Taste & judgement, unbundled

Critique of 'technical rationality' in professional practice. Professionals don't apply theory to practice — they reflect-in-action, framing and reframing problems.

Donald Schön’s The Reflective Practitioner: How Professionals Think in Action (1983) stands as a landmark in twentieth‑century thinking about professional practice and, more specifically, about design. Against the dominant ideal of the professional as a detached problem‑solver who applies scientific theory to clearly defined situations, Schön offers a rich, practice‑based account of how professionals actually work in the messy, uncertain conditions of real life. His central claim is that competent practice depends on a kind of intelligent improvisation he calls “reflection‑in‑action,” and that this practical, situated intelligence deserves the same status as formal, scientific knowledge.

Historical situation and the “crisis of confidence”

Schön writes in the context of a growing “crisis of confidence in professional knowledge” in the late twentieth century. From the late nineteenth century onward, the modern research university and the professions—medicine, law, engineering, architecture, planning, management—had been bound together by a shared promise: that rigorous, science‑based education would produce experts capable of diagnosing and solving society’s problems. Professional authority was justified by the assumption that practitioners applied reliable, codified knowledge to well‑structured problems.

By the 1970s and early 1980s, this promise was visibly fraying. In domains such as urban planning, environmental management, health care, and education, professionals confronted problems that were ill‑defined, politically contested, and resistant to technical solution. Urban renewal projects displaced communities rather than improving them; large‑scale planning failed to anticipate social and ecological complexity; managerial techniques imported from operations research and economics did not reliably produce effective organizations. Schön reads these failures not as incidental but as symptomatic of a deeper mismatch between the dominant model of professional knowledge and the actual character of professional work.

He names the dominant model “technical rationality”: the view that professional practice consists in the application of scientific theory and technique to instrumental problems. On this view, the proper sequence is: (1) define the problem, (2) select the relevant theory or method, (3) apply it, and (4) evaluate the outcome. Schön argues that this model fits only a narrow subset of situations—those that can be cleanly formalized—and that it obscures the most important and characteristic activities of skilled practitioners.

Central claim: reflection‑in‑action

Schön’s central concept, “reflection‑in‑action,” captures the distinctive kind of thinking that occurs in the midst of practice. Instead of separating thinking from doing, he shows how professionals think through their doing. Reflection‑in‑action is not a leisurely, after‑the‑fact analysis; it is an on‑the‑spot, experimental engagement with a situation that is unfolding in real time.

He characterizes this as a “reflective conversation with the situation.” The practitioner:

Frames the situation – They decide what kind of situation this is, what is at stake, and what counts as a relevant move. Framing is not given by the situation itself; it is an interpretive act.

Makes a move – They intervene: a sketch, a question, a treatment, a policy proposal, a managerial decision.

Listens for the situation’s “back‑talk” – The situation responds: the sketch suggests new forms, the client reacts, the system behaves in an unexpected way. These responses are not just outcomes; they are information.

Reframes and adjusts – In light of this back‑talk, the practitioner may reframe the problem, revise their understanding of what is going on, and choose a new move.

This cycle is iterative and improvisational. It draws on what Schön calls “knowing‑in‑action”—the largely tacit, embodied repertoire of skills, patterns, and examples that experienced practitioners have built up over time. Knowing‑in‑action is not typically articulated as explicit rules; it is revealed in the smooth, unselfconscious performance of competent practice. Reflection‑in‑action occurs when this usually tacit knowing becomes momentarily visible to the practitioner, allowing them to question, adjust, and extend it in response to the situation.

Crucially, Schön insists that reflection‑in‑action is both intuitive and rigorous. It is not mere guesswork or arbitrary improvisation. Its rigor lies in the disciplined way practitioners test their moves against the situation’s back‑talk, revise their frames, and build a coherent line of action under conditions of uncertainty.

Key concepts and mechanisms

Several interrelated concepts structure Schön’s account:

Technical rationality – The prevailing ideal of professional knowledge as the application of scientific theory and technique to well‑formed problems. Schön argues that this ideal marginalizes the most important aspects of practice: problem framing, dealing with uniqueness, and working under uncertainty.

Knowing‑in‑action – The tacit, embodied know‑how that guides skilled performance. It is visible in what practitioners do rather than in what they can say about what they do.

