The Capability Atrophy Cascade

CV-023 documents a biological floor: the neurotransmitters, BDNF, sleep architecture, gut-brain axis, and chemical body burden that determine whether the brain can function at all. CV-021 documents a cognitive floor: the attentional, epistemic, social, and motivational prerequisites without which democratic function falls below threshold. Between these two floors is a specific, documentable layer that neither paper addresses — the layer where biological capacity becomes cognitive capacity through practice.

This is the skill/knowledge layer. It contains practiced capabilities — the judgment a physician builds through thousands of diagnoses, the craft a machinist develops through years of fabrication, the legal reasoning a judge accumulates through decades of adjudication, the writing capacity a policymaker needs to evaluate competing claims. These are not innate. They are not biological givens. They are constructed through sustained, deliberate practice on a functional biological substrate. And they are the infrastructure through which biological hardware becomes democratic prerequisite.

The distinction matters because interventions at different layers require different mechanisms. Restoring BDNF levels (hardware) does not automatically restore the professional judgment that took decades to build (skill/knowledge). Measuring prerequisite failure (CV-021) without understanding the skill layer that produces prerequisites mislocates the intervention point. The ICS corpus has documented the floor and the ceiling of cognitive capacity. This paper documents the staircase between them — and the compound process by which that staircase is collapsing.

HC-020 documents a non-linear capability depreciation mechanism with three phases. Phase 1: slow decline, as pre-automation practitioners compensate for gaps and maintain institutional knowledge through direct transmission. Phase 2: accelerating decline, as the compensating generation retires and the remaining practitioners cannot maintain transmission at scale. Phase 3: threshold crossing, as the practitioner base drops below the minimum required for tacit knowledge transmission and the capability enters irreversible decline.

The mechanism is cognitive offloading. When external systems handle tasks that humans previously performed, the brain reduces investment in maintaining the capability. This is not speculation. Ward et al. (2017, Journal of the Association for Consumer Research) demonstrated that the mere presence of a smartphone — even when powered off — reduces available cognitive capacity. The brain, detecting that an external system can handle the cognitive load, allocates resources away from the capability that the external system has assumed.

Dahmani and Bohbot (2020, Scientific Reports) provide longitudinal evidence in a different domain: GPS use correlated with spatial memory decline at r=−0.68 over 3.2 years (N=50 cross-sectional, N=13 longitudinal retest). The dose-dependent relationship is the depreciation curve in miniature — more offloading, more degradation, measured across time.

Dell’Acqua et al. (2023, Harvard Business School) document the frontier effect: among 758 BCG consultants, those using AI completed tasks 12.2% faster and at 40% higher quality — but when tasks fell outside AI’s capability frontier, AI access worsened performance by 19 percentage points. The capability that AI replaces does not survive the replacement. When the AI fails, the human fails with it.

HC-020 identifies four leading indicators for tracking the depreciation curve at population scale: declining apprenticeship registrations, declining deliberate practice hours in professional training, declining entry-level hiring standards, and declining performance on 30-day delayed competency tests (the interval at which short-term workarounds expire). The FAA has already detected this pattern in aviation: Advisory Circular 120-111 mandates minimum manual flight hours to counteract automation-driven skill degradation in pilots. No equivalent mandate exists for cognitive faculties across domains.

The PISA data presents a similar complexity. OECD assessments show a cross-sectional dose-response relationship within the 2022 cohort: students with less than one hour of daily leisure device time scored approximately 50 points higher in mathematics than those exceeding five hours per day. Students distracted by peers’ devices scored approximately 15 points lower — equivalent to three-quarters of a school year. But the 15-point temporal decline between 2018 and 2022 cannot be causally decomposed by technology-dependence variables. COVID-19 disruption is the dominant confounding factor. The dose-response data is correlational. It does not prove that technology dependence caused the historical decline.

Prior automation waves targeted domain-specific capabilities sequentially. The Jacquard loom automated weaving. The assembly line automated sequential manufacturing. The spreadsheet automated tabular calculation. In each case, one domain’s capabilities were replaced while other domains’ capabilities remained unaffected. Mokyr (1990, The Lever of Riches) documents this pattern across the full history of technological displacement: each wave was domain-limited, sequential, and bounded.

