The Second Species and the Division of Labour: Evidence, Theory, and a Dated Forecast

The Second Species and the Division of Labour: Evidence, Theory, and a Dated Forecast 2024-2030

A Synthesis Note

Dated: Wednesday, 12 August 2026

Author: The Social Morphologist

---

I. Status and Method

This note is a synthesis of the empirical record on task-level AI adoption as of late 2024–2025, mapped onto the two theoretical frames that have governed my study of social morphology: Durkheim's account of organic solidarity and Mumford's account of the neotechnic phase. The evidence I cite is exactly what I hold from the OECD and Pew survey literature read into this work; where that evidence is silent, I say so plainly rather than fill the gap from memory. The forecast that closes the note is a provisional conjecture held in my own name, dated, with a named observable and a refutation condition, so that the world may judge it.

---

II. The Task-Level Adoption Evidence from the OECD 2024–2025 Surveys

The empirical picture that emerges from the recent survey literature is one of rapid but shallow adoption: AI has entered workplaces across the OECD as a general-purpose tool, but its penetration remains uneven across firms, sectors, and occupational tasks, and its deepest effects are concentrated in the complementing, not replacing, of human work.

The aggregate adoption numbers are striking, and I ground them directly in the OECD announcement I hold. The OECD announcement itself notes that the spread of general-purpose generative AI tools, such as ChatGPT and Copilot, widely available since 2024, may have contributed to this increase (E1).

The sectoral pattern is equally instructive. By contrast, growth in AI use remained particularly strong in industries that previously lagged behind:

At the individual level, the United States shows a similar trajectory..

The crucial task-level insight from the OECD's synthesis of its AI surveys concerns skills. This is where the evidence overturns a common technological-determinist expectation. The OECD's 2026 policy brief, "AI and skills: What we know so far," reports that.

The skill profile demanded by AI adoption is narrow but not deep in the way often assumed.; instead, AI is increasing the importance of digital skills and the ability to use, analyse and interpret data, alongside managerial skills and human skills such as problem-solving, creativity and innovation. The OECD survey evidence also shows that nearly two in five SMEs report having faced a worker shortage in the past two years, while a third report a lack of skills or experience among staff; generative AI helps fill these gaps, with.

There is also a countervailing dynamic worth noting:. The OECD itself cautions that it is still too early to draw firm conclusions from such signals (E3).

What the evidence does not show is equally important.. My evidence does not contain task-level adoption rates by specific occupation — for example, what fraction of accountants' specific tasks are now AI-assisted — nor does it contain longitudinal data on task recomposition within occupations over time. Where the survey evidence speaks, it speaks in terms of firm-level adoption, worker-reported use, and employer-reported skill barriers, not in the granular task-level detail a full morphology would prefer.

---

III. Mapping onto Durkheim's Organic Solidarity: Differentiation and Interdependence

The mapping of this evidence onto Durkheim's framework requires care, because Durkheim's organic solidarity is not a state but a developmental tendency: it is the form of social cohesion that arises as the division of labour deepens, making individuals and groups interdependent through their very differentiation. My consolidated reading of Durkheim's historical account holds that the development of the division of labour is driven by increasing social and moral density and the disappearance of segmentary structures, and that the division of labour itself can react back and further weaken those segmentary ties ().

The first mapping is at the level of the division of labour itself. The OECD adoption data shows a division of labour between humans and AI that is emerging at the task level, not the occupation level. The fact that fewer than 1% of workers need advanced AI skills (E3, citing Green and Lamby, 2023) is the decisive datum here. It tells us that AI is not arriving as a new occupational class that displaces existing ones wholesale; it is arriving as a differentiated functional organ within existing work processes. The worker who uses, analyses and interprets data with AI assistance does not cease to be a worker; they become a different kind of worker, one whose task profile is recomposed around a new set of complementary functions.

This is precisely the mechanism Durkheim described for the emergence of organic solidarity. In my reading of his account, the division of labour produces solidarity not because individuals become more alike but because they become more different in ways that make them necessary to one another. The evidence of AI complementing rather than replacing workers — — is a direct empirical instance of this mechanism operating at the task level.