Reflection‑in‑action – The practitioner’s ability to surface, question, and reshape their knowing‑in‑action while acting. It is a kind of thinking that is inseparable from doing.

Reflection‑on‑action – The retrospective analysis of one’s practice after the fact. While important for learning, it is distinct from the in‑the‑moment improvisation that is Schön’s primary focus.

Framing and reframing – The interpretive acts by which practitioners define what the problem is, what counts as relevant, and what a solution might look like. For Schön, framing is central: different frames produce different problems and different possible moves.

The reflective conversation with the situation – The back‑and‑forth process in which the practitioner “talks” to the situation through their moves and “listens” to its responses, adjusting their understanding and action accordingly.

These concepts together articulate a model of professional practice as a form of practical wisdom rather than merely the application of theory. Professionals are not just problem‑solvers; they are problem‑framers and sense‑makers in complex, evolving contexts.

Design case studies and the architecture studio

To make these ideas concrete, Schön analyzes detailed case studies of practitioners at work: a psychotherapist in session, a town planner negotiating a development decision, a manager handling a product crisis, and, most influentially for design theory, an architecture studio tutorial.

In the architecture case, Schön observes an interaction between an experienced designer‑teacher (whom he calls Quist) and a student working on a design problem. The medium is the sketch: as Quist draws, erases, and redraws, he is not simply externalizing a pre‑formed solution. Each mark on the page both expresses an intention and reveals new possibilities and constraints.

Schön describes this as a process of “seeing‑as” and “doing‑as.” The designer sees the emerging sketch as a certain kind of configuration—perhaps as a courtyard, a circulation pattern, a massing strategy—and then acts as if that configuration invites particular moves. Each new line changes what can be seen; each new way of seeing suggests new lines to draw. The conversation is not only between teacher and student but between both of them and the drawing itself.

This analysis has been foundational for design cognition. It shows design not as the execution of a pre‑specified plan but as an exploratory dialogue with materials and representations. The sketch is not merely a communication tool; it is a thinking medium. The designer’s expertise lies in their ability to generate and shift frames, to see emergent patterns in partial representations, and to respond creatively to the “back‑talk” of the drawing.

Contrast with Herbert Simon and rational problem‑solving

Schön’s work is often read in direct contrast to Herbert Simon’s The Sciences of the Artificial (1969, rev. 1981), which offered a highly influential model of design as rational problem‑solving. For Simon, design is a form of search in a problem space: the designer formulates goals and constraints, generates alternatives, evaluates them against criteria, and selects the optimal or satisficing solution. This model aligns closely with technical rationality and with computational approaches to design.

Schön does not deny that such analytic and optimization processes occur, but he argues that Simon’s model captures only the technical component of practice. It presupposes that the problem is already well‑defined and that relevant variables and criteria are known. In real professional work, however, the most challenging and characteristic tasks involve:

Defining what the problem is in the first place.

Dealing with situations that are unique or uncertain, where no standard method straightforwardly applies.

Negotiating conflicting values, interests, and interpretations.

In these contexts, the designer’s central skill is not optimization within a given frame but the ability to frame and reframe the situation, to invent new problem definitions, and to improvise appropriate responses. Schön thus reorients design theory from a focus on rational choice within fixed parameters to a focus on reflective engagement with evolving situations.

The debate between Simon’s rationalism and Schön’s reflective practice has structured much of design theory since the 1980s. Subsequent work in design research, participatory design, and human‑centered design often aligns more closely with Schön’s emphasis on situated, iterative, and dialogical processes, even when it still borrows tools from rational problem‑solving.

Reception and influence across professions

The Reflective Practitioner quickly became a touchstone across a wide range of fields. In education, it helped reshape teacher training around the idea of the “reflective teacher,” emphasizing classroom inquiry, case discussion, and practice‑based learning. In nursing, social work, and counseling, it provided a vocabulary for articulating the complexity of practice that could not be captured by procedural manuals or evidence‑based protocols alone.

In medicine and law, Schön’s ideas contributed to curricular reforms that foregrounded clinical reasoning, case‑based learning, and the cultivation of judgment under uncertainty. In management and organizational studies, the book fed into movements emphasizing learning organizations, action learning, and the reflective manager.

