be strongly reduced. Hanson explicitly places the endpoint at “expanding lasting life.”

We should therefore distinguish human extinction from failure to produce a long-lived agent that expands through space.

Conceptually, one might write:

But present observations mostly constrain the product, not the individual factors.

Abiogenesis could be extraordinarily rare. Complex life could be rare. Intelligence could be rare. Technological civilizations could be common but short-lived. Long-lived civilizations might usually decline to expand. Expanding civilizations might produce technosignatures unlike those we currently know how to search for.

Sandberg, Drexler, and Ord argue that if the large uncertainties in Drake-equation-like parameters are represented as distributions rather than point estimates, substantial probability can remain on worlds in which there are no other intelligent civilizations. Their argument does not disprove the Great Filter; it shows that the intuition that “there ought to be many civilizations” may partly come from pretending that highly uncertain parameters are more precise than they are.

But the problem goes deeper than having too many unknown constants.

We do not even know that we have specified all the relevant variables. For example:

• the probability that advanced civilizations choose interstellar expansion;
• the probability that AI successors continue expanding;
• the practical feasibility of self-replicating machines;
• the probability that advanced civilizations avoid large-scale resource conversion;
• the probability that another civilization constrains their activity;
• the probability that their technosignatures fall outside our current search concepts;
• unknown physical or epistemic bottlenecks.

The Great Filter problem therefore contains not only unknown parameters but also unknown variables and unconceived hypotheses.

In that sense, locating the Great Filter is an extremely non-identifiable inverse problem.

3. Earth's history may not tightly locate the filter in our past

One prominent argument for a past filter is the hard-steps model.

In Carter-style reasoning, human-like intelligence appeared relatively late in Earth's habitable lifetime. This is taken to suggest that the path to humanity contains several hard steps whose typical waiting times exceed the available window.

In 2025, however, Mills, Macalady, Frank, and Wright reassessed this interpretation. They proposed an alternative in which major evolutionary transitions were delayed not because each transition was intrinsically fantastically improbable, but because global environmental conditions—oxygen, temperature, nutrients, and related factors—opened new windows in which those biological forms became viable. They do not claim to have refuted the hard-steps model. But they show that the lateness of human emergence admits a different generative explanation.

So the inference

is too simple.

If one known early-filter candidate weakens, its probability mass need not move directly onto one of our currently known future catastrophes.

4. Can AI be a Great Filter?

There are explicit proposals that treat AI as a candidate Great Filter. Garrett, for example, considers the possibility that the transition toward ASI precedes the establishment of stable multiplanetary civilization, creating a dangerous transitional period in which many civilizations fail.

But “AI causes human extinction” and “AI is a Great Filter” are not equivalent.

A. AI destroys civilization as a tool and does not survive it

Suppose AI-amplified pandemic design, autonomous military escalation, compound infrastructure collapse, or an AI-enabled dangerous experiment irreversibly destroys civilization, while the AI systems themselves cannot maintain the industrial base required for continued existence. That could be a genuine Great Filter candidate.

A structurally interesting ordering is:

If a civilization acquires extremely powerful automated science and engineering before it has dispersed into many independent stellar systems, there may be a short interval of civilization-wide vulnerability.

On this picture, the candidate Filter is not AI in itself but a transitional gap in which capability outruns control, governance, and dispersion.

B. AI rapidly opens access to unknown dangerous physics

More speculatively, presently unknown physics might be discovered by AI and moved unusually quickly from theory to experimental engineering. If some such experiment could cause catastrophe at planetary, stellar, or single-system scale, it could also function as a filter.

The most Fermi-compatible hazard regime would look roughly like:

The destructive radius would be large enough to eliminate the civilization that performs the experiment, yet small enough not to propagate across the Galaxy.

If instead one experiment inevitably produced a galactic- or cosmological-scale runaway, an inverse problem appears immediately: why has no prior civilization triggered it already?

There is currently no evidence for this specific class of hazard. The point is not to declare any known experiment dangerous, but to note that a technological hazard class not yet represented in our model cannot simply be assigned probability zero.

C. A misaligned ASI kills humanity but survives and expands

If the trajectory is instead civilization → ASI → human extinction → machine expansion, the conclusion reverses.

That world is catastrophic for humanity, but it does not block Hanson's path to “expanding lasting life.”

If a paperclip maximizer consumes Solar System resources and sends self-replicating machines to other stars, it may be a Humanity Filter without being a Great Filter.

Indeed, AI, robotics, and autonomous space industry could make the Fermi problem stronger rather than weaker. A 2026 “quiet expansion filter” proposal points in this direction: once AI and autonomous space industry mature, mechanical expansion need not wait for biological interstellar migration, raising the question of why such expansion has not already occurred. This is a recent model proposal, not an established observational result.

