AI targets language, reasoning, and judgment — the same cognitive substrate across all domains simultaneously. If this atrophies, it does not atrophy in one domain. It atrophies in the human.
Every prior wave of automation targeted different capabilities in each domain. The mechanization of agriculture targeted physical labor in farming. The automation of manufacturing targeted repetitive motor tasks in factories. The computerization of finance targeted arithmetic and record-keeping in banking. Each wave was domain-specific: the capability it replaced in one domain was different from the capability it replaced in another.
The current wave of AI automation is structurally different. It targets language, reasoning, and judgment — cognitive faculties that are not domain-specific but domain-general. The same underlying capabilities that AI automates in legal analysis are the same capabilities it automates in medical diagnosis, financial assessment, educational instruction, software engineering, scientific research, creative production, and policy analysis. The faculties are common. The automation is simultaneous.
This is not a claim about “unprecedented simultaneity” — that framing is historically imprecise and invites objection. It is a specific structural observation: prior automation targeted domain-specific capabilities sequentially; current automation targets common cognitive faculties concurrently. The distinction is not about speed or scale. It is about the architecture of what is being automated.
Prior automation targeted different capabilities in each domain: physical strength in manufacturing, arithmetic in finance, repetitive precision in assembly. The current wave targets language, reasoning, and judgment — the same cognitive substrate across all domains. The fragility is not cross-domain. It is cross-capability: single-point fragility at the level of human cognition itself.
When Stage 2 (tacit knowledge non-transmission) and Stage 3 (single-point fragility) arrive in one domain, the damage is contained. When they arrive in multiple domains simultaneously because the same underlying faculty is atrophying across all of them, the damage compounds. The common faculty is the single point of failure — and it fails everywhere at once.
Joel Mokyr (1990), in The Lever of Riches: Technological Creativity and Economic Progress, documented the historical pattern of automation waves targeting domain-specific capabilities sequentially. The mechanization of textile production (1760s–1840s) targeted spinning and weaving — specific manual skills in a specific industry. The resistance and adaptation played out within that industry over decades. The resistance to automation in agriculture (1850s–1950s) was a separate process targeting different capabilities in a different domain on a different timeline.
This sequential, domain-specific pattern meant that the human cognitive infrastructure was never simultaneously stressed across multiple domains. When agricultural labor was automated, manufacturing absorbed the displaced workforce — using different capabilities. When manufacturing was automated, services absorbed the displaced workforce — again using different capabilities. The common cognitive faculties — language, reasoning, judgment — were never the target. They were the substrate that enabled adaptation.
The historical resilience of human economies to automation waves depended on the availability of an unautomated cognitive substrate. Workers displaced from automated domains could retrain because the faculties required for retraining — language comprehension, logical reasoning, judgment under uncertainty — were not themselves being automated. The adaptation mechanism assumed the persistence of the adaptation substrate.
The common faculty problem is what happens when the adaptation substrate itself becomes the target of automation. If the faculties that enable retraining, career transition, and cognitive flexibility are themselves atrophying under automation-driven disuse, the historical adaptation mechanism does not function.
The World Economic Forum Future of Jobs Report (2025) documents the simultaneous feasibility of automation across professional domains that share no domain-specific skills but share cognitive faculties. Legal analysis, medical diagnosis, financial assessment, software development, educational instruction, scientific literature review, policy analysis, and creative writing are all experiencing automation of language processing, pattern recognition, and evaluative judgment at comparable timescales.
McKinsey Global Institute (2023) documented this cross-domain simultaneity in workforce impact projections: the domains experiencing the largest automation exposure share language and reasoning as their primary cognitive substrates, despite having no overlap in domain-specific technical skills.
The specificity of this observation matters. The claim is not that AI automates “everything.” The claim is that AI automates a specific set of cognitive faculties — language processing, pattern-based reasoning, and evaluative judgment — that happen to be the same faculties used across all knowledge-work domains. The targeting is specific. The impact is general because the target is general.