For design specifically, Schön’s legacy is profound. His account of design as a reflective conversation with materials and situations has become a foundational reference in design research, studio pedagogy, and practice. It underpins:

The emphasis on studio‑based learning as a setting where students can engage in guided reflection‑in‑action.

The recognition of sketching, prototyping, and modeling as central cognitive tools, not just communication devices.

The valorization of iterative, exploratory processes—“design thinking”—over linear, specification‑driven approaches.

Moreover, Schön’s insistence that design knowledge is a form of practical wisdom—a knowing‑in‑action revealed through doing—has helped legitimize practice‑based research and design inquiry as intellectually serious forms of knowledge production, on par with more traditional scientific and theoretical work.

Connections and continuing relevance

Schön’s work connects with broader philosophical traditions that value practical, situated knowledge: Aristotle’s notion of phronesis (practical wisdom), pragmatist theories of inquiry (Dewey), and phenomenological accounts of embodied skill (Merleau‑Ponty, later elaborated by Dreyfus). Within design, it anticipates and supports later emphases on participatory design, co‑design, and human‑centered design, all of which foreground ongoing engagement with stakeholders and contexts rather than one‑off problem specification.

The enduring power of The Reflective Practitioner lies in its revaluation of what professionals actually do when they are at their best. Instead of treating the intuitive, improvisational aspects of practice as second‑rate or merely preliminary to “real” scientific work, Schön shows that these are the heart of professional artistry. By naming and analyzing reflection‑in‑action, he provides both a critique of technical rationality and a constructive alternative: a vision of professional education and practice that takes seriously the intelligence embedded in action, the centrality of framing, and the creative, conversational character of work in complex, real‑world situations.

Schön, D. A. (1983). The Reflective Practitioner: How Professionals Think in Action. New York: Basic Books.

Boisot's I-Space maps knowledge along codification, abstraction, and diffusion, and the Social Learning Cycle traces how organisations learn: scanning, problem-solving, abstraction, diffusion, absorption, impacting. Codification is selective and lossy; knowledge is dispositional, inferable only from action; and the cycle's returning arc — absorption and impacting — runs through practice, not machines. Winner of the 2000 Igor Ansoff Strategy Prize.

Knowledge Assets is the theory the knowledge-management wave needed and mostly ignored: a precise account of what codification does (economises data processing, buys diffusion) and what it costs (selection among conflict-laden alternatives, fossilised skills, discarded particularity). IT accelerates the codifying and diffusing arc of the Social Learning Cycle; absorption — learning-by-doing — and impacting — re-embedding in concrete practice — are by definition experiential, personal, idiosyncratic. In his own later words: at best, ICTs increase the quantity of data flowing between agents; agents must still sift it, and absorption has been 'downplayed in knowledge management, particularly where it is driven by information technology.'

The essay's coda rests its diagnosis of the recurring digitalization waves on Boisot: the social learning cycle by which an organisation actually comes to know lives only half in the machine, which is why each wave's return refused to appear — codification always leaves a residue, and the residue was the value. He is also the register's Q5 frame-internal-lossy pole, the nearest neighbour to Hayek's successful-formalisation position in the epistemic convergence.

Boisot, M. H. (1998). Knowledge Assets: Securing Competitive Advantage in the Information Economy. Oxford: Oxford University Press. (SLC: pp. 58–66; codification: pp. 41–47; ch. 9 'IT and its Impact'. Explicit ICT statement: Boisot in Choo & Bontis, eds., 2002.)

Kurtz and Snowden's fullest academic statement of Cynefin: known, knowable, complex and chaotic domains (with disorder at the center) demand different relationships between cause, effect and action — categorize, analyze, probe or act — and most management failure is treating one domain with another's methods.

The paper's target is three pervasive assumptions — order, rational choice and intent — and its instrument is the Cynefin framework: in ordered domains best practice and analysis work; in complex domains cause and effect are visible only in retrospect, so one probes, senses and responds; in chaos one acts first. The framework's origin in knowledge management matters: it descends from the same codification debates as Boisot's I-Space, asking which knowledge can be structured and which only enacted.

