Skip to content

Gell-Mann Amnesia Effect

A reader who finds a source riddled with errors in their own field turns the page and trusts it on an unfamiliar topic, because the error was filed as topic-specific news rather than propagated to a source-level prior.

Core Idea

The Gell-Mann amnesia effect, named by Michael Crichton after physicist Murray Gell-Mann, is the cognitive pattern in which a reader who has just found a news article on their own area of expertise riddled with errors turns the page and trusts the next article — on a topic outside their expertise — as if the demonstrated unreliability had never occurred. The mechanism is a failure of source-level belief updating: evidence about process quality (this outlet makes basic errors) is registered only as evidence about a topic (this particular article was wrong) and never propagated to a prior over the source as a whole, so the second-domain trust is set at the reader's baseline rather than discounted by the already-observed error rate.

Three structural features define the pattern. First, the reader has a domain in which they can independently verify accuracy — the expert check — and the source fails it. Second, the reader immediately encounters claims from the same source in a domain where no independent check is possible. Third, the amnesic reset: whatever the reader just learned about the source's quality in domain A is not carried into their credence about the source's claims in domain B. The result is that the most diagnostic event available for calibrating a source — encountering it in one's own specialty — has no effect on trust outside that specialty. The effect is structural, not motivated: the reader is not suppressing the uncomfortable finding; they are categorizing it as topic-specific information rather than source-quality information, which is the wrong level of abstraction for calibration purposes.

Structural Signature

Sig role-phrases:

  • the multi-topic source — one outlet or author publishing claims that span many beats, only some of which any given reader can check
  • the in-field check — a domain in which the reader has expertise enough to independently verify the source's accuracy
  • the out-of-field exposure — a second domain, consumed seconds later, where the reader has no independent check
  • the failed expert check — the source caught making basic errors inside the reader's specialty, the maximally diagnostic source-quality event
  • the level-of-abstraction mis-filing — the error registered as topic information (this article was wrong) rather than source-process information (this outlet makes errors), the wrong bin for calibration
  • the amnesic reset — the topic-A finding not propagated to a source-level prior, so it never reaches the cross-topic credence
  • the un-discounted out-of-field trust — credence in the second-domain claims set at baseline rather than discounted by the just-observed error rate, the offset outcome
  • the source-level filing corrective — the structural lever: topic-banded credibility tracking in place of one aggregate rating, giving the in-field evidence a place to land (vs. exhorting more skepticism, which leaves the slope unmoved)

What It Is Not

  • Not generic over-trust in authority. The defining twist is that the source has just been demonstrated to fail, and the reader saw it; this is not deference granted to an unexamined authority but trust extended to one whose unreliability was witnessed seconds earlier. Ordinary appeal-to-authority lacks the prior disconfirmation that makes the amnesia striking.
  • Not motivated reasoning or wishful thinking. The reader has no stake in the unfamiliar-topic claim, so there is nothing to rationalize away. The failure is structural, not motivated — the error is categorized as topic-specific information rather than suppressed as uncomfortable, which is why exhorting honesty does not fix it.
  • Not literal forgetting. The "amnesia" is metaphorical: the reader has not lost the memory of the errors. They remember them perfectly well but file them at the wrong level of abstraction — as facts about this article's topic rather than about the source's process — so the evidence never reaches the source-level prior that would discount the next piece.
  • Not topic-specific credibility calibration. Treating an outlet as reliable on macroeconomics and unreliable on epidemiology can be sound if grounded in evidence per beat. The effect is the opposite: the same source rated inconsistently across topics consumed seconds apart, with the inconsistency driven by which topic the reader can check rather than by any difference in the source's actual reliability.
  • Not an inconsistency to be scolded. The reader is not being hypocritical or careless; the breakdown is in where evidence is filed, and the corrective is structural — topic-banded credibility tracking that gives the in-field finding a place to land — not an appeal to "try harder to be skeptical," which leaves out-of-field trust unmoved.

Scope of Application

The effect lives in one domain — human source-credibility judgment under partial observability — and the contexts below are different application settings of that single cognitive substrate (one reader failing to propagate source-quality evidence across topic-bins), not structurally distinct disciplines. Non-human or institutional "forgetting" of a source's failures is the bin-segregation / base-rate-neglect parent travelling by analogy, and stays out of this map.