HC-023 documents the architectural shift. Current AI systems target domain-general faculties — language, reasoning, pattern recognition, judgment — not domain-specific skills. These faculties are the shared cognitive substrate across all knowledge-work domains. A lawyer uses the same language faculty as a physician. A policymaker uses the same reasoning faculty as an engineer. A journalist uses the same judgment faculty as a financial analyst. When AI automates language itself, every domain that depends on language is affected simultaneously.

HC-023 audits eight domains currently experiencing concurrent AI automation: legal reasoning, medical diagnosis, financial analysis, software development, educational assessment, scientific writing, policy analysis, and creative production. This is not a forecast. McKinsey Global Institute (2023) and the World Economic Forum (2025) project future workforce impact, and those projections should be treated as projections, not documented simultaneity. But the observation that AI systems are already deployed across all eight domains simultaneously — handling language, reasoning, and judgment tasks in each — is verifiable in the present tense.

Haldane and May (2011, Nature) provide a structural analogy from banking networks: when financial institutions share common exposures, a stress that affects one institution propagates through the shared substrate to all institutions simultaneously. The systemic risk is architectural, not incidental — it emerges from the shared structure, not from the individual failures. CV-026 applies this structural insight to the cognitive domain: when the same faculties underlie all knowledge work, atrophy in the shared faculty propagates to every domain that depends on it. This is an analogical application of Haldane and May’s framework, not a mathematical one. The claim is architectural — shared substrate produces correlated failure — not that banking equations apply to cognition.

Autor and Dorn (2013, American Economic Review) documented the hollowing of middle-skill jobs through routine task automation. Autor (2024) extends this: automation innovations have shown intensifying demand-eroding effects over four decades. But Autor also notes that AI could reverse polarization with deliberate policy — the market alone does not produce that outcome. This is consistent with HC-021’s Non-Reconstruction Principle: recovery does not occur under market conditions.

HC-021 documents the condition under which capability atrophy becomes permanent: when tacit knowledge — the knowledge embedded in practice, in muscle memory, in professional intuition that cannot be fully articulated or codified — is lost through automation-driven disuse, it is not rebuilt under market conditions.

The theoretical foundation is Polanyi (1966, The Tacit Dimension): “We can know more than we can tell.” Collins (2010, Tacit and Explicit Knowledge) extends the taxonomy to somatic tacit knowledge (embodied skill) and collective tacit knowledge (knowledge distributed across a community of practice). Both forms depend on sustained practice for maintenance. Both are lost through disuse. Neither can be reconstructed from codified records alone, because the knowledge was never fully codified in the first place.

HC-021 identifies four recovery conditions that must coexist for tacit knowledge to be rebuilt: (1) recognition that the knowledge has been lost, (2) institutional will to invest in recovery, (3) surviving practitioners who can transmit the knowledge, and (4) cultural valuation of the capability sufficient to sustain the recovery effort. HC-021’s finding: these four conditions never coexist at the moment of need. By the time the loss is recognized, the surviving practitioners have retired or died. By the time institutional will emerges, the cultural valuation has shifted to the replacement technology.

Budzyn et al. (2025) provide evidence that the irreversibility gate may operate faster than HC-020’s three-phase model suggests. Endoscopists showed measurable deskilling after AI exposure within the timeframe of a single clinical trial — not after a generation of automation, but within months. If the depreciation curve begins this quickly, the window for intervention narrows accordingly.

Brynjolfsson, Li, and Raymond (2023, NBER) document a related mechanism from the other direction: among 5,179 customer support agents, AI produced a 34% improvement for novice workers by providing best-practice responses in real time. The implication: novice workers may never develop the diagnostic judgment that experienced agents built through years of practice, because the AI provides the answer before the cognitive work of developing it occurs. The novice performs well — but does not build the tacit knowledge that would allow independent performance if the AI were removed. Goldman Sachs’s equity trading floor illustrates the endpoint: at peak in 2000, approximately 600 traders; by 2017, two equity traders and 200 computer engineers (Chavez, Harvard IACS, January 2017). The tacit knowledge of 598 traders did not transfer to the two who remained or to the 200 engineers who replaced them.