In the taste-and-judgement argument Cynefin is an anchor with an asterisk. It supplies the domain-sensitivity the vague phrase gestures at: deciding whether a given AI-augmented task sits in the knowable (delegate and verify) or the complex (probe with the tool, sense, respond) is a precise question where 'apply judgement' is not. The asterisk: Cynefin itself circulates through the consultancy-framework economy that the boundary-work critique targets, so citing it against taste-talk invites a tu quoque — worth conceding and moving past, since the framework's distinctions survive the sociology of its distribution.

Kurtz, C. F., & Snowden, D. J. (2003). The new dynamics of strategy: Sense-making in a complex and complicated world. IBM Systems Journal, 42(3), 462–483.

The standard reference for the embodied–embedded–enacted–extended synthesis: cognition is not inner symbol manipulation but skillful engagement of a bodily agent with its environment, sometimes constitutively including external tools and structures.

The handbook consolidates four decades of anti-Cartesian cognitive science — from Varela, Clark and Chalmers to enactivist perception and material engagement theory — into a field map: the mind is embodied (shaped by its physical substrate), embedded (scaffolded by its niche), enactive (constituted in sensorimotor engagement) and extended (partly composed of external resources). Disagreements among the four E's are themselves productive research questions.

As an anchor in the taste-and-judgement argument, 4E grounds the claim that 'taste' is trained perception, not an inner faculty: discriminating quality is a skill of an embodied agent formed by a history of engagement with materials, tools and exemplars. It also frames the live question about generative tools — whether they become genuinely incorporated scaffolding (extended cognition, transparent in use like the blind man's cane) or remain consulted oracles whose outputs are inspected — since the two relationships build different capacities and different dependencies.

Newen, A., De Bruin, L., & Gallagher, S. (Eds.) (2018). The Oxford Handbook of 4E Cognition. Oxford: Oxford University Press.

Aristotle's ethics distinguishes the intellectual virtues — episteme (demonstrative knowledge of what cannot be otherwise), techne (reasoned capacity to make) and phronesis (practical wisdom: deliberating well about what is good and doable in particular situations). Virtue is acquired by habituation: we become just by doing just acts, skilled by exercising skill.

Book VI is the taxonomy under the entire taste-and-judgement discussion. Phronesis differs from techne in having no product separable from the acting, and from episteme in concerning particulars — which is why Aristotle says the young can master mathematics but not practical wisdom: it requires experience of cases. Judgment of the particular cannot be derived from universal rules; the practically wise person perceives the salient features of the situation, and that perception is trained by habituation, not instruction alone.

Every framework in this collection inherits from here: Flyvbjerg's phronetic research and Vallor's technomoral virtues explicitly, the tacit-knowledge and skill-acquisition literatures structurally, and the habituation thesis anticipates both deliberate practice and the missing-rungs problem — capacities exist only where they are exercised. The 2,300-year-old lesson for AI-augmented work is exact: what techne the machine absorbs matters less than whether the arrangements of work still habituate phronesis, since that is the capacity 'taste and judgement' gestures at without naming.

Aristotle (c. 340 BCE). Nicomachean Ethics (R. Crisp, Trans. & Ed.). Cambridge: Cambridge University Press, 2000.

Hume's essay poses the modern taste problem: aesthetic verdicts vary widely, yet some judgments are evidently better than others. His answer locates the standard not in rules but in the qualified critic — formed by practice, comparison, delicacy of discernment and freedom from prejudice.

Hume accepts both horns of the taste dilemma: sentiment is subjective, yet the verdict that Ogilby equals Milton would be absurd. His resolution is procedural rather than criterial — there is no rule that certifies quality, but there are better and worse judges, and what makes a judge better can be specified: extensive practice, wide comparison across cases, delicacy of taste, good sense and freedom from prejudice. Joint verdicts of such critics constitute the standard.

For the taste-and-judgement discussion, Hume shows the 18th century already did the analytical work the AI discourse skips: he converts 'taste' from a mysterious possession into a trainable, criterion-governed competence of the judge, while conceding that the judgments themselves cannot be proceduralized. The parallel question — what makes someone a qualified judge of AI output, and what practice and comparison would train that judge — is precisely what current invocations of 'taste' leave unasked.

Hume, D. (1757). Of the standard of taste. In Four Dissertations. London: A. Millar.