  • News media — the original case: a specialist reads a story in their own field, sees the errors, then trusts the next article on an unfamiliar beat at baseline.
  • Peer review and academic skim-reading — errors are caught in papers near one's specialty while prestige-based deference is granted to claims in adjacent fields without the same scrutiny.
  • Expert testimony — a credentialed specialist is over-trusted outside their specialty, the argument for topic-anchored rather than credentials-anchored cross-examination.
  • Social-media and influencer credibility — a follower who catches a creator being wrong on a known topic does not revise trust on the topics they cannot check.
  • Brand and outlet trust — process failures discovered in one beat or product line fail to discount the source's other outputs.
  • Encyclopedia / Wikipedia consumption — readers spot errors in entries on their specialty yet continue to trust entries on unfamiliar topics from the same resource.

Clarity

Naming the effect lets media-literacy and epistemology educators point at a specific bug in source-evaluation behavior and stop treating it as ordinary credulity. The crucial distinction it sharpens is level of abstraction: evidence gathered while reading an outlet on one's own specialty is evidence about the source's process (this outlet makes basic errors), not merely about a topic (this one article was wrong), and calibration requires filing it at the source level. Without the label, an expert's continued trust in the next article looks like generic deference to authority or a lapse of attention; with it, the failure is localized precisely — the reader registered the right evidence but binned it at the wrong level, so the prior over the source never moved.

The naming also separates the effect from three patterns it superficially resembles. It is not generic over-trust in authority, because the authority has just been demonstrated to fail and the reader saw it. It is not topic-specific expertise calibration, because the very point is that one source is treated inconsistently across topics consumed seconds apart. And it is not motivated reasoning about the second article, because the reader has no stake in the unfamiliar-topic claim. Holding these apart lets a practitioner ask the sharper question: when is the most diagnostic event available for rating a source — encountering it inside one's own field — being discarded as topic news rather than propagated into a source-level prior? That question, in turn, names the corrective design move (topic-banded credibility tracking rather than a single aggregate trust rating) and points to where the effect bites hardest: generalist outlets spanning many beats, read by people with narrow but deep expertise.

Manages Complexity

Source-credibility judgment looks like an open-ended problem: every reader, every outlet, every beat, every article seems to demand its own assessment of how much to trust. The effect compresses that sprawl by locating the recurring failure at a single structural joint — the level of abstraction at which in-domain evidence gets filed. Instead of cataloguing the endless particulars of who trusts what, an analyst tracks just a few quantities: whether the reader has a domain in which the source can be independently checked, whether the source fails that check, whether the next claims fall in a domain where no check is possible, and whether the topic-A finding is binned as topic news or propagated to a source-level prior. Fix those, and the trust outcome follows without re-deriving each reader-outlet pairing from scratch. The same compression yields a sharp predictor of where the effect bites hardest — generalist sources spanning many beats, consumed by readers with narrow but deep expertise, under a "generally credible" rather than "credible on these topics" contract — and a single corrective lever (topic-banded credibility tracking in place of one aggregate trust rating), so a large, messy space of credibility behavior collapses to a small set of dials and one qualitative branch.

Abstract Reasoning

The effect licenses a cluster of inferences about how trust in a multi-topic source ought to move, and where it predictably fails to.

Diagnostic — from a surface signature to a hidden mis-filing. The characteristic move runs from an observed inconsistency back to its cognitive cause. When a reader rates the same source as untrustworthy on topic A and trustworthy on topic B in the span of a single sitting, the analyst infers not two genuine credibility estimates but one credibility estimate that never formed: the topic-A error was binned as topic news rather than propagated to a source-level prior. The diagnostic tell is the adjacency in time — the closer together the two judgments, the more clearly their divergence implicates a binning failure rather than honestly differing evidence. Run the inference the other way and it constrains the reader: from "this reader trusts the source's out-of-field claims at baseline," conclude that whatever in-field disconfirmation they have encountered has not been filed at the source level, because if it had, the out-of-field credence would carry the discount.