The preceding sections document individual mechanisms: the depreciation curve (HC-020), the common faculty architecture (HC-023), the irreversibility gate (HC-021). Each is documented independently. This section documents what happens when they interact — and this interaction is CV-026’s unique analytical contribution. The compound cascade model that follows is a proposed theoretical framework, not an established empirical finding. Evidence classification is provided for each pathway.

Tier A Language atrophy propagates across domains. Sparrow, Liu, and Wegner (2011, Science) established that people do not encode information they expect to remain externally accessible — the brain offloads to the external system. When language production is offloaded to AI (writing assistance, auto-completion, summarization), the practiced capability of precise written expression atrophies. A lawyer who cannot write clearly cannot construct an argument. A physician who cannot articulate a differential diagnosis cannot practice clinical reasoning. A policymaker who cannot evaluate competing written claims cannot govern. The language faculty is shared infrastructure. Its atrophy carries into every domain.

Tier A Automation bias accelerates the depreciation curve. HC-018 documents the complacency paradox: Goddard, Roudsari, and Wyatt (2012, BMJ Quality & Safety) found an automation bias effect of d=0.74 on decision quality — a large effect in which higher AI accuracy produces decreased human oversight quality. This is not a failure of attention. It is a rational metacognitive response: when the system is usually right, the effort of independent verification yields diminishing returns. But the consequence is that the very accuracy that makes AI useful also reduces the conditions for maintaining human capability. The more AI improves, the faster the depreciation curve accelerates. Air France Flight 447 (228 deaths, BEA 2012) and the Boeing 737 MAX (346 deaths, MCAS authority-capability gap) document the consequence when automated systems fail and the humans who should intervene have atrophied capabilities.

Tier B Cross-domain reasoning atrophy compounds non-linearly. Taatgen (2013, Psychological Review) proposes the PRIMs (Primitive Information Processing Elements) architecture: cognitive skills that share underlying processing elements transfer between domains. The reverse implication — that atrophy of shared elements would propagate across domains — is theoretically grounded in the architecture but has not been empirically tested for degradation. Hanushek et al. (2025, Science Advances, N=3,263 adults) found that both literacy and numeracy decline with below-average skill usage, documenting parallel disuse effects. But no evidence shows that decline in one domain causes decline in the other. The parallel disuse is consistent with the cascade hypothesis but does not confirm cross-domain causation.

Tier C The compound interaction exceeds any source paper’s prediction. When HC-020’s depreciation curve operates on HC-023’s cross-domain architecture, through HC-021’s irreversibility gate, while HC-022’s backup vacuum eliminates redundancy — the result is a compound process that no single source paper documents. The depreciation in one domain cannot be quarantined because the same faculty is involved in all domains. The irreversibility applies across all domains simultaneously. The backup failure is systemic, not local. This compound interaction model is CV-026’s analytical framework. It is consistent with the documented individual mechanisms. It has not been empirically tested as a compound.

HC-022 documents what happens when the cascade reaches the point of system failure: the Flash Crash ($1 trillion in market value lost in 36 minutes), Knight Capital ($440 million in 45 minutes, SEC Release 34-70694), the Boeing 737 MAX (346 deaths), the 2003 Northeast Blackout (55 million affected, triggered by an alarm system software bug). These are domain-specific failures, not cross-domain cascade evidence. They demonstrate that single-point fragility in individual systems has catastrophic consequences. CV-026 argues that the common faculty architecture means this fragility pattern will manifest across all domains as the cascade progresses — an analytical extension of HC-022’s documented findings, not a direct empirical claim.

The Capability Crisis series documents the social manifestation of this cascade already in progress. These are not separate crises. They are Stage 1 indicators of the Common Faculty Depreciation — the observable output of practiced capabilities atrophying at the population level.

