The Loss of Formation: Artificial Intelligence and the Erosion of Expertise
Carl Jean
Abstract
This essay argues that artificial intelligence threatens expertise not primarily by replacing skilled labor, but by displacing the formative processes through which skill, judgment, and authority are developed. It introduces the concept of formative displacement, defined as the removal or compression of the developmental sequences—repetition, error, correction, and gradual internalization—through which practitioners acquire capacities that cannot be fully articulated or automated. Drawing on the work of Hubert Dreyfus, Harry Braverman, and Andrew Abbott, the essay situates artificial intelligence not at the level of expert performance, but at the level of formation itself. While contemporary systems are celebrated for increasing efficiency by automating routine, high-volume tasks, those tasks historically constituted the substrate on which expertise was built. When that substrate is removed, institutions may retain the appearance of expertise while gradually degrading its substance. This produces a temporal externality: the costs of formative displacement are deferred, becoming visible only as current expert cohorts are replaced by practitioners who have not undergone comparable developmental processes. Artificial intelligence is therefore not merely a tool that reshapes work, but a system that reorganizes the conditions under which human capability emerges. Its legitimacy depends not only on what it produces, but on whether it preserves the processes through which future expertise can still be formed.
I. Expertise as Formation, Not Performance
Expertise is often defined in terms of performance—the ability to produce correct or effective outcomes—but this definition obscures the processes through which such outcomes become possible. To become an expert is not simply to accumulate knowledge, but to undergo a transformation in perception, judgment, and responsiveness that emerges through sustained engagement with difficulty. This transformation is gradual, structured, and resistant to compression. It depends on repetition, failure, feedback, and the progressive internalization of distinctions that cannot be fully articulated in advance.
The account developed by Hubert Dreyfus and Stuart Dreyfus provides a precise framework for understanding this process. In their five-stage model of skill acquisition—novice, advanced beginner, competent, proficient, expert—progression occurs not through the accumulation of rules, but through a shift in how situations are perceived. Novices rely on context-free rules; experts respond intuitively to patterns that have become meaningful through experience. The crucial point is that this transition cannot be skipped. The later stages depend on capacities that are formed only through passage through the earlier ones.
This account is further grounded in the work of Michael Polanyi, whose concept of tacit knowledge establishes that expertise involves forms of understanding that cannot be fully articulated. As Polanyi argues, we know more than we can tell, and it is precisely this inarticulable dimension that is developed through sustained engagement with practice. The formative stages of expertise are therefore not merely preparatory; they are the only conditions under which tacit capacities can emerge. If these stages are bypassed, what is lost is not simply experience, but the development of knowledge that cannot be recovered through instruction or observation alone.
This implies a structural constraint on expertise: it requires time spent in positions of limited competence. Early-stage work—routine, repetitive, and often low-stakes—is not peripheral to expertise but constitutive of it. It is through this work that practitioners learn to notice what matters, to discriminate signal from noise, and to develop the judgment that defines expert performance.
Expertise is not what one produces, but what one becomes capable of producing through formation.
II. Formative Displacement and the Reorganization of Skill
Artificial intelligence intervenes precisely at the level where formation occurs. Rather than replacing experts outright, it removes or compresses the developmental stages through which expertise is produced. This process can be described as formative displacement: the reallocation of formative labor from human practitioners to automated systems.
The analysis of Harry Braverman provides a historical analogue, but the distinction is critical. Braverman’s account of deskilling describes the reduction of worker expertise through the fragmentation of tasks and the transfer of knowledge into machines. Workers retain employment but perform simplified functions requiring less skill. Formative displacement, by contrast, operates prior to this stage. It does not reduce existing expertise; it prevents expertise from forming in the first place. Where deskilling degrades competence, formative displacement interrupts its development. Where Braverman’s model is immediate and visible, formative displacement is delayed and structural, emerging only across time as new cohorts fail to acquire the capacities their roles require.
Artificial intelligence extends this logic into cognitive domains previously resistant to such intervention. Tasks that once required judgment—first-pass analysis, preliminary diagnosis, early-stage research—are now performed by systems trained on large datasets. Practitioners remain involved, but their engagement shifts from participation in formative processes to interaction with completed outputs.
This dynamic is further illuminated by Donald Schön’s account of the reflective practitioner, in which professional judgment develops through iterative engagement with uncertain situations. Schön emphasizes that expertise is not the application of fixed knowledge, but a process of knowing-in-action that emerges through doing and reflecting simultaneously. Formative displacement interrupts this process by reducing the occasions for such engagement, weakening the conditions under which reflective judgment can develop.