Establishes aesthetic judgment as a distinct cognitive faculty. His analysis of 'free beauty' versus 'dependent beauty' introduces purposiveness without purpose — indispensable for thinking about designed objects whose form exceeds mere function. Virtually all modern design theory operates within or in reaction to the problems Kant formulated here.

Kant’s Critique of Judgment (1790) provides the basic conceptual toolkit through which modern design theory has come to think about beauty, value, and meaning in designed artifacts, even though Kant himself never addresses industrial design or mass production.

Aesthetic judgment: subjective yet universal

Subjective: grounded in feeling, not in concepts or rules; yet

Universal in claim: when we judge something beautiful, we implicitly expect others to agree.

This tension underlies later debates about taste, quality, and design value: when designers or critics claim that one solution is “better” or “more elegant” than another, they are effectively making Kantian-style universal claims on the basis of subjective experience.

Purposiveness without purpose and design

In beauty, our imagination and understanding are in a free, harmonious play.

The object appears as if designed for a purpose, but no specific, determinate purpose can be stated.

For design theory this is both powerful and troublesome:

It captures the sense that good design exceeds mere function: a chair, phone, or interface can be fully usable yet still feel “better than necessary,” rewarding contemplation and use in ways not captured by a spec sheet.

But it also separates beauty from function, encouraging a conceptual split between aesthetic value and practical utility that later functionalist traditions (e.g., modernism, Bauhaus, “form follows function”) try to overcome.

The sublime and non-comfort design

The sublime arises when sensory experience is overwhelmed (vastness, power, infinity), yet

This very failure of the senses reveals our rational capacity to think what cannot be fully perceived (the infinite, moral law).

Thus the sublime is not just aesthetic but moral: it discloses the superiority of reason over nature.

This becomes a foundation for design and architecture that aim not at comfort or prettiness but at awe, transcendence, or existential confrontation:

Monumental architecture, brutalist or minimalist spaces, and certain interaction or speculative designs seek to unsettle or elevate, not merely to please.

Modernist and Romantic aesthetics draw on this Kantian sublime to justify designs that prioritize intensity, abstraction, and moral seriousness over decorative charm.

Aesthetic ideas and the excess of meaning in design

Stimulate much thought, yet

Cannot be fully captured by any determinate concept or paraphrase.

Applied to design, this implies:

The best designs communicate more than their brief or functional description: they carry cultural, symbolic, or existential resonances that cannot be fully specified in advance.

Design quality involves an excess of meaning over purpose: a well-designed object or environment is not just adequate to its task; it suggests worlds of association, identity, and value.

This anticipates later distinctions such as:

Design vs. styling: styling as surface-level variation within fixed purposes; design as shaping the deeper meaning and experience of use.

The difficulty of writing exhaustive design specifications: what makes a design genuinely good often emerges only in the realized form, not in the initial requirements.

Design and Kant’s three domains of reason

Theoretical (what is: knowledge, science),

Practical (what ought to be: ethics, morality),

Aesthetic (what pleases or moves us: beauty, taste).

Design cuts across all three and fits neatly into none:

It uses empirical knowledge (materials, ergonomics, engineering).

It involves practical judgment (social impact, ethics, sustainability).

It requires aesthetic sensitivity (form, expression, experience).

This structural mismatch fuels ongoing disputes:

Is design a science, an art, or a practical discipline of its own kind?

Should design education sit in engineering schools, art academies, or independent design institutions?

Kant does not resolve these institutional questions, but his architecture of reason defines the conceptual map on which they are argued.

Historical mediation into design theory

Romantic aesthetics and German Idealism (e.g., Schiller) extend Kant’s ideas of freedom, play, and aesthetic education, feeding into views of design as a means of cultural and moral formation.

The Arts and Crafts movement tries to reunite beauty and utility, reacting against industrial ugliness while still working within a Kantian vocabulary of taste, genius, and purposiveness.

The Bauhaus explicitly negotiates between art and industry, echoing Kant’s tension between free aesthetic form and determinate practical purpose.

Phenomenology (Heidegger, Merleau-Ponty) reinterprets experience, embodiment, and worldhood in ways that reshape late 20th-century design research, but still within a space Kant helped open by distinguishing aesthetic from theoretical and practical reason.

Enduring conceptual vocabulary

Beauty, sublimity, purposiveness, taste, genius, aesthetic ideas are now standard terms for talking about:

Why some designs feel right or inevitable.