Analytical status and counterexamples. The Common Faculty Problem is a proposed analytical category, not an established finding in the automation literature. The claim that prior automation waves did not target common cognitive faculties is a simplification that requires qualification: the printing press transformed memory practices and oral tradition; calculators displaced mental arithmetic; GPS has measurably reduced navigational cognition (Bohbot et al., 2017). The distinction proposed here is one of scope and simultaneity — not that prior technologies left cognition untouched, but that none targeted multiple core faculties concurrently through a single delivery mechanism operating across all knowledge-work domains at comparable timescales. Whether this difference in degree constitutes a difference in kind is an empirical question that the current evidence base cannot definitively answer.
Andrew Haldane and Robert May (2011), in their Nature paper on systemic risk in banking networks, established a framework for understanding how interconnected systems amplify failure. Their insight: in highly interconnected networks, the failure of one node propagates to other nodes not because the nodes are similar but because they share common dependencies. The risk is in the shared infrastructure, not in the individual components.
Applied to the common faculty problem: the cognitive faculties targeted by AI automation are the shared infrastructure of all knowledge-work domains. Language, reasoning, and judgment are the common dependencies. If these faculties atrophy through Stages 1–3, the atrophy propagates across all domains that depend on them — not because the domains are similar, but because they share the same cognitive substrate.
This is why Stage 4 is qualitatively different from Stage 3. Stage 3 (single-point fragility) is domain-contained. A failure in automated trading does not directly cause a failure in automated medical diagnosis. Stage 4 is substrate-level: if the common cognitive faculties atrophy, they atrophy in the human, and the human carries that atrophy into every domain they enter.
The fragility is not that AI fails in many domains. The fragility is that the same human faculty fails in all of them — because it was the same faculty in all of them, and it atrophied once.
Stage 4 can be detected through cross-domain correlation of Stage 2–3 indicators:
Cross-domain Stage 2 correlation. If tacit knowledge non-transmission (Stage 2) arrives in legal practice, medical training, software engineering, and financial analysis within the same decade, the common faculty mechanism is the most parsimonious explanation. Domain-specific causes would produce domain-specific timelines.
Common capability decline in new workforce entrants. If new practitioners across multiple unrelated domains show declining performance in the same cognitive capabilities — independent reasoning, unassisted writing, judgment under ambiguity — the decline is not domain-specific. It is faculty-level.
Adaptation failure. If workers displaced by AI automation in one domain show reduced capacity to retrain for non-automated roles — specifically in the cognitive faculties required for retraining rather than in domain-specific skills — the adaptation substrate is degrading.
Simultaneous dependency normalization. “Can’t do it without AI” appearing as a normalized response in workforce surveys across multiple unrelated professional fields at comparable timepoints signals that the common faculty, not domain-specific skill, is the site of atrophy.
The collapse gradient from Stage 0 (Extractive Deployment) through Stage 4 (The Common Faculty Problem) maps a trajectory from recoverable atrophy to civilizational fragility. HC-024a (The Early Warning Record) makes this gradient empirically testable — specifying the measurements, thresholds, and institutional mechanisms required to detect each stage transition before it becomes irreversible. HC-024 (What Prevention Actually Requires) specifies the structural conditions that arrest the gradient: the resilience floor that ensures human capability persists alongside AI capability.
The gradient is not inevitable. Every stage transition has detectable leading indicators. Every stage has intervention points. But the interventions become more expensive and less effective at each stage. The argument for early action is not precautionary. It is economic: acting at Stage 1 costs orders of magnitude less than responding at Stage 3.
Internal: This paper is part of The Collaboration (HC series), Saga XI. It draws on and contributes to the argument documented across 31 papers in 2 series.
External references for this paper are in development. The Institute’s reference program is adding formal academic citations across the corpus. Priority papers (P0/P1) have complete references sections.