CC-001 documents the readiness crisis: 77% of Americans aged 17–24 are ineligible for military service (DOD Qualified Military Available study, 2020; up from 71% in 2017). Only 1% are both eligible and inclined. The disqualifiers are not purely physical: 21.1% youth obesity prevalence (2021–2023), but also cognitive and behavioral factors that reflect the compound degradation documented in CV-023 and the capability atrophy documented here. The military is one of the few institutions that maintains standardized readiness benchmarks across decades — providing a rare longitudinal signal in a landscape where most capability measures are domain-specific and incomparable across time.

CC-002 documents the hollow pipeline: 3.5 million unfilled skilled trades positions alongside $1.77 trillion in student debt and 52% graduate underemployment (Burning Glass/Strada 2024). Forty-five percent of underemployed graduates remain underemployed after a decade. Germany’s dual system — in which approximately 50% of young people enter apprenticeship — provides the counterfactual: a system that maintained skill transmission infrastructure has a different capability profile. The American dismantling followed a specific timeline: the Smith-Hughes Act (1917) built it, A Nation at Risk (1983) began redirecting it, the College-for-All consensus (1990s) abandoned it, and No Child Left Behind (2001) completed the institutional pivot away from skill transmission toward credential production.

CC-004 provides the synthesis that connects the CC data to CV-026’s framework. CC-004 identifies that demand removal and attention capture are “distinct but convergent” processes operating on the same cohort during the same decades — roughly 1975 through 2005. The demand for practiced capability was removed by institutional restructuring (credential over competence) while the attention infrastructure was captured by algorithmic systems. CV-026 closes this observation: the two processes converge because both drive the depreciation curve. Demand removal eliminates the conditions for practice. Attention capture eliminates the cognitive resources practice requires. The result is HC-020’s Phase 2 — accelerating decline — driven from two directions simultaneously.

CC-003 maps three removals that compound the cascade: the removal of consequence from failure, meaning from struggle, and obligation from citizenship. The Victorian Britain parallel is instructive: when 40% of Boer War recruits were rejected on physical grounds, the 1904 Inter-Departmental Committee on Physical Deterioration led to institutional reforms (school meals, medical inspections, physical education) that produced a capable fighting force within a generation. The parallel confirms that capability degradation is reversible — but only through institutional intervention, not market dynamics. It also confirms the timeline: even with institutional will, recovery required a generation.

The cascade documented in this paper is invisible to existing measurement systems. This is not an accident. It is a structural consequence of how capability is measured.

EI-004 documents a seven-stage loop through which detection instruments become captured by the institutions they were designed to measure: Real Discovery → Institutional Formation → Frontier Migration → Survival Mechanism Activation → Outsider Suppression → Scale Escalation → Redefinition of Success. EI-004 documents this loop in three domains: particle physics, psychiatry, and consciousness research. CV-026 proposes that the same loop operates in workforce analytics and capability measurement — an analytical extension of EI-004’s framework, not a documented instance.

The proposed application: workforce analytics measures domain-specific outcomes — productivity metrics, output volume, error rates, credential completion — not the shared cognitive substrate beneath all domains. When AI boosts productivity metrics (Dell’Acqua et al. 2023: 12.2% more tasks, 25.1% faster, 40% higher quality), the measurement system records improvement. The depreciation of human capability that the improvement masks is invisible to the metric. The instrument has been captured by the measurement paradigm: what the instrument can see (output) becomes the definition of capability, and what it cannot see (the human capacity to perform without the AI) disappears from institutional view.

EI-005 explains why researchers who might identify the cross-domain pattern face structural disincentives. Academic careers are organized by domain. A computer scientist studying automation bias has no institutional incentive — and no disciplinary infrastructure — to collaborate with a labor economist studying skill polarization, a neuroscientist studying cognitive offloading, and an education researcher studying credential inflation. The cross-domain cascade is invisible not because the evidence is absent but because no discipline is positioned to see it.

This section is the structural centerpiece of CV-026. It maps four specific pathways through which CV-023’s biological degradation translates through the skill/knowledge layer into CV-021’s prerequisite failure. Each pathway is classified by evidence tier. The first two are well-documented. The third is mechanistic inference. The fourth is a proposed analytical framework. The strength of CV-026’s argument does not depend on all four pathways being equally established. It depends on the first two being sufficient to demonstrate the translation mechanism, with the third and fourth offered as theoretically grounded extensions.