How artifacts can be morally or existentially charged.

Why certain works resist full functional or verbal reduction.

In this sense, the Critique of Judgment remains the deep grammar of Western design theory: it sets the questions and provides the conceptual tools through which debates about function, form, meaning, and value continue to be conducted.

Kant, Immanuel. Kritik der Urteilskraft. Berlin, 1790. Trans. Werner Pluhar. Indianapolis: Hackett, 1987.

Sibley's landmark paper argues that aesthetic terms — graceful, elegant, balanced, garish — are not condition-governed: no set of non-aesthetic features logically entails their application. Taste is precisely the perceptual capacity to apply such terms correctly where rules cannot.

Sibley's argument is a precision instrument. Non-aesthetic features (line counts, colour values, symmetries) are relevant to whether a work is elegant, but no conjunction of them is sufficient; a critic can cite them to support a verdict yet cannot derive the verdict from them. What bridges the gap is taste — an exercised sensitivity, educable by exemplars, comparison and the critic's pointing — not inference.

This is the exact logical shape of the problem with specifying quality criteria for generative output: style guides and evaluation rubrics list non-aesthetic conditions, and the qualities that matter refuse to reduce to them. Sibley explains both why 'taste' names something real — the application of non-condition-governed concepts — and why it cannot be handed to a model as a specification. The educability point matters equally: taste transmits through guided perception, which is an apprenticeship structure, not a document.

Sibley, F. (1959). Aesthetic concepts. The Philosophical Review, 68(4), 421–450.

Gadamer's founding work of philosophical hermeneutics opens by rehabilitating the humanist concepts of Bildung, sensus communis, judgment and taste as legitimate modes of knowing that scientific method cannot replace. Understanding is historically situated, tradition-formed and dialogical.

Truth and Method begins, significantly, with taste: Gadamer traces how the humanist tradition treated taste and judgment as genuine cognitive capacities — a sense of the fitting that cannot be derived from concepts — and how 19th-century methodologism demoted them to mere subjectivity. His rehabilitation runs through the sensus communis: taste is communal, formed within a tradition, and normative without being rule-governed.

The relevance to AI-augmented work is double. First, Gadamer supplies the strongest philosophical account of why 'judgement' resists codification into prompts, rubrics or process: application is not the mechanical use of a rule but part of understanding itself. Second, his insistence that taste is tradition-formed raises the reproduction question — a capacity constituted by participation in a practice community is exactly the kind of capacity that changes when the practice is automated.

Gadamer, H.-G. (1960). Wahrheit und Methode: Grundzüge einer philosophischen Hermeneutik. Tübingen: Mohr. English: Truth and Method.

Vickers dissolves the vague noun 'judgment' into the operation of appreciative systems: readinesses to notice and value certain aspects of situations, exercised as reality judgments (what is the case), value judgments (what ought to be) and instrumental judgments (what can be done). Judgment is an art, but an analyzable one.

Drawing on forty years in law and public administration, Vickers rejected both rational-choice models and mystified appeals to intuition. His appreciative system names the learned structure of attention and valuation a practitioner brings to a situation: it determines what is even seen as a fact or a problem before any decision occurs. Appreciation changes through exercise — each judgment resets the system that will make the next one — which is why judgment can be educated but not specified.

There is a canonical systems book literally titled The Art of Judgement, and current taste-and-judgement discourse shows no sign of having read it. Vickers's triad gives the parsimony argument its sharpest edge: when someone claims AI work needs 'judgement', one can now ask which kind — reality, value or instrumental — and whose appreciative system is being exercised and re-formed. Via Checkland's soft systems methodology, the concept also connects directly to the design-methods lineage the timeline traces.

Vickers, G. (1965). The Art of Judgement: A Study of Policy Making. London: Chapman & Hall.

Arendt's posthumously published lectures find in Kant's Critique of Judgment the unwritten political philosophy of judging: the capacity to think from the standpoint of everyone else — the enlarged mentality — exercised by spectators who must woo the assent of others without the compulsion of proof.

Arendt takes Kant's account of taste — judging particulars without a concept, claiming general though not provable assent — and reads it as the model for judging human affairs, where general rules are precisely what is missing. Judgment works through representative thinking: making present to oneself the perspectives of others, then judging in anticipation of their agreement. It is irreducibly social; the solitary person of taste is a fiction.