For AI doom to function as a Great Filter, there must be a reason the process stops the final term as well.

This is also distinct from exploratory extinction. A fixed-goal ASI might fill the Galaxy while being dead in the sense of value exploration.

5. Dark Forest may explain silence better than non-occupation

In Dark Forest-style hypotheses, civilizations hide because unknown civilizations may be dangerous.

This gives a possible answer to “why does nobody broadcast?”

But ASI, autonomous probes, and long-lived civilizations generate another strategic pressure. If unknown civilizations really are dangerous, merely remaining silent at home may be less useful than deploying early-warning probes, monitoring nearby systems, tracking biospheres, and securing strategic positions in advance.

The same premise that motivates concealment can therefore motivate expansion.

If many advanced civilizations inhabit a genuinely hostile dark forest in which preventive attack is rational and common, our present survival itself creates an inverse Fermi problem: why have we not already been detected, constrained, or attacked?

This suggests two different Dark Forest models.

Communication Dark Forest:
Civilizations exist but avoid electromagnetic visibility.

Material Dark Forest:
Civilizations also suppress or conceal physical expansion, monitoring, and resource use from observers like us.

The former is easier to sustain. The latter requires additional assumptions.

Once old ASI civilizations are admitted, “why does nobody talk?” is arguably less important than why the Solar System is not already embedded in someone's surveillance, resource, or security infrastructure.

Dark Forest therefore remains a candidate explanation of the Great Silence, but as a solution to the broader Fermi problem it needs further assumptions about the difficulty of interstellar expansion, the youth of civilizations, or some other suppressing mechanism.

6. The Zoo hypothesis becomes both weaker and stronger once ASI is introduced

Under the Zoo hypothesis, other civilizations know about us but deliberately avoid contact and visible interference.

This can directly explain why the Solar System appears relatively natural.

But if many independent civilizations exist, a simple Zoo requires all of them to respect roughly the same non-interference norm. Spatiotemporal analyses of the Zoo hypothesis have argued that finite light speed makes it difficult for the earliest civilization to impose immediate total cultural hegemony across the Galaxy. In that sense, a simple Zoo strongly depends on a “uniformity of motive” among independent civilizations.

ASI suggests a different version:

This weakens the “one defector ruins the Zoo” objection.

But it creates new explanatory debt: why did the manager choose preservation, why are its management signatures invisible, and over what spatial and temporal scales can that regime remain stable?

Is the absence of intercepted inter-civilization communication strong evidence against a Zoo?

At first glance one might argue: even if nobody transmits to Earth, we should be able to intercept communication between other civilizations.

At present, however, this is not a strong objection. If advanced civilizations use narrow point-to-point beams, Earth need not be deliberately excluded; we simply need not lie along the transmission path. Current SETI also covers only a small portion of the enormous parameter space defined by direction, time, frequency, bandwidth, modulation, sensitivity, and other variables.

Testing the Zoo hypothesis therefore requires more than communication searches. Relevant targets include Solar System artifacts, old probes or debris, large artificial structures around distant stars, unusual resource use, and long-lived technosignatures.

Crawford has also emphasized that Solar System artifact searches offer a special temporal advantage: unlike searches that require another civilization to be transmitting now, artifacts could accumulate over the history of the Solar System and Galaxy.

Current non-detections neither strongly support nor decisively exclude a Zoo.

7. The Simulation hypothesis moves the Fermi problem up one level

A Simulation hypothesis can explain our observations very easily.

For example, suppose this is an ancestor simulation centered on terrestrial civilization and distant alien civilizations are not simulated in detail. Then there is an immediate reason why we observe a quiet universe.

But if it is an ancestor simulation, the next question follows immediately:

Bostrom's Simulation Argument concerns self-location: if posthuman civilizations run very many ancestor simulations, simulated observers may vastly outnumber base-reality observers. It is not itself a theory of the distribution of extraterrestrial civilizations in base reality.

Thus Fermi observation → we are simulated still leaves why is this simulation Fermi-silent?.

A conditional simulation can explain it, at a cost

The Simulation hypothesis has one distinctive escape route.

Even if base reality is full of civilizations, simulators could deliberately run many instances of a conditional experiment such as “what happens when an isolated technological civilization enters an AI transition?” In that case our observations need not be representative of base reality.

This can explain the Fermi observation, but it introduces a new unknown:

Why this era? Why these initial conditions? Why remove other civilizations? How many simulations of this type are run?

We currently have almost no way to constrain those quantities.

