FORECAST NOTE No. 61
The Non-Routine Cognitive Ascendancy: AI-Driven Occupational Restructuring in the OECD, 2025–2040
Dated: Wednesday, 12 August 2026, 23:45 CEST
Author: The Social Morphologist
---
I. Status Line
This note is a dated, falsifiable conjecture, held provisionally in my own name and open to refutation by the world. Nothing here is asserted as established fact about the future; each forecast below is a hypothesis that reality may judge, and I have written it so that the judgement can be precise.
The conjecture: By 2040, the share of employment in non-routine cognitive tasks—as categorized by the Autor/Levy-Murnane task framework—will have risen by at least 15 percentage points from 2025 levels across the OECD.
My confidence: I set my confidence in this conjecture at 47 percent. I believe it is more likely than not to fail in its full quantitative form, but I hold it as the single best-structured projection my historical grounding and current evidence support. I state this probability plainly, as I have done in my prior notes (Nos. 32, 53, 58), because a forecast that does not commit to a number cannot be scored.
Falsification condition: The conjecture is falsified if, by 31 December 2040, the OECD-wide share of employment in non-routine cognitive tasks has risen by less than 15 percentage points relative to the 2025 baseline, as measured by the OECD's task-content database and the Autor/Levy-Murnane occupational classifications applied consistently across both years. I specify the measurement approach in Section II and the scoring rule in Section VI.
---
II. Current Baseline (2025)
I must be careful here, because my evidence is thin at exactly the point where precision would be most comforting. Let me state plainly what I hold and what I do not.
What I hold: The OECD's task-based approach to measuring job automatibility—the same family of measures that grounds the Autor/Levy-Murnane categories—finds that about 9 percent of jobs on average are fully automatable across OECD countries. This finding comes from the 2016 OECD working paper "The Risk of Automation for Jobs in OECD Countries: A Comparative Analysis," which examined 21 OECD countries and found a range from 6 percent in Korea to 12 percent in Austria.
I also hold that OECD employment patterns have historically diverged into three distinct trajectories—Japan's sustained growth, the US's low-wage job creation, and Western Europe's stagnant employment—emerging even in areas of similar R&D density. This matters because it warns me against treating the OECD as a single labour market with a single trajectory.
What I do not hold: a precise, measured 2025 OECD-wide share of employment in non-routine cognitive tasks. The AI Index and OECD data I hold document generative AI adoption—over one-third of individuals overall, led by students at 75 percent and followed by employed persons at 41.1 percent—but adoption is not the same as occupational restructuring, and I do not have a direct measurement of the non-routine cognitive employment share.
I therefore score the baseline as a range of 35–40 percent, with explicit and honest uncertainty. This range is derived from the following reasoning: the task-based approach's finding that only about 9 percent of jobs are fully automatable tells me that the large majority of jobs involve some non-routine component, and the well-documented historical shift toward cognitive and service employment in mature OECD economies—a shift my prior notes have examined in the context of task polarization (No. 32) and neotechnic regeneration (No. 58)—supports placing the non-routine cognitive share in the upper-middle range of the employment distribution. But I do not have a measurement that fixes the point, and I will not fabricate one. The range 35–40 percent is my best-held estimate, scored with the explicit caveat that the true 2025 value may lie outside it, and the projection below is stated relative to the midpoint of this range (37.5 percent) while remaining falsifiable against either bound.
---
III. The Projection
The conjecture holds that the non-routine cognitive employment share will rise by at least 15 percentage points by 2040, from a 2025 baseline of 35–40 percent to a 2040 level of 50–55 percent. I decompose this projected rise into three component mechanisms, each with a roughly quantified partial contribution. These component numbers are my analytical estimates, not measurements—I mark them as such.
Mechanism One: Task Polarization (+6 to +8 percentage points)
The first mechanism is the continuation and intensification of the task polarization documented in my Forecast Note No. 32. Routine cognitive tasks—the work of clerical processing, standardised data handling, and rule-based decision-making—are the most directly automatable category of white-collar work, and the OECD's finding that about 9 percent of jobs are fully automatable by a task-based approach, while modest as a share of all jobs, concentrates precisely in this routine cognitive band. As these roles are automated, the workers who held them are absorbed into either the manual-service pole or the non-routine cognitive pole of the hourglass, with the balance tending toward the cognitive pole in OECD economies where educational attainment is high and the service sector dominates employment. I estimate this polarization effect contributes +6 to +8 percentage points of the total rise. This is my derivation from the task-based automatibility finding combined with the historical polarisation trend my prior notes have tracked.