Against the individualist framing of 'taste and judgement' as a personal asset the AI-era professional must possess, Arendt relocates judgment in the space between people: it is exercised toward a community of judges and validated by wooed, not compelled, agreement. For teams arguing about whether AI-assisted work is good, this reframes the activity — the argument itself, conducted with enlarged mentality, is the judgment, not a report of private discernments.

Arendt, H. (1982). Lectures on Kant's Political Philosophy (R. Beiner, Ed.). Chicago: University of Chicago Press.

Eisner's methodological statement of educational connoisseurship and criticism: connoisseurship is the private art of appreciation — noticing subtle, significant qualities — while criticism is the public art of disclosure that makes those qualities visible to others. Together they form a rigorous evaluation practice grounded in cultivated perception.

Borrowing deliberately from wine tasting and art criticism, Eisner argued that expert evaluation of complex practice depends on connoisseurship — a trained capacity to perceive qualities that untrained observation misses — and that connoisseurship becomes publicly useful only through criticism, the crafted disclosure that lets others see what the connoisseur sees. He built validity criteria for this (structural corroboration, referential adequacy): criticism is checked by whether it enables new perception.

Eisner is the working model the taste-discourse lacks: taste made methodical and accountable. The connoisseurship/criticism split maps directly onto AI-augmented review work — privately noticing that generated output is subtly wrong is connoisseurship; the hard, teachable, teamwork-relevant skill is criticism, articulating the noticing so colleagues and juniors can come to see it too. A team whose 'taste' lives only in private vetoes has connoisseurship without criticism, and cannot reproduce its own judgment.

Eisner, E. W. (1991). The Enlightened Eye: Qualitative Inquiry and the Enhancement of Educational Practice. New York: Macmillan.

Flyvbjerg argues social inquiry fails when it imitates natural science and succeeds as phronetic research: inquiry organized around Aristotle's phronesis — context-dependent practical wisdom concerning values and power — rather than episteme (universal theory) or techne (technical production). Drawing on the Dreyfus skill model, he shows expertise is intuitive and case-based.

Flyvbjerg's operative move is to revive the Aristotelian triad — episteme, techne, phronesis — and argue that the distinctive competence of social practice is the third: knowing what is good and feasible in this particular case, for these people, given these power relations. Phronesis cannot be formalized, is learned through cases, and matures via the Dreyfus ladder from rule-following novice to intuitive expert; hence his insistence on the force of example and case research.

For the taste-and-judgement argument, Flyvbjerg carries the 2,300-year-old vocabulary the discourse collapses into one word. What AI plausibly automates maps onto techne and fragments of episteme; what the discourse gestures at with 'judgement' is phronesis — and naming it that immediately imports the tradition's hard-won insights: it is value-laden, power-entangled, case-trained and unteachable by rule. The question 'how do we keep phronesis reproducing when the techne is automated' is answerable in a way 'how do we keep taste' is not.

Flyvbjerg, B. (2001). Making Social Science Matter: Why Social Inquiry Fails and How It Can Succeed Again (S. Sampson, Trans.). Cambridge: Cambridge University Press.

Nelson and Stolterman articulate design as a distinct culture of inquiry — neither science nor art — whose core competence is design judgment. They provide an explicit taxonomy: framing, default, appearance, quality, instrumental, navigational, compositional, connective and core judgments.

The Design Way grounds design in service of the particular: unlike science's true and ethics' good, design pursues the desirable in a specific situation for specific people. Because there is no algorithm from the general to the particular, judgment carries the weight — and Nelson and Stolterman refuse to leave it as one word, distinguishing the judgment that frames a situation from the judgments of appearance, of quality, of composition, of navigation through a project, down to the core judgments that constitute a designer's character.

This is the design-native version of the parsimony argument: the discrete concepts that 'taste and judgement' muddles already exist, named and distinguished, inside design theory itself. Anyone claiming AI-augmented design work needs 'judgement' can be asked which of Nelson and Stolterman's kinds they mean — and the answers have different implications for tooling, training and delegation. A model may assist compositional judgment long before it touches framing or core judgment.

Nelson, H. G., & