Formalizes the “anthropic shadow” effect: because observers can only find themselves in a history without an extinction-level catastrophe, the observed historical frequency of near-extinction events (e.g. large impacts, supervolcanic eruptions) is a biased underestimate of the true underlying rate, with the bias growing as the event’s fatality-probability approaches 1. Develops a Bayesian model relating true rate, observed rate, and the population/civilization survival function, and applies it to impactor and supervolcano statistics, showing naive frequentist estimates can understate risk by orders of magnitude for near-certain-fatality event classes. relevance_note: gives the general quantitative machinery for how much “we survived X so far” undercounts risk when survival is a precondition for observing at all — directly bears on whether cosmic-ray/collider survival arguments license the confidence claimed.
Conditioning fact and Bayesian model (§2-4)
O-17 - Human observers exist today, so every historical record is conditioned on no observer-extinguishing catastrophe having occurred
This is the conditioning fact (“E, the fact of our present-day existence”) on which the paper’s Bayesian anthropic-shadow machinery operates: events that would have extinguished observers are unobservable in our past regardless of their true frequency.
Link to original
A-19 - Anthropic-shadow correction - observed catastrophe frequencies underestimate true rates, without bound as survival probability goes to zero
Reasoning
Link to original
- Single-event toy model: let B2 = “the catastrophe occurred in the past interval” with prior P, and let Q = P(humanity survives | catastrophe). The evidence is E = our present existence, with P(E|B2) = Q and P(E|not-B2) = 1. Bayes: P(B2|E) = PQ / ((1-P) + PQ).
- Overconfidence parameter eta := P(a priori)/P(a posteriori) = (1 - P + PQ)/Q. For Q = 1 (harmless events) eta = 1: the record is unbiased. As Q falls, eta rises; eta → infinity as Q → 0. Worked example: Q = 0.1, P = 0.5 (a Toba-scale event at ~1/Myr over human evolution) gives eta = 5.5 - the true probability is 5.5x the naive past-frequency estimate.
- Generalization (§3): for N possible disasters with per-slot probability alpha and lethality beta, P(k observed disasters and observers | alpha, beta) = C(N,k) alpha^k (1-alpha)^(N-k) (1-beta)^k; with uniform priors, P(alpha, beta | O, k) is depressed at high beta - the surviving observer’s record systematically under-samples severe events (“anthropic shadow”, the biased-sampling region above the anthropic compatibility boundary in the time-severity plane).
- The bias is cumulative with classical selection effects (erosion of traces etc.) and must be corrected after them when constructing the a-priori distribution used for prediction; the terrestrial cratering record (Earth Impact Database: no >100-km craters in the recent past except Chicxulub) illustrates the censored region.
- Scope: the bias afflicts hazards that (1) could have destroyed our species or its predecessors, (2) are uncertain, and (3) whose frequency estimates rest largely on terrestrial records.
Application to physics-disaster safety arguments (§2, §6)
A-20 - Cosmic-ray and solar-system-survival safety arguments are unreliable for observer-extinguishing event classes
Reasoning
Link to original
- Apply the toy model at Q = 0: eta = (1 - P + PQ)/Q diverges - the posterior credence licensed by “we (and our planet’s habitability) are still here” is zero information about P, since every possible observer, in every world with any value of P, makes the same observation. Survival evidence discriminates between rival risk levels only through histories in which observers survive, and for Q = 0 classes all such histories look identical.
- Hence the Hut-Rees inference (natural cosmic-ray collisions on Earth/Moon exceed accelerator collisions by orders of magnitude, and we survived 4.5 Gyr of them, so accelerators are safe for the foreseeable future) is partially misleading: the Earth-survival leg supplies no anthropically clean information for observer-extinguishing outcomes. The paper notes the divergent Hut-Rees sub-argument that remains valid - the count of collisions in any conceivable accelerator is minuscule beside nature’s - while the survival-frequency leg fails uncorrected, and the LSAG report repeats the uncorrected form via the solar-system-duration argument.
- The unreliability applies equally to naturally occurring and human-induced members of the class; anthropogenic hazards generally leave no deep-history record at all, physics disasters being the rare exception where past-frequency reasoning is even attempted.
- Repair route acknowledged by the paper: conditioning on external evidence instead - e.g. Tegmark & Bostrom’s use of planetary-age distributions and Earth’s late formation date bounds the vacuum-decay rate below ~1e-9 per year - shows the shadow can be circumvented, not that the uncorrected survival argument stands.
Scope limit (§2, §5)
A-21 - Evidence not conditioned on terrestrial survival - astronomical observations - largely escapes the anthropic shadow
Reasoning
Link to original
- The shadow arises from sampling only histories compatible with our existence. Evidence whose generating process would look the same whether or not Earth-bound observers survived - e.g. star counts, stellar mass functions, supernova/GRB frequencies measured in external galaxies similar to the Milky Way - is (to good approximation) not filtered by our survival, so inference from it needs no anthropic correction. (Approximation: extreme cases, e.g. a nearby event that would also have prevented astronomers from existing, retain some residual conditioning.)
- The paper’s own worked instance: the conclusion that Earth’s destruction by an encounter with a passing star, neutron star, or black hole is extremely improbable cannot be obtained solely from our planetary system’s undisturbed past history (that inference is shadowed), but admitting the external astronomical information renders it “bias-free and persuasive”.
- Corollary the paper draws for hazard classes generally: distribution frequencies of large cosmic explosions inferred from distant external galaxies decrease the anthropic bias in those estimates, whereas terrestrial-record-based rates (craters, ice cores) remain shadowed; the amount of admissible external information is uneven across hazard types.
- This is the scope limit on the shadow correction: it partitions survival-style safety evidence (shadowed, per the companion argument) from external astrophysical evidence (largely clean), rather than discrediting both.