When systems perform the work that forms the worker, the worker’s development is altered.
III. The Counterargument: Efficiency, Augmentation, and the Redistribution of Expertise
The strongest objection to this account is that it misinterprets the role of efficiency in technological change. Artificial intelligence, on this view, does not erode expertise but redistributes it. By automating routine tasks, it allows practitioners to focus on higher-order functions—interpretation, strategy, and decision-making—that more accurately reflect the essence of expert work.
This argument is compelling because it is historically grounded. Technologies that reduce cognitive burden have often expanded human capability. The calculator did not eliminate mathematics; it enabled more complex forms of abstraction. Digital tools did not eliminate design; they accelerated it. Artificial intelligence may represent a continuation of this pattern, freeing practitioners from low-level tasks in order to engage more fully with complex ones.
Efficiency does not eliminate expertise; it changes where expertise is required.
Yet this argument depends on a crucial assumption: that the tasks being removed are not essential to the development of expertise itself. The question is not whether efficiency is beneficial, but whether the new structure preserves the conditions under which expertise is formed.
IV. The Limits of Efficiency: Formation, Practice, and Dependency
The augmentation account holds only if the processes eliminated by automation are not necessary for the development of expertise. Research on skill acquisition, particularly the work of K. Anders Ericsson, suggests that they are. Expertise depends on deliberate practice—sustained engagement with tasks that require effort, feedback, and correction. Observing correct outcomes is insufficient; expertise emerges through the process of attempting, failing, and refining performance.
Artificial intelligence alters this structure by enabling practitioners to bypass these processes. When systems generate solutions directly, opportunities for deliberate practice are reduced. Over time, this can produce a form of dependency, in which practitioners rely on systems for tasks they have not learned to perform independently.
When the path to expertise is removed, expertise itself becomes unstable.
V. Case Study: Radiology and the Transformation of Diagnostic Expertise
Radiology provides a clear example of how formative displacement operates in practice. Diagnostic expertise in radiology is developed through repeated exposure to medical images, requiring practitioners to interpret subtle variations, recognize patterns, and refine judgment through feedback.
First, process displacement. AI systems increasingly perform first-pass image analysis, identifying patterns associated with specific conditions. This reduces the volume of cases that trainees must interpret independently, limiting their exposure to the raw material of formation.
Second, feedback reduction. When AI systems provide correct or near-correct outputs, the opportunity for error—and therefore for correction—is diminished. Trainees encounter fewer situations in which their interpretations are challenged, reducing the feedback loop that drives learning.
Third, judgment dependency. As reliance on AI-generated outputs increases, practitioners may defer to system recommendations rather than developing independent judgment. This creates a dependency in which correct performance is achieved without the corresponding development of expertise.
As Atul Gawande observes in his work on clinical performance, expertise in medical domains depends on institutional conditions that sustain repeated, supervised engagement with difficulty—conditions that are increasingly altered by the integration of automated systems.
Accurate diagnoses do not guarantee the formation of diagnostic judgment.
Similar dynamics are emerging in law, where AI systems perform document review and preliminary analysis, reducing the experiential foundation through which junior associates develop legal reasoning.
VI. Temporal Externality and the Deferred Cost of Efficiency
One of the most significant features of formative displacement is its temporal structure. The effects are not immediately visible. Current experts, trained under earlier conditions, continue to perform at high levels, masking the underlying shift in formation.
This creates a temporal externality, in which the costs of efficiency are deferred. Institutions benefit from increased productivity in the short term while undermining the conditions necessary for long-term expertise. As current expert cohorts are replaced by practitioners who have not undergone comparable developmental processes, the degradation of expertise becomes visible.
This temporal structure also reflects the broader reorganization of labor described by Kate Crawford, who argues that artificial intelligence systems operate as infrastructures built on the extraction and aggregation of human activity. From this perspective, the displacement of formative processes is not merely a local effect within professions, but part of a larger structural shift in which the conditions of knowledge production are reorganized at scale. The result is a system that preserves outputs while redistributing—and often obscuring—the labor through which those outputs were historically formed.
The loss of expertise is not immediate; it is inherited.
VII. Professional Authority, Jurisdiction, and Responsibility
The erosion of formative processes has consequences not only for skill but for authority and responsibility. Andrew Abbott’s account of professional jurisdiction emphasizes that professions maintain authority through their control over specialized knowledge and its application. Artificial intelligence disrupts this structure by performing tasks that were previously central to professional expertise.