Pathway	CV-023 Vector → CV-026 Atrophy → CV-021 Failure	Evidence
Tier A	Degraded sleep architecture + D2 receptor downregulation → reduced capacity for sustained practice (Walker 2017; Budzyn et al. 2025) → accelerated depreciation curve → attentional prerequisite failure	Sleep → attention is well-established. Reduced attention → reduced practice capacity is directly measurable. Budzyn 2025 provides clinical deskilling evidence.
Tier A	BDNF deficit + sedentary environment → reduced neuroplasticity for skill acquisition (Cotman & Berchtold 2002; Erickson et al. 2011) → tacit knowledge non-transmission → epistemic prerequisite failure	BDNF-neuroplasticity link is robust. Reduced plasticity → impaired skill acquisition is established in motor learning literature.
Tier B	Chronic cortisol + gut dysbiosis → impaired prefrontal function and social cognition → degraded professional judgment → social prerequisite failure	Chronic cortisol → impaired prefrontal function is documented (McEwen 2006). Prefrontal function → professional judgment is logically necessary. The specific chain (cortisol → social cognition → domain judgment) is inferred, not directly demonstrated as a unit.
Tier C	Wellness inversion + chemical body burden → reduced motivational capacity for demanding work → demand removal normalization → motivational prerequisite failure	Proposed framework synthesizing CV-023’s wellness data with CC-004’s demand removal mechanism. No single study documents this chain. Consistent with available evidence but unverified as a compound.

The first pathway is the most thoroughly documented. CV-023 establishes that sleep architecture is degraded by blue-spectrum light exposure (Chang et al. 2015), algorithmic engagement extending waking hours (NR-001), and chronic stress elevating cortisol (McEwen 2006). Walker (2017) documents the cognitive consequences: degraded sleep impairs attention, memory consolidation, and executive function. Budzyn et al. (2025) demonstrate that these cognitive impairments produce measurable deskilling in clinical practice. The chain from degraded sleep to degraded professional capability to degraded democratic prerequisite is documented at each link.

The second pathway operates through neuroplasticity. CV-023 documents the BDNF deficit from sedentary environments (IT-003). Cotman and Berchtold (2002, Trends in Neurosciences) establish that BDNF is the primary molecular mechanism for experience-dependent neuroplasticity. Erickson et al. (2011, PNAS) demonstrated that aerobic exercise increases hippocampal volume by 2% in older adults. The reverse is the relevant finding: reduced physical activity reduces the neuroplasticity on which skill acquisition depends. When the brain’s capacity for new learning is diminished at the hardware level, the skill/knowledge layer cannot maintain or build the capabilities that the prerequisite layer requires.

The third pathway is mechanistic inference. McEwen (2006) documents that chronic cortisol elevation produces hippocampal atrophy, prefrontal remodeling, and amygdala hypertrophy. These structural changes degrade the cognitive architecture for social reasoning and emotional regulation. The inference — that degraded social cognition degrades professional judgment in domains requiring interpersonal assessment — is logically necessary (a judge cannot assess credibility without social cognition) but has not been directly demonstrated as a causal chain from cortisol to domain-specific judgment failure.

The fourth pathway is the weakest. The connection between chemical body burden, motivational capacity, and the normalization of demand removal is a proposed framework. CC-004 documents demand removal as a historical process. CV-023 documents wellness system capture. The synthesis — that biological degradation reduces the motivational substrate on which demanding practice depends, contributing to cultural acceptance of reduced capability demands — is consistent with the evidence but untested. It is included here as a theoretical extension, not an empirical claim.

HC-023 coined “The Common Faculty Problem” to identify the architectural vulnerability: that current AI targets domain-general faculties rather than domain-specific skills. CV-026’s “Common Faculty Depreciation” names what happens when that vulnerability is exploited by the depreciation curve over time. The Problem is the condition. The Depreciation is the process. The Problem is static — it exists as long as AI targets common faculties. The Depreciation is dynamic — it unfolds along HC-020’s non-linear curve, through HC-021’s irreversibility gate, producing HC-022’s backup vacuum, while EI-004’s detection failure ensures the process remains invisible to existing measurement.