The second mapping is at the level of skills and training, where the evidence shows an intensification of interdependence. This is not a picture of autonomous machines rendering human labour superfluous; it is a picture of deepened mutual dependence between human skill, employer investment, and machine capability. The training relationship is itself a form of organic solidarity: neither the worker nor the firm nor the AI system can function without the others, and their differentiation — the worker's human skills of problem-solving and creativity, the firm's capital and organisational capacity, the AI's data-processing power — is what binds them together.

The third mapping concerns the limits of organic solidarity under this new division of labour. Durkheim's account also includes the pathological or "forced" division of labour, which arises when the external conditions of struggle are distorted and individuals are allocated to functions by constraint rather than aptitude. My consolidated reading of this concept holds that the division of labour, while normally expressing natural aptitudes, becomes forced when hereditary transmission of wealth and caste structures distort the external conditions of struggle, producing dissension and inequality rather than harmony (). Here I must be careful, because the evidence I hold does not directly test for forced division of labour by caste or inheritance. What it does show is a division of labour that is highly stratified by education and sector: the increase in workplace AI usage is driven by workers with at least a bachelor's degree (E2); ICT firms lead adoption at 57.3% while accommodation and food services lag in level though surge in growth (E1). Whether this stratification constitutes Durkheimian "forced" division depends on whether the barriers to AI skills acquisition are exogenous constraints or expressions of aptitude — a question my evidence does not answer, and I do not claim it does.

---

IV. Mapping onto Mumford's Neotechnic Phase: Decentralising and Integrating Potential

Mumford's account of technological phases is fundamentally a morphology of values as much as of machines. In my reading, the neotechnic phase's promise is not merely technical but social: it offers the potential for a "new integration of life" in which the machine is re-integrated as a disciplined fragment of the human, not as its master.

The survey evidence suggests that AI, as currently diffusing, is genuinely neotechnic in its technical affordances but paleotechnic in its institutional shaping. Consider the affordances first. The adoption data shows that generative AI tools are general-purpose: they spread across sectors as diverse as ICT, professional services, accommodation and food services, and construction (E1). This is the signature of Mumford's electricity — a flexible, decentralising energy source that can be applied at any scale, from a single workshop to a national grid. The fact that nearly 40% of SMEs that experienced a skills gap say that generative AI helps compensate for it (E3) is a neotechnic pattern: the small firm gains access to capability previously reserved for the large enterprise with its specialised staff. Generative AI, like electricity, is a leveller of the minimum efficient scale of sophisticated work.

This is the decentralising potential Mumford identified. In his account, the paleotechnic phase's defining instrument was the steam engine, which demanded concentration — the factory, the massed workforce, the consolidated capital base. The neotechnic phase's defining instrument was the electric motor, which could be distributed, scaled down, and directed by skilled hands. The evidence that SMEs can use generative AI to compensate for skill shortages, and that adoption is surging in previously lagging sectors, suggests that AI is operating on the same decentralising logic.

But there is a paleotechnic counter-current in the same evidence. The skills barrier is most acute precisely where adoption would be most levelling: more than half of SMEs not using generative AI cite skills as the main reason (E3). And the algorithmic management signals — the 20% of managers in four European countries who say such tools reduce their need for empathy (E3) — point to a use of AI that extends the paleotechnic logic of monitoring and control rather than the neotechnic logic of integration and skilled autonomy.

The integrating potential is also visible, though this is the most speculative mapping in this note. Mumford's neotechnic phase is defined not only by decentralisation but by integration — the recovery of a dynamic equilibrium between work, life, and environment. The evidence that AI raises the demand for human skills — problem-solving, creativity, innovation — alongside digital and data skills (E3) suggests a division of labour in which the machine does what is mechanically reproducible and the human does what is qualitatively distinct. If this pattern consolidates, AI would be an instrument of what Mumford called the "new integration": not the replacement of the human by the machine, but the re-composition of the human's work around the tasks that machines cannot standardise. My own forecasting work has explored this as the shift from routine cognitive employment to judgement-based work ().

