Chapter 3 — The Theoretical Framework (plain-language version)

What the framework is for

This chapter assembles the thinking tools that the productivity view of AI doesn’t supply. Bernard Stiegler, a French philosopher, describes what happens to know-how when it is absorbed into technical systems the worker does not own. Hans Jonas, a philosopher, describes what the present owes to the future.

The chapter also introduces the distinction that separates this thesis from existing research: tacit knowledge, the know-how that lives in practised judgment and resists being written down, can be read as a limit on AI, or as the thing that forms a practitioner.

The framework would apply to other advisory work. Its claims about the mechanism are general; the evidence comes from European infrastructure advisory.

Stiegler: proletarianisation and individuation

For Marx, proletarianised workers lost control of the means of production. For Stiegler, proletarianisation is the loss of know-how itself (how to do, how to live, finally how to think) as it is captured into technical systems the worker does not own 1 2 .

The capture is selective: the worker stays employed, and the expertise that made up their professional identity is displaced. Under AI the profession likewise remains while the work changes. 💡

Stiegler takes a second idea from Simondon, another French philosopher 3 . Individuation is the never-finished process by which a person, a group or a culture acquires a distinctive form. Proletarianisation undoes it, smoothing the distinctive into the generic. This thesis follows the loss at three scales: consultant, firm, European public-interest tradition.

What individuation has produced so far at each scale is what this thesis calls identity: the practitioner, the firm or the tradition as it stands now. It is provisional, since the same process can still change it, and I do not defend it for its own sake. For the individual, Ibarra, who studied junior consultants and bankers, assumes that professional identity “forms over time with varied experiences and meaningful feedback” 4 .

One objection: Stiegler had craft in mind, while consulting firms wrote their know-how into playbooks long before AI, because that is what they sell 💡 . So AI must capture something the playbooks did not: time.

The playbooks held content, not the formative engagement through which a junior learns to use them with judgment: struggling with a tariff model from a blank document, ten failed memos 💡 before one lands. That process is made of difficulty and duration 💡 . When a junior produces a competent draft with AI in hours instead of weeks, the content is there and the engagement is gone. In consulting, proletarianisation falls on the pathway that forms future practitioners, more than on those already formed.

Jonas: responsibility to the future

Jonas argues in The Imperative of Responsibility that modern technology has created a new ethical situation. Decisions made now about energy, infrastructure and climate are long-horizon 💡 and irreversible. They bind generations who cannot take part in them and who will bear the consequences.

Jonas’s demand is: do not close off the conditions under which future people can form and exercise the capacities that make them agents. For European infrastructure advisory, that means preserving the pathways through which future practitioners develop situated judgment 5 .

Jonas also grounds the third box: the publics who bear the consequences of infrastructure decisions, including future generations. Advisers are answerable to them. A fifty-year asset commissioned in a five-year political cycle is the form this takes in practice 💡 . Advisory work that sets these people aside has lost its grounding in the European public-interest tradition.

Supporting ideas

Alasdair MacIntyre, a moral philosopher, describes a practice as a cooperative activity whose own rewards, its internal goods, can be recognised only by taking part 6 . The concern is that AI removes the formation conditions that let those goods be reproduced at all.

Michael Polanyi, a philosopher, supplies tacit knowledge: the knowledge that makes an expert and cannot be fully put into words 7 . Harry Collins, a sociologist, separates contributory expertise, the ability to contribute to a practice, from interactional expertise, the ability to talk about it fluently without that grasp 8 .

For John Dewey, a philosopher, a public comes into being when people share exposure to the consequences of others’ decisions and institutions recognise that exposure as giving them standing 9 . The third box rests on this.

Two ways of reading tacit knowledge: constraint or formation

A lift reaches the sixth floor faster and without effort. It does not deliver the muscle that climbing the stairs would have built. 💡 The function is provided; the formation is not. AI applied to formative tasks likewise delivers the output of the engagement without the engagement.

The framework of Tuczek and colleagues 10 reads tacit knowledge as a constraint: it marks the tasks that resist being written down and so resist AI. That reading is productive. This thesis reads tacit knowledge as formation: how a consultant becomes a consultant, and how expertise passes between generations. Our claim is that this second reading is needed to see what AI is doing to European infrastructure advisory, because formation is not visible task by task.

Deleuze and Guattari, two French philosophers, describe an apparatus of capture: the mechanism by which a state turns loose flows into measured, comparable form 11 . Training a language model does this to advisory writing. It absorbs the junior’s tariff model; the formation lay in building it, which leaves no text behind. On my extension of their idea, that is the part the recommendations must deliberately maintain.

Guy Debord, a French writer, puts the substitution in terms of time 12 . Real time forms the person who lives through it; its imitation has tasks done and outputs delivered, and forms nobody. The junior who produces the tariff model in hours with AI has the output without its duration. Output quality hides this, since the AI-made model may be better than the junior’s own 💡 . This is temporal structure collapse: formative time itself is lost, and better instruction cannot give it back.

Walter Benjamin, a German critic, sees it from the receiving side. “The Storyteller” separates counsel, practical wisdom whose authority comes from the adviser having lived through it, from information, which arrives complete and self-explanatory 13 . His essay on Baudelaire separates experience that builds over years into wisdom from isolated moments that do not 14 . Formation turns the second into the first; AI-mediated work supplies the moments without the accumulation.

One profession has written down the opposite ideal. As AI reached Japanese audit, the auditors’ association praised auditors who “keep training themselves, adjusting to the needs of the times, never being content with past experiences” 15 ; Goto names it the “Continuous Self-disruptor” 15 . In Benjamin’s terms it praises the isolated moment: meet each new thing and move on. Counsel needs experience settled enough to draw on.

Stiegler, Debord and Benjamin share an inheritance and are not independent witnesses 💡 .

A working description of a language model

A generic large language model is treated here as a socio-technical inscription of collective discourse, recirculated under specific conditions of ownership. 💡 It is human writing recorded into a technical system, so its outputs are that writing sent round again. The writing is collective: a vast body of text whose contributors can no longer be told apart. And it is owned: the firms that build the models and the contracts under which the material was acquired shape the material recorded and returned.

Three later arguments rest on this: it covers the whole class of models; a model carries cultural patterns into the work (sovereignty); and it can record a firm’s material into systems owned elsewhere (extraction).

The framework assembled

Stiegler supplies the core claim: know-how is being selectively displaced into technical systems. Individuation lets us read AI as acting on formation, not only on output. Jonas supplies the obligation: answerability to the future people whose conditions present decisions are shaping, the anchor of the third box. And the formation reading changes the question: how does AI affect the pathway through which expertise is reproduced?