The relationship to CV-023 and CV-021 is directional. CV-023 documents what degrades (biological substrate). CV-021 documents what fails (democratic prerequisites). CV-026 documents how the degradation becomes the failure — the specific translation through practiced capabilities, the mechanisms that make the translation irreversible, and the detection failure that makes the cascade invisible. Without this paper, the corpus documents a broken floor and a collapsing ceiling with no explanation of the structural failure that connects them.

Not a summary of HC-020 through HC-023. Those papers document individual stages of a collapse gradient. CV-026 documents the compound interaction across stages when they hit the same substrate simultaneously. The individual mechanisms are the inputs. The cascade is the finding.

Not an argument against AI. The target is extractive deployment specifically — AI deployed for efficiency optimization without capability preservation design, which HC-020 identifies as Stage 0. AI that preserves and enhances human capability (deployment inversion, HC-003) is explicitly excluded from the critique. The paper is about the architecture of deployment, not the technology itself.

Not a claim that the cascade is primarily AI-driven. CC-004’s demand removal timeline begins in the 1970s — decades before widespread AI deployment. The institutional dismantling of skill transmission infrastructure (Smith-Hughes to NCLB), the cultural shift from competence to credential, and the economic hollowing of middle-skill jobs (Autor 2013) all predate AI. AI accelerates and compounds a pre-existing cascade. It did not create it.

Not a prediction of civilizational collapse. Every stage documented here has detectable leading indicators (HC-024a) and identifiable intervention points. The Victorian Britain parallel (CC-003) demonstrates that capability degradation is reversible through institutional intervention. The FAA’s mandated manual flight practice demonstrates that detection and mitigation are possible within a single domain. The paper documents the cascade to enable detection and response, not to argue helplessness.

Not a claim that the cascade is inevitable. The compound interaction model includes evidence at multiple tiers — from well-documented individual pathways (Tier A) to proposed analytical frameworks (Tier C). The honest presentation of evidence classification throughout this paper is itself a design choice: the cascade is documentable and detectable, the intervention points are identifiable, and the compound model invites empirical testing rather than fatalistic acceptance. The Common Faculty Depreciation is named so that it can be measured, tracked, and interrupted.

Cognitive Offloading & Deskilling

Adrian F. Ward et al., “Brain Drain: The Mere Presence of One’s Own Smartphone Reduces Available Cognitive Capacity,” Journal of the Association for Consumer Research 2, no. 2 (2017): 140–154.

Betsy Sparrow, Jenny Liu, and Daniel M. Wegner, “Google Effects on Memory: Cognitive Consequences of Having Information at Our Fingertips,” Science 333, no. 6043 (2011): 776–778.

Ewa Risko and Sam Gilbert, “Cognitive Offloading,” Trends in Cognitive Sciences 20, no. 9 (2016): 676–688.

Aleksandra Budzyn et al., “Endoscopist Deskilling Risk After Exposure to Artificial Intelligence in Colonoscopy,” The Lancet Gastroenterology & Hepatology 10, no. 10 (2025): 896–903.

Fabian Dell’Acqua et al., “Navigating the Jagged Technological Frontier: Field Experimental Evidence of the Effects of AI on Knowledge Worker Productivity and Quality,” Harvard Business School Working Paper 24-013 (2023).

Erik Brynjolfsson, Danielle Li, and Lindsey R. Raymond, “Generative AI at Work,” NBER Working Paper 31161 (2023).

Louisa Dahmani and Véronique D. Bohbot, “Habitual Use of GPS Negatively Impacts Spatial Memory During Self-Guided Navigation,” Scientific Reports 10, 6310 (2020).

Automation Bias & System Failure

Charlotte Goddard, Reza Roudsari, and Jeremy C. Wyatt, “Automation Bias: A Systematic Review of Frequency, Effect Mediators, and Mitigators,” BMJ Quality & Safety 21 (2012): 592–600.

Mary L. Cummings and Marc Britton, “Regulating Next-Generation Artificial Intelligence,” Duke University (2020).

BEA, “Final Report on the Accident on 1st June 2009 to the Airbus A330-203 Registered F-GZCP Operated by Air France Flight AF 447” (2012).

Securities and Exchange Commission, “In the Matter of Knight Capital Americas LLC,” SEC Release No. 34-70694 (2013).

Tacit Knowledge & Irreversibility

Michael Polanyi, The Tacit Dimension (University of Chicago Press, 1966).

Harry Collins, Tacit and Explicit Knowledge (University of Chicago Press, 2010).

Marty Chavez, Goldman Sachs, Harvard IACS presentation (January 2017); reported in MIT Technology Review (February 7, 2017).

Cross-Domain Architecture & Labor

Joel Mokyr, The Lever of Riches: Technological Creativity and Economic Progress (Oxford University Press, 1990).

Andrew G. Haldane and Robert M. May, “Systemic Risk in Banking Ecosystems,” Nature 469 (2011): 351–355.

David H. Autor and David Dorn, “The Growth of Low-Skill Service Jobs and the Polarization of the US Labor Market,” American Economic Review 103, no. 5 (2013): 1553–1597.

David H. Autor, “Applying AI to Rebuild Middle Class Jobs,” NBER Working Paper 32140 (2024).

McKinsey Global Institute, “The Economic Potential of Generative AI” (2023).

World Economic Forum, “Future of Jobs Report 2025” (2025).

Cognitive Transfer & Skill Degradation

Niels Taatgen, “The Nature and Transfer of Cognitive Skills,” Psychological Review 120, no. 3 (2013): 439–471.

Eric A. Hanushek et al., “Age and Cognitive Skills: Use It or Lose It,” Science Advances 11, eads1560 (2025).

Giovanni Sala and Fernand Gobet, “Cognitive Training Does Not Enhance General Cognition,” Trends in Cognitive Sciences 23, no. 1 (2019): 9–20.

PISA & Educational Measurement

OECD, “PISA 2022 Results” (2023/2024).

FAA Advisory Circular 120-111, “Upset Prevention and Recovery Training” (2015).

DOD Qualified Military Available Study (2017, 2020).

Burning Glass Institute / Strada Education Foundation, “The Permanent Detour” (2024).

Biological Substrate

Matthew Walker, Why We Sleep: Unlocking the Power of Sleep and Dreams (Scribner, 2017).

Carl W. Cotman and Nicole C. Berchtold, “Exercise: A Behavioral Intervention to Enhance Brain Health and Plasticity,” Trends in Neurosciences 25 (2002).

Kirk I. Erickson et al., “Exercise Training Increases Size of Hippocampus and Improves Memory,” PNAS 108, no. 7 (2011): 3017–3022.

Bruce S. McEwen, “Protective and Damaging Effects of Stress Mediators,” Dialogues in Clinical Neuroscience 8 (2006).

ICS Cross-References

HC-020: The Capability Atrophy Mechanism — The Depreciation Curve.

HC-021: The Tacit Knowledge Problem — The Non-Reconstruction Principle.

HC-022: The Single-Point Fragility Record — The Human Backup Vacuum.

HC-023: The Common Faculty Problem — The Common Faculty Problem.

HC-018: The Automation Bias Record — The Complacency Paradox.

HC-024a: The Early Warning Record — The Stage Indicators.

CC-001: The Readiness Crisis — The Readiness Crisis.

CC-002: The Hollow Pipeline — The Hollow Pipeline.

CC-003: The Engineered Softness — Engineered Softness.

CC-004: The Collapse Is One Event — Demand Removal.

EI-004: The Instrument Capture Loop — The Instrument Capture Loop.

EI-005: The Incentive Architecture — The Incentive Architecture.

CV-021: The Epistemic Floor Collapse — The Cognitive Prerequisites Failure.

CV-023: The Biological Substrate Erasure — The Compound Biological Degradation.