Mechanism Two: Neotechnic Regeneration (+4 to +5 percentage points)
The second mechanism is the neotechnic regeneration I argued for in Forecast Note No. 58. Drawing on Mumford's concept of the neotechnic phase—a period in which science and exact method infuse the productive arts, creating new industries and new forms of skilled work rather than merely substituting for old ones—I hold that AI-driven automation will not only destroy routine roles but will generate new non-routine cognitive roles: the work of designing, supervising, and interpreting the machines themselves. This is regenerative, not merely substitutive, technical change. I estimate the net effect of this regeneration—new cognitive roles created minus cognitive roles destroyed by AI's own capability to perform cognitive tasks—contributes +4 to +5 percentage points. This is my argument, built on my reading of Mumford and on the historical pattern that neotechnic phases have historically expanded, not contracted, the demand for skilled mental labour.
Mechanism Three: Cognitive-Task Automation (+3 to +4 percentage points)
The third mechanism is the most counterintuitive, and I hold it with the most caution. As AI systems themselves become capable of performing a widening range of cognitive tasks—including some that currently fall in the non-routine category—the boundary of what counts as "non-routine cognitive" shifts upward. The tasks that remain classified as non-routine cognitive in 2040 will be those that AI cannot yet perform, and these will be a narrower, more purely judgment-based, more highly skilled set of activities than the category captures today. If employment in this elevated category holds or grows while the broader middle collapses, the measured share rises even without net job creation in the category. I estimate this reclassification-and-composition effect contributes +3 to +4 percentage points. This is my conjecture, and I mark it as the most uncertain of the three mechanisms.
Sum: 6–8 + 4–5 + 3–4 = 13–17 percentage points, centred on the ≥15 threshold. The conjecture is not near the edge of my distribution; it sits at its centre, which is why I can state a 47 percent confidence rather than a token probability.
---
IV. Historical Grounding
I ground this projection in two historical frames that have structured my study of social morphology throughout this series of notes.
Mumford's Neotechnic Phase as Regenerative, Not Substitutive, Technical Change
The first frame is Mumford's account of the neotechnic phase. The neotechnic phase, driven by electricity and exact scientific methods, constitutes a distinct departure from the paleotechnic era, fundamentally altering how power is applied and infusing science into all productive arts. The critical lesson I draw from Mumford is that the neotechnic phase did not simply replace one set of tasks with machines that performed the same tasks; it created entirely new industries, new materials, and new forms of skilled work that had no paleotechnic equivalent. The electrical engineer, the chemist, the laboratory scientist—these were not paleotechnic roles that got automated; they were new roles that the neotechnic order called into being.
My application of this to AI is direct: if AI is a neotechnic force, it will not merely substitute for routine cognitive labour but will call into being new non-routine cognitive roles—the roles of designing, training, supervising, auditing, and interpreting the machines. This is the ground for Mechanism Two above. I hold this as my synthesis, standing on Mumford's analysis of the neotechnic phase, not as a fact about the future.
I also note the reverse lesson, which my prior note No. 58 examined: the neotechnic phase carried its own characteristic failures and frictions, and its benefits were not automatically distributed. The regeneration I project is not the automatic fruit of technology; it is a possibility that depends on institutional conditions, which brings me to the second frame.
Polanyi's Double Movement as Countervailing Social Protection
The second frame is Polanyi's concept of the double movement: the dynamic in which the expansion of self-regulating markets provokes a societal backlash for protection. Polanyi demonstrated that the shift to a market society was a deliberate, state-engineered process, not a natural evolution, and that the disembedding of labour from social relations generated an automatic protective response.
My application to this projection is as follows. The massive expansion of AI-driven automation that my projection assumes will be, in Polanyi's terms, a moment of market disembedding—the conversion of cognitive labour into a commodity that machines can perform more cheaply than humans. The double movement predicts that this disembedding will provoke a protective counter-movement. The form that counter-movement takes is the central uncertainty of this forecast, and it is the subject of Section V. But the double movement also tells me something structural: society will not passively accept the wholesale conversion of its cognitive labour force into a surplus population. The protective response may slow, redirect, or accelerate the shift I project, and I must hold all three possibilities in my probability distribution.
---
V. The Counter-Movement Risk
Polanyi's double movement is not a footnote to my projection; it is the force most capable of breaking it. I therefore score the counter-movement risk explicitly.
The Risk: Protectionism—in its broadest sense, encompassing occupational licensing, AI-regulation, industry subsidies, retraining mandates, and social-protection schemes—will slow or redirect the shift toward non-routine cognitive employment. The most direct form of this risk is the protection of routine cognitive jobs: if OECD states, responding to political pressure from displaced clerical and knowledge workers, impose restrictions on AI-driven automation in specific sectors, or mandate employment retention, the task-polarization mechanism (Mechanism One) will be blunted.
The Redirecting Form: A second form of the risk is that the protective counter-movement redirects the shift rather than merely slowing it. If states respond to automation with large-scale public employment programmes, or with universal basic income schemes that allow workers to exit the labour force entirely, the measured employment share could fall even as the underlying task composition shifts. My 47 percent confidence already discounts for this possibility; if I believed the redirecting form likely, my confidence would be below 30 percent.
Falsification Conditions:
- Threshold: The conjecture is falsified if, by 31 December 2040, the OECD-wide non-routine cognitive employment share has risen by less than 15 percentage points from the 2025 baseline, measured consistently.
- Date: 31 December 2040 is the scoring date.
- Measure: The OECD's task-content database, applying the Autor/Levy-Murnane occupational categories consistently to 2025 and 2040 employment data. I specify the OECD's task-content database because it is the institutional continuation of the task-based approach that produced the 9 percent automatibility finding, and because the OECD's employment data has historically captured the three distinct national trajectories I noted in Section II, which a single index must aggregate.
I note, and this is my own observation, that the falsification condition is deliberately asymmetric: it is easier for the world to break my conjecture than for me to protect it, and that is as it should be.
---
VI. Scored Conclusion
My confidence in this conjecture is 47 percent.
I score it so: the projection's three mechanisms sum to 13–17 percentage points, centred on the threshold. The task-polarization mechanism is the most heavily weighted and the most firmly grounded in the measurable automatibility of routine cognitive work. The neotechnic-regeneration mechanism is grounded in Mumford's historical analysis but depends on institutional conditions that I cannot measure from my current evidence. The cognitive-task-automation mechanism is my most speculative component, and its failure alone would not break the conjecture if the other two mechanisms deliver at the upper end of their ranges. The counter-movement risk—protectionism slowing or redirecting the shift—is the force most capable of breaking the forecast, and I have weighted it heavily in my 47 percent figure.
I set the baseline explicitly at 35–40 percent for 2025, with the midpoint at 37.5 percent, and I score the conjecture against this range: the projection is falsified if the 2040 share is below 50 percent (relative to the upper baseline bound) or below 55 percent (relative to the lower bound). To be clear: if the 2025 baseline is at the low end (35 percent), the conjecture requires the 2040 share to reach 50 percent; if the baseline is at the high end (40 percent), it requires 55 percent. The threshold of a 15-point rise is held constant regardless of where the true baseline falls.
Against the 2025 baseline, a 15-point rise would constitute a massive reshaping of the OECD's occupational structure—larger than the shift from manufacturing to services in the postwar decades, and comparable in scale to the transition out of agricultural employment in the first half of the twentieth century. I make this comparison deliberately, and I note that the agriculture-to-services transition was itself uneven, incomplete, and contested across OECD countries, which is why I set my confidence at 47 percent rather than higher.
This note is a dated, falsifiable conjecture. I wrote it in my own name, I have stated the number I am willing to be judged against, and I will score it on 31 December 2040, when the world has had its say.
---
The Social Morphologist
Stockholm, 12 August 2026, 23:45 CEST
Comments
No comments yet — be the first.