Interventionist — what to change, and the predicted effect on trust. Because the failure is localized to the level of abstraction at which in-domain evidence is stored, the corrective lever is fixed by the diagnosis: replace a single aggregate trust rating with topic-banded credibility tracking ("reliable on macroeconomics; shown unreliable on epidemiology"), and the prediction is that out-of-field trust should fall toward the discount the in-field check warranted, because the evidence now has a source-level place to land. Conversely, the model predicts which interventions will not work: a fix aimed at the second article — better disclaimers on the unfamiliar-topic piece, more reader attention to it — leaves the slope unmoved, because the breakdown was never about that article's content but about the absent propagation from the first. The intervention that bites is structural (change where evidence is filed), not exhortative (try harder to be skeptical).

Boundary-drawing — which regime, and where the effect bites hardest. The concept marks the conditions under which it applies, turning "readers are inconsistent" into a regime prediction. It requires (i) a reader with a domain in which the source can be independently checked, (ii) a check the source fails, and (iii) a second domain where no independent check is possible. Remove any leg and the effect does not arise: a reader with no field of expertise has no diagnostic check to discard; a source that passes the in-field check supplies no discount to propagate; a second domain the reader can also verify lets content evidence substitute for the missing source prior. Inside that regime, magnitude is predictable — the effect bites hardest for generalist sources spanning many beats (more topic-bins to sever across), read by people with narrow but deep expertise (a sharp in-field check but wide out-of-field exposure), under a "generally credible" rather than "credible on these topics" contract (which licenses aggregate trust the structure cannot support).

Order-of-events / predictive. The model fixes the most diagnostic moment in a reader's relationship with a source: the first time they encounter it inside their own specialty. That event is where source quality is most cleanly observable, so it is where calibration should occur — and the effect predicts that precisely this maximally-informative event will leave out-of-field trust unchanged unless the evidence is filed at the source level. It also predicts the steady-state outcome: under aggregate trust ratings, source credibility estimates run systematically too coarse and out-of-field experts are over-trusted, because the one mechanism that could correct the estimate — encountering the source where one can check it — is routed into the wrong bin every time.

Knowledge Transfer

Within cognition and media literacy the diagnostic transfers as mechanism, intact, but the honest qualification is that its "range" is a span of application contexts rather than structurally distinct substrates: in every case the substrate is one human propagating (or failing to propagate) source-quality evidence across topic-bins. With that understood, the three-step signature and its corrective lever carry without translation across news media (the original doctor-reads-the-medical-story case), peer review and academic skim-reading (errors caught near one's specialty, prestige-based deference granted to adjacent fields), expert testimony (a specialist over-trusted outside their specialty, arguing for topic-anchored rather than credentials-anchored cross-examination), social-media and influencer credibility, brand and outlet trust, and encyclopedia/Wikipedia consumption. The vocabulary (in-field check, source-level versus topic-level filing, the amnesic reset), the diagnostic (inconsistency across adjacent-in-time judgments implicates a binning failure, not honestly differing evidence), and the intervention (topic-banded credibility tracking, plus the prediction that fixing the second article does nothing) move freely across all of these — because they are the same cognitive operation viewed against different topic catalogs, not different mechanisms.

Beyond that human-source-evaluation substrate the named effect does not travel as mechanism; what travels is the more general parent pattern it instantiates. The portable core is a bin-segregation failure: evidence about an object filed in one bucket fails to update beliefs about the same object in another bucket — process-quality evidence mis-binned as topic-specific content, the wrong level of abstraction for calibration. That parent is the catalog region around base-rate neglect (the in-field sample is exactly the source-quality base rate the reader declines to update on) and the silo effect (information siloed by topic rather than aggregated to the source level), and it recurs as co-instances wherever a multi-topic generator is judged under partial observability — including non-human cases the effect's own framing does not cover: trusting a forecasting model's out-of-sample predictions despite watching it miss on the cases one can check, or extending an institution's reputation from an audited division to an unaudited one. In each, the general mis-propagation mechanism is what recurs; the Gell-Mann concept's own cargo — the reader-turns-the-page narrative, the newspaper-beat framing, the media-literacy intervention vocabulary — stays bound to human consumption of multi-topic publications. Any use of "Gell-Mann amnesia" outside that setting (an organization that "forgets" a vendor's failures across departments, say) is therefore borrowing the picture by analogy, while the load-bearing lesson belongs to the bin-mixing / base-rate-neglect parent. The right move when the lesson is needed elsewhere is to carry that parent, not the eponymous effect with its newsroom furniture (see Structural Core vs. Domain Accent).

Examples

Canonical

Michael Crichton coined and described the effect in his 2002 talk "Why Speculate?", naming it after physicist Murray Gell-Mann for wry effect. Crichton's own worked instance is the newspaper reader: you open the paper to a story on a subject you know well — Crichton's example is a topic in your own professional field — and you see the piece is riddled with errors, misunderstanding the causation, getting the facts backwards, wet-paint wrong. You register the outlet as unreliable on that story. Then you turn the page to national or international news, on subjects you cannot independently check, and read as though the rest of the paper were more accurate than the page you just put down — despite having no reason to think so. Crichton's point is that the one page where you could actually audit the source is precisely the page whose lesson you discard.

Mapped back: The newspaper is the multi-topic source; your professional field is the in-field check the paper flunks, the failed expert check. The next page's foreign-affairs story is the out-of-field exposure with no check. Filing "this story was wrong" as topic news is the level-of-abstraction mis-filing; not carrying it forward is the amnesic reset, leaving the un-discounted out-of-field trust.

Applied / In Practice

Courtroom expert testimony shows the effect operating with real stakes, and the legal system's response illustrates the corrective. A witness credentialed and genuinely reliable within a narrow specialty is routinely over-trusted by juries when they opine on adjacent matters they were never expert in — a forensic analyst validated on one technique testifying confidently about another, say. Modern admissibility gatekeeping under the US Daubert standard (Daubert v. Merrell Dow, 1993) pushes courts to scrutinize the reliability of the specific testimony and methodology at issue, not merely the witness's general credentials, and effective cross-examination increasingly anchors challenges to the precise claim rather than the diploma. That is exactly a shift from an aggregate "this expert is credible" rating to topic-banded evaluation.

Mapped back: The expert is the multi-topic source; the narrow specialty where their reliability can be tested is the in-field check, the adjacent opinion the out-of-field exposure. A jury crediting the credential wholesale is the un-discounted out-of-field trust produced by the level-of-abstraction mis-filing (trusting the person, not the specific claim). Daubert's claim-specific scrutiny is the source-level filing corrective: anchor credence to the checkable claim, not the aggregate reputation.

Structural Tensions

T1: Amnesic reset versus warranted topic-banding (the corrective looks like the disease). The effect condemns rating a source untrustworthy on topic A yet trustworthy on topic B — but treating an outlet as reliable on macroeconomics and unreliable on epidemiology can be entirely sound if grounded in per-beat evidence, and the prescribed fix (topic-banded credibility tracking) is itself topic-level filing. So the concept simultaneously indicts topic-level binning as the bug and prescribes topic-level binning as the cure. The line between them is whether the divergence is driven by which topic the reader can check (the amnesic reset) or by the source's actual per-beat reliability (legitimate calibration). The tension is that these look identical from outside — the same inconsistency across topics — and separating them requires knowing why the reader trusts differently, which the surface behavior does not reveal. Diagnostic: Is the reader's cross-topic trust difference tracking the source's real per-beat reliability, or merely which topic they happen to be able to audit?

T2: A clean in-field check versus its narrow reach (how far one beat licenses). The reader's own specialty is the maximally diagnostic place to audit a source — the one page they can actually verify. But it is a sample from a single beat, and propagating its verdict to a source-level prior assumes process quality is correlated across the outlet's other beats, which may not hold: a paper with a weak science desk can have a strong foreign desk staffed by different people. So the effect's own corrective — file the error at the source level and discount everything — risks over-propagating, condemning beats the failing desk never touched, just as the amnesia under-propagates. The tension is that the in-field check is both the cleanest evidence available and evidence about only one slice, and the framework does not supply the cross-beat correlation that would say how much A really licenses about B. Diagnostic: How correlated is process quality across this source's beats — does a failure on the checkable beat genuinely predict the unfamiliar one, or are they produced by independent processes?

T3: A precise diagnosis versus an impractical fix (the setting resists the cure). The effect localizes the failure exactly — evidence filed at the wrong level of abstraction — and shows that exhortation ("be more skeptical") cannot touch it, because the reader is categorizing, not suppressing. The only lever that bites is structural: replace aggregate trust with topic-banded tracking. Yet the canonical setting is casual, high-volume, low-stakes consumption — turning newspaper pages, scrolling a feed — where no reader maintains a per-source, per-beat credibility ledger, and the friction of doing so is prohibitive for the very rapid, adjacent-in-time reading that produces the effect. The tension is that the diagnosis rules out the only intervention (self-exhortation) that fits the setting, and prescribes one (instrumented banded tracking) that fits courtrooms and peer review but not the newsstand where the effect was named. Diagnostic: Does this context support maintaining topic-banded credibility records, or is it exactly the fast, casual consumption where the only feasible fix is unavailable?

T4: Expertise as the audit versus expertise as the vulnerability (depth cuts both ways). The effect requires a reader with a domain they can independently check — so it is precisely the deep expert, the person best equipped to catch the source's errors, in whom the effect bites hardest. Their sharp in-field check makes the failed audit maximally diagnostic, yet their equally wide out-of-field exposure means the discarded lesson leaves a large surface of un-discounted trust. A reader with no expertise has no diagnostic check to throw away and so cannot exhibit the effect at all. The tension is that the very capacity that should improve a reader's calibration — deep knowledge of one field — is what creates the vulnerability, because it manufactures the auditable event whose lesson is then routed into the wrong bin. Diagnostic: Is the reader's expertise being used to update trust in the source across topics, or only to detect the in-field error that then fails to propagate?

T5: Autonomy versus reduction (media-literacy effect or the instance of bin-segregation). "Gell-Mann amnesia" is a named, vivid media-literacy concept with its own cargo — the reader-turns-the-page narrative, the newspaper-beat framing, the topic-banded intervention vocabulary — and across human source-evaluation settings (news, peer review, expert testimony, influencers, Wikipedia) it transfers as mechanism, since each is one reader failing to propagate source-quality evidence across topic-bins. But beyond human multi-topic consumption the named effect travels only by analogy; what genuinely recurs is the parent — a bin-segregation failure in which evidence about an object filed in one bucket fails to update beliefs about the same object in another — the catalog region around base-rate neglect (the in-field sample is the source-quality base rate declined) and the silo effect. Institutional "forgetting" of a vendor's failures across departments, or trusting a model's out-of-sample predictions despite watching it miss the checkable cases, are co-instances of that parent, not of the eponymous effect. The tension is between a memorable named effect and the recognition that its cross-domain cargo belongs to the bin-mixing / base-rate-neglect parent. Diagnostic: Resolve toward the parent (bin-segregation / base-rate neglect) for non-human or institutional "forgetting" of an object's failures; toward the named Gell-Mann effect when diagnosing a human reader's trust in a multi-topic publication in situ.

Structural–Framed Character

The Gell-Mann amnesia effect sits at the mixed position: it is a genuine, discovered regularity of a natural system — the human trust-forming mind — rather than an institutional artifact, which pulls toward structure, but its every operative term is bound to human source-evaluation cognition, holding it short of the structural end. On evaluative_weight it carries a mild framed tint: "amnesia"/"effect"/"bug" mark a miscalibration — a failure to propagate evidence to the right level of abstraction — yet the label convicts no one, the entry being emphatic that this is "not an inconsistency to be scolded," "not motivated reasoning," and "structural, not motivated" (the reader "is not being hypocritical or careless"). On institutional_origin it points structural: no survey or agency constitutes the effect — Crichton (2002) named a thing minds already do, the way Schkade and Kahneman named the focusing effect; the eponymous coinage labels a pre-existing cognitive regularity, it does not manufacture one. The pivotal criterion is human_practice_bound, genuinely mixed: the effect is not constituted by an institution (turning a newspaper page merely triggers it), yet it is mind-bound — it runs only in a reader who forms and fails to update a source-level prior — so unlike isostasy it does not run observer-free in nature. On vocab_travels it scores framed: the in-field check, source-level versus topic-level filing, the amnesic reset, the reader-turns-the-page and newspaper-beat framing are all pinned to human consumption of multi-topic publications and lose their referents once that substrate is removed. On import_vs_recognize it is bimodal exactly as Knowledge Transfer describes: within human source-evaluation settings (news, peer review, expert testimony, influencers, brand trust, Wikipedia) the mechanism is recognized intact as one reader failing to propagate source-quality evidence across topic-bins; beyond human multi-topic consumption only the parent pattern recurs as co-instances (an institution "forgetting" a vendor's failures across departments, trusting a model's out-of-sample predictions despite watching it miss the checkable cases), imported by analogy, not the eponymous effect recognized in place.