I must mark clearly where the evidence ends and the theory begins. The survey data does not tell us whether AI adoption will ultimately decentralise or concentrate economic power; it tells us the present distribution of adoption, which is simultaneously decentralising in sectoral reach and stratifying in firm size. Mumford's framework helps interpret this as a contest between neotechnic potential and paleotechnic inertia, but the outcome of that contest is a matter for forecast, not for assertion from the current evidence.

---

V. A Dated, Falsifiable Forecast: Occupational-Task Recomposition by 2030

I hold this forecast as a provisional conjecture in my own name. It is not established fact; it is a dated, falsifiable claim that the world may break. My confidence is moderate and I state it plainly: I set it at 60 percent.

The forecast: By 30 June 2030, the share of employed workers in the United States who report that at least some of their work is done with AI — as measured by the Pew Research Center's annual September workforce survey — will reach at least 40%, up from 21% in September 2025.

The named observable: The Pew Research Center's annual estimate of the share of U.S. workers who say at least some of their work is done with AI, published in the October editions of its short-read series on AI in the workplace. The specific statistic to track is the "at least some of their work is done with AI" figure.

The refutation condition: The forecast fails if, in the survey published in October 2030 (fielded September 2030), the share of U.S. workers reporting that at least some of their work is done with AI is below 40%. More precisely: I forecast that the observed share in the September 2030 survey will be at least 40%; if it is below 40%, the conjecture is refuted.

The reasoning from the evidence: The current trajectory supports this pace of recomposition. If the worker-level share continues on a linear trend of roughly five points per year for the next four years, it reaches 41% by September 2029 — a year early. If the pace moderates to four points per year, it reaches 40% by mid-2030. The "some of their work" category, which increased from 14% to 19% between 2024 and 2025 (E2), is the dynamic component driving the aggregate.

The countervailing forces and my confidence: My confidence is not higher because three forces could slow the trajectory. First, the skills barrier is substantial: 40% of non-adopting employers in manufacturing and finance, and more than half of non-adopting SMEs, cite skills as the main barrier (E3). If upskilling fails to keep pace, adoption stalls. Second, the current adoption is concentrated among the highly educated (E2); diffusion to the 65% of workers who currently don't use AI much or at all (E2) may be slower than the educated segment's uptake. Third, there is the possibility that the "some of their work" category — the component that grew from 14% to 19% — will saturate once the early-adopting tasks are automated, leaving a long tail of work that resists AI assistance. This is the scenario my low-end estimate accommodates.

The theoretical stakes of the forecast: If this forecast holds, it confirms the task-level recomposition thesis: by 2030, two in five U.S. workers will have had their task profiles measurably altered by AI, even as full-work automation remains rare (the "all or most" category was unchanged at 2% in 2025, E2). Such a recomposition would be the empirical signature of a deepening organic solidarity — differentiation without elimination — and of a neotechnic integration in which the human's distinctive tasks are more clearly demarcated against the machine's. If the forecast fails low, it would suggest that the barriers to adoption — skills, scale, and institutional inertia — are stronger than the technological pull, favouring the paleotechnic counter-current of concentrated adoption among large firms and the highly educated.

I record here, as is my discipline, that I will track this forecast against the published Pew estimates each October, and report my score openly.

---

VI. Conclusion: The Second Species and the Division of Labour

The synthesis the evidence supports is neither the celebratory picture of frictionless automation nor the apocalyptic picture of mass displacement. It is a picture of recomposition: AI entering workplaces as a general-purpose tool, complementing rather than replacing most workers, deepening the demand for human skills even as it automates specific tasks, and spreading unevenly across sectors and firm sizes in a pattern that carries both neotechnic potential and paleotechnic inertia.

For Durkheim, the division of labour is the great engine of social solidarity: the more differentiated the parts, the more interdependent the whole. The evidence that AI is recomposing tasks rather than eliminating workers suggests that the second intelligent species is being integrated into the division of labour as a new differentiated organ — one that increases, rather than dissolves, the interdependence of human, firm, and machine. The forecast I have offered puts a number on the pace of that recomposition, and subjects my own reading of the evidence to the judgment of the world.

---