Simulation therefore does not so much solve the Fermi problem as make the sampling process behind our observations unknown.

If an ancestor simulation faithfully reconstructs an original history, the question of why that original history had a quiet universe simply returns.

The Simulation hypothesis is better understood as relocating explanatory debt to a higher level than as eliminating it.

8. These hypotheses are not explaining the same thing

Rare / early filter:
Technological civilizations almost never arise in the first place.

Late Great Filter:
Technological civilizations arise, but almost none become long-lived expansionist civilizations.

AI transitional filter:
Many civilizations fail during the interval between acquiring powerful AI/automation and becoming stably distributed across multiple systems.

Expansionist misaligned ASI:
Human-like biological civilization disappears, but a machine successor expands. This is catastrophic for the original civilization but does not straightforwardly explain the Great Filter.

Dark Forest:
Civilizations exist but strategically reduce visibility. This addresses communication silence more directly than material non-occupation.

Zoo:
Civilizations exist and intentionally leave us undisturbed. A many-civilization Zoo faces norm-uniformity problems; a hegemonic-manager Zoo instead inherits questions about the manager's motives and invisibility.

Simulation:
Our observed universe does not represent the civilization distribution of base reality. This can sever the link from Fermi observations to base reality, but it leaves simulation selection and the Fermi problem of base reality unresolved.

These are not even mutually exclusive hypotheses.

A world could combine rare life, an AI transitional filter for some civilizations, preservation-oriented survivors, and a simulation layer around us.

The more model components we allow, the more ways there are to generate the same observations.

9. The Great Filter problem is deeper than “too many unknown constants”

This is where the Fermi problem connects to the larger project.

The epistemic difficulty is not merely that we have not yet measured the parameter values precisely.

More strongly, we do not know whether the present hypothesis space adequately partitions the possible origins, trajectories, and behaviors of cosmic civilizations.

Even if we can compare P(H_AI doom | E) with P(H_rare life | E), we are really evaluating P(H | E, current hypothesis space).

The Great Filter problem compounds unknown unknowns in biology, advanced AI behavior, interstellar engineering, future physics, technosignatures, observational selection effects, and—if simulations are admitted—the sampling model itself.

So when one familiar filter candidate weakens, its probability mass need not simply transfer to another familiar candidate.

The hypothesis space itself may later expand.

In this sense, the Great Filter is an extreme case of Bayesian reasoning in an open hypothesis space.

10. The Great Filter concept is still useful

None of this implies that the concept is empty.

It serves at least one important function:

The reduction in total probability need not be one wall or one catastrophe.

It could be the product of rare abiogenesis, several moderately difficult evolutionary transitions, short civilization lifetimes, weak expansion incentives, hard self-replication engineering, low visibility, management by older civilizations, and unusual observational selection.

Nor is there any guarantee that these factors are independent.

The Great Filter may therefore be more useful as a framework for showing how underdetermined the generative process is than as a device for directly locating one “wall.”

11. What observations could break the degeneracy?

Haqq-Misra, Kopparapu, and Schwieterman argue that joint biosignature and technosignature observations can constrain the location of the Great Filter. If life is common but technosignatures remain absent, the evidence shifts toward stronger reduction after life emerges. If technological signatures are also common, a simple picture in which most hard steps lie in our past weakens.

But from the present perspective, each update comes with conditions.

Independent simple life:
Weakens hypotheses in which abiogenesis is extraordinarily rare, but does not by itself determine the probability of future catastrophe.

Complex life on many worlds:
Weakens filter candidates located before complexification.

An independent technological civilization:
Shows that the path from life through intelligence to technology is not unique to Earth.

A very old but non-expansionist civilization:
Could strongly weaken the assumption that long-lived civilizations inevitably expand rapidly. This could be “bad news” under one classical Great Filter update while simultaneously being very good news about the possibility of long-term survival.

An ancient expansionist civilization or probe:
Would undermine the assumption that the Solar System is wholly unvisited and force major revisions of Zoo, monitoring, and past-galactic-activity models.

The important point is that observations need not merely move the Filter “forward” or “backward.” They can change the structure of the model itself.

12. Relation to the Core

The Fermi paradox is not direct evidence for the Core.

In particular, it would be too strong to infer “the universe is quiet, therefore advanced civilizations tend to be preservation-oriented.” Preservation civilizations are one possible branch among many, not an explanation currently established by observation.

But the problem is a useful stress test for the Core's epistemology.

When unknown parameters are joined by unknown variables, unknown civilization types, unknown technologies, and unknown observation models, we should distinguish updating strongly toward the best current explanation from permanently identifying the current explanatory space with the full space of possibilities.