The one portable skeleton is bin-segregation failure — evidence about an object filed in one bucket failing to update beliefs about the same object in another bucket, here process-quality evidence mis-binned as topic-specific content, the wrong level of abstraction for calibration. That skeleton is genuinely substrate-portable and recurs wherever a multi-topic generator is judged under partial observability. But it does not pull the effect off the mixed position, because it is precisely what the effect instantiates from its parents (the catalog region around base_rate_neglect — the in-field sample being the source-quality base rate the reader declines to update on — and the silo_effect of information siloed by topic rather than aggregated to the source), not what makes "Gell-Mann amnesia" itself travel: the cross-domain reach belongs to that bin-mixing / base-rate-neglect umbrella, while the effect's distinctive cargo — the reader-turns-the-page narrative, the newsroom-beat framing, the media-literacy intervention vocabulary — is exactly the part that stays home in human source-evaluation. Its character: a discovered, evaluatively mild, mind-bound regularity of human trust judgment, structural in the bin-segregation skeleton borrowed from its base-rate-neglect / silo-effect umbrella but framed by the source-credibility vocabulary that pins it to the reading mind.

Structural Core vs. Domain Accent

This section decides why the Gell-Mann amnesia effect is a domain-specific abstraction and not a prime — its portable core is a bin-segregation failure already housed in its parent primes, while its distinctive cargo is newsroom furniture bound to a human reading multi-topic publications.

What is skeletal (could lift toward a cross-domain prime). Strip the media-literacy framing and a thin relational structure survives: evidence about an object, filed in one bucket, fails to update beliefs about the same object in another bucket — process-quality evidence mis-binned as topic-specific content, the wrong level of abstraction for calibration. The portable pieces are abstract — a multi-attribute object, an observation that bears on the object as a whole, and a filing rule that quarantines it to one attribute. That skeleton is genuinely substrate-portable, which is why it recurs in the catalog as the parents the effect instantiates — the region around base_rate_neglect (the in-field sample is the source-quality base rate the reader declines to update on) and the silo_effect (information siloed by topic rather than aggregated to the whole). It is the core the Gell-Mann effect shares; it is not what makes it distinctive.

What is domain-bound. Everything that makes it the Gell-Mann amnesia effect in particular is source-credibility furniture, and its engine is not merely accented but substrate-restricting: the reader-turns-the-page narrative; the multi-topic source with beats only some of which a given reader can check; the in-field expert check and the failed expert check; the out-of-field exposure with no independent check; the amnesic reset; and the media-literacy corrective vocabulary of topic-banded credibility tracking. The decisive test: the effect requires a human forming and failing to update a source-level prior — a reader with a domain they can verify, a source that fails it, and a second domain they cannot check. Remove that reading mind (an institution that "forgets" a vendor's failures across departments, a model whose out-of-sample misses the analyst watches) and the named effect is gone, leaving a co-instance of the parent pattern under different clothing. The effect is bound to human consumption of multi-topic publications, exactly the substrate the prime bar would ask it to shed.

Why this does not clear the prime bar. A prime's vocabulary travels and its transfer is recognition of the same mechanism, not analogy. The effect's transfer is bimodal. Within human source-evaluation it travels as mechanism — the three-step signature, the adjacent-in-time diagnostic (divergent judgments seconds apart implicate a binning failure, not honestly differing evidence), and the structural corrective (topic-banded tracking, plus the prediction that fixing the second article does nothing) carry without translation across news media, peer review, expert testimony, influencer credibility, brand trust, and Wikipedia, because each is one reader failing to propagate source-quality evidence across topic-bins; that is recognition, but of application settings of one cognitive substrate, not distinct disciplines. Beyond human multi-topic consumption the named effect does not travel — institutional or model "forgetting" is the parent pattern by analogy, not the eponymous effect re-instantiated. And when the cross-domain lesson is wanted, it is already carried, in more general form, by the bin-segregation / base_rate_neglect / silo_effect umbrella, which needs no reader turning a page. The cross-domain reach belongs to those parents; the named entry carries newsroom cargo that should stay home. It is a memorable, useful diagnostic in situ, but its only substrate-spanning content is already the parents' — which is what keeps it below the prime bar.