As authority becomes distributed across human and machine systems, responsibility remains attached to human agents. Practitioners are expected to justify decisions and bear the consequences of outcomes, even when those outcomes depend on processes they did not perform. This creates a structural tension: authority is redistributed, but responsibility is not.
In this condition, practitioners may be held accountable for judgments they are less equipped to make independently. The link between formation and responsibility is weakened. Responsibility presupposes the capacity for judgment; when that capacity is underdeveloped, accountability becomes increasingly formal rather than substantive.
The consequence is not simply a redistribution of expertise, but a transformation in the structure of accountability itself. Professionals may be required to justify decisions that depend on processes they did not perform and capacities they were not fully formed to develop. In such cases, responsibility becomes detached from formation, producing a condition in which individuals are held accountable for judgments they cannot independently reproduce. This is not a failure of individual competence, but a structural consequence of formative displacement.
When judgment is outsourced, responsibility becomes harder to locate.
VIII. The Threshold of Expertise
The dynamics of formative displacement converge at a threshold. Below this threshold, artificial intelligence supports expertise by enhancing feedback and enabling practice. Beyond it, the conditions necessary for formation are compromised.
This threshold is defined by the interaction of three conditions. Process displacement removes the tasks through which expertise is developed. Feedback reduction limits opportunities for correction and learning. Judgment dependency replaces internalized capacities with reliance on external systems. Individually, these conditions may be manageable. Together, they alter the structure of expertise itself.
When all three conditions converge, practitioners can produce correct outcomes without having undergone the processes that would allow them to generate those outcomes independently. At this point, expertise is no longer being formed, even as its outputs persist.
Below the threshold, tools train experts. Beyond it, they replace the processes that make experts possible.
Conclusion
Artificial intelligence does not eliminate expertise. It transforms the conditions under which expertise is formed, maintained, and exercised. The central question is not whether AI can perform expert tasks, but whether human practitioners will continue to develop the capacities that define expertise.
What is at stake is not performance, but the formation of the capacity to perform.
Final Doctrine
Artificial intelligence is epistemically legitimate only when it preserves the conditions necessary for the formation of expertise—requiring practitioners to engage in the processes through which knowledge, judgment, and adaptability are developed.
Related Reading
If this essay examines how artificial intelligence disrupts the formation of expertise, the next essay, The Work That Refuses to End: Creativity, Authorship, and the Limits of Artificial Intelligence, turns to a parallel question: what remains of creativity when the processes that once formed the creator are increasingly externalized. It argues that while artificial intelligence can generate convincing outputs, it cannot replace the ongoing, unfinished work through which authorship is developed and sustained. Together, the essays trace a shared boundary—between what can be produced and what must still be formed.
Join the Conversation
What happens when the systems we rely on begin to shape not just what we produce, but what we are capable of becoming? If you found this argument compelling—or if you disagree—I invite you to share your perspective in the comments. How is artificial intelligence affecting the formation of expertise in your field? Subscribe to follow the full series as it traces the changing conditions of creativity, responsibility, and human development in the age of intelligent systems.
Works Cited
Abbott, Andrew. The System of Professions: An Essay on the Division of Expert Labor. University of Chicago Press, 1988.
Braverman, Harry. Labor and Monopoly Capital: The Degradation of Work in the Twentieth Century. Monthly Review Press, 1974.
Crawford, Kate. Atlas of AI: Power, Politics, and the Planetary Costs of Artificial Intelligence. Yale University Press, 2021.
Dreyfus, Hubert L., and Stuart E. Dreyfus. Mind over Machine: The Power of Human Intuition and Expertise in the Era of the Computer. Free Press, 1986.
Dreyfus, Hubert L. On the Internet. 2nd ed., Routledge, 2009.
Ericsson, K. Anders, Ralf Th. Krampe, and Clemens Tesch-Römer. “The Role of Deliberate Practice in the Acquisition of Expert Performance.” Psychological Review, vol. 100, no. 3, 1993, pp. 363–406.
Gawande, Atul. Better: A Surgeon’s Notes on Performance. Metropolitan Books, 2007.
Pasquinelli, Matteo. The Eye of the Master: A Social History of Artificial Intelligence. Verso, 2023.
Polanyi, Michael. The Tacit Dimension. University of Chicago Press, 1966.
Schön, Donald A. The Reflective Practitioner: How Professionals Think in Action. Basic Books, 1983.
Comments
Post a Comment