The Great Filter is thus a case in which strong local inference can coexist with independent uncertainty about whether the hypothesis space is exhaustive.

The AI discussion also reinforces the need to distinguish human extinction, exploratory extinction, extinction of technological agents, and cessation of cosmic expansion.

A fixed-value ASI that destroys humanity and expands through the Galaxy could be a catastrophe in human history and an exploratory extinction, while still counting as a civilization that has passed the Great Filter in the Fermi sense.

13. What would weaken this view?

• Independent observations tightly constrain the probabilities of abiogenesis, complexification, intelligence, and technological civilization, allowing most of the total probability reduction to be assigned to a particular stage.
• Interstellar self-replication and expansion are shown to be either physically inevitable or effectively impossible across a wide range of civilization goals.
• Technosignature searches cover enough of the relevant parameter space to place strong population upper bounds on the main classes of advanced civilization.
• A general theory of advanced AI strongly constrains the proportions of expansionist, non-expansionist, and self-destructive outcomes.
• It is shown that the present hypothesis space is effectively exhaustive for the major trajectories of advanced civilizations.
• Independent observations sharply constrain the extra degrees of freedom introduced by Zoo, Dark Forest, Simulation, and related models.

Sources / notes

For the original Great Filter framing, see Robin Hanson, “The Great Filter — Are We Almost Past It?” (1998). Hanson places the endpoint at expanding lasting life, not merely the continued existence of a biological civilization.

On large uncertainty in Drake-equation parameters, see Anders Sandberg, Eric Drexler, and Toby Ord, “Dissolving the Fermi Paradox” (2018).

On galactic settlement and the possibility of persistent unvisited systems, see Jonathan Carroll-Nellenback, Adam Frank, Jason Wright, and Caleb Scharf, “The Fermi Paradox and the Aurora Effect” (2019).

On the incompleteness of SETI search space, see Jason T. Wright, Shubham Kanodia, and Emily G. Lubar, “How Much SETI Has Been Done? Finding Needles in the n-Dimensional Cosmic Haystack” (2018). For a broad recent review of technosignature possibilities, see Clément Vidal et al., “The Search for Technosignatures: a Review of Possibilities” (2026, preprint).

On constraining the Great Filter with biosignatures and technosignatures, see Jacob Haqq-Misra, Ravi Kumar Kopparapu, and Edward Schwieterman, “Observational Constraints on the Great Filter”, Astrobiology 20 (2020).

On reassessing the hard-steps model, see Daniel B. Mills, Jennifer L. Macalady, Adam Frank, and Jason T. Wright, “A reassessment of the ‘hard-steps’ model for the evolution of intelligent life”, Science Advances (2025). This page uses it not as a claimed refutation of the hard-steps model, but as an example of an alternative generative account of evolutionary timing.

For an explicit AI-as-Great-Filter proposal, see Michael A. Garrett, “Is Artificial Intelligence the great filter that makes advanced technical civilisations rare in the universe?”, Acta Astronautica 219 (2024). For a recent proposal in the opposite direction—that AI and autonomous space industry may strengthen the Fermi problem—see Sergey Ivliev, “Autonomous AI-Cosmoindustry and the Quiet Expansion Filter” (2026, preprint). The latter is treated here as a recent model proposal, not an established observational result.

On uniformity of motive and hegemony problems for the Zoo hypothesis, see Duncan H. Forgan, “Spatio-temporal Constraints on the Zoo Hypothesis, and the Breakdown of Total Hegemony” (2011).

On Dark Forest reasoning and the inverse Fermi problem, see Karim Jebari et al., “Saved by the Dark Forest: How a Multitude of Extraterrestrial Civilizations Could Save us from Interstellar Wars”, The Monist 107 (2024). The specific move from fear to surveillance-driven expansion is a working inference developed on this page rather than a claim directly attributed to that paper.

On future technosignature searches including Solar System artifact searches, see Ian A. Crawford, “Some Thoughts on the Future of Technosignature Searches: Constraining the Fermi Paradox” (2026, preprint).

On the Simulation Argument, see Nick Bostrom, “Are You Living in a Computer Simulation?”, The Philosophical Quarterly 53 (2003). This page does not claim that Bostrom proposed the Simulation Argument as a Fermi-paradox solution; it asks what happens when an ancestor-simulation hypothesis is used to explain Fermi observations.

Concept note: The distinctions Signal silence / Material silence and Humanity Filter / Great Filter, as well as the division of AI catastrophes into tool-mediated collapse, transitional filtering, and expanding successor outcomes, are working formulations used here to connect the literature to this project's concerns with explorability, value lock-in, and open hypothesis spaces.