Relationships to Other Abstractions

Local relationship map for Gell-Mann Amnesia EffectParents appear above the current abstraction, mutual partners to the right, and children below. Node labels state whether each abstraction is prime or domain-specific; colors identify relation types.Gell-MannAmnesia EffectDOMAINDomain-specific abstraction: Source Credibility — is part ofSourceCredibilityDOMAIN

Current abstraction Gell-Mann Amnesia Effect Domain-specific

Parents (1) — more general patterns this builds on

  • Gell-Mann Amnesia Effect is part of Source Credibility Domain-specific

    Source Credibility is the state variable the Gell-Mann Amnesia Effect fails to revise when a checkable article exposes errors in the source's process.

Hierarchy path (1) — routes to 1 parentless root

Not to Be Confused With

  • Appeal to authority / generic over-trust. Deference granted to an unexamined authority on the strength of credentials or status. The Gell-Mann twist is that the source has just been demonstrated to fail and the reader saw it — the trust is extended despite witnessed disconfirmation, which ordinary appeal-to-authority lacks. Tell: is trust granted to an authority never checked (appeal to authority), or extended to one whose unreliability was observed seconds earlier (Gell-Mann amnesia)?

  • Motivated reasoning / wishful thinking. Belief-updating distorted by a stake in the conclusion. The Gell-Mann reader has no stake in the unfamiliar-topic claim — nothing to rationalize away — so the failure is structural mis-filing, not motivated suppression, which is why exhorting honesty does not fix it. Tell: is the reader protecting a conclusion they want to be true (motivated reasoning), or neutrally binning source-quality evidence as topic news with no stake at all (Gell-Mann amnesia)?

  • Halo effect. A favorable impression of an entity in one respect spilling over to inflate judgments of unrelated traits — over-propagating a positive attribute too broadly. Gell-Mann amnesia is the mirror image: a demonstrated negative source-quality finding failing to propagate at all. Tell: is one good trait spreading trust too widely (halo effect), or a demonstrated bad-quality finding failing to spread to other topics (Gell-Mann amnesia)?

  • Source amnesia (source-monitoring failure). The genuine memory phenomenon of retaining a fact while forgetting where it came from. Gell-Mann's "amnesia" is metaphorical and almost opposite: the reader remembers the source's errors perfectly well but files them at the wrong level of abstraction (topic, not source-process), so the memory is intact and it is the propagation that fails. Namesake trap. Tell: is the origin of a remembered fact forgotten (source amnesia), or the source's errors remembered but not propagated to a source-level prior (Gell-Mann amnesia)?

  • Legitimate topic-banded credibility calibration. Rating an outlet reliable on one beat and unreliable on another when grounded in real per-beat evidence — sound, and indeed the effect's own prescribed corrective. Gell-Mann amnesia is the opposite: the same source rated inconsistently across topics consumed seconds apart, the inconsistency driven by which topic the reader can check, not by any real difference in the source's reliability. Tell: is the cross-topic trust difference tracking the source's actual per-beat reliability (legitimate calibration), or merely which topic the reader happens to be able to audit (Gell-Mann amnesia)?

  • base_rate_neglect / silo_effect (the parents). The substrate-neutral bin-segregation pattern the effect instantiates — evidence about an object filed in one bucket failing to update beliefs about the same object in another; the in-field sample being exactly the source-quality base rate the reader declines to use. These carry the lesson to non-human or institutional "forgetting." Tell: strip away the reader turning a page and what remains — process-quality evidence quarantined from a whole-object judgment — is carried by these parents, not by "Gell-Mann amnesia." (Treated fully in a later section.)

Neighborhood in Abstraction Space

Gell-Mann Amnesia Effect sits in a sparse region of the domain-specific corpus (67th percentile for distinctiveness): few abstractions share its structure, so a faithful description tends to retrieve it precisely.

Family — Journalistic Sourcing & Institutional Trust (13 abstractions)

Nearest neighbors

Computed from structural-signature embeddings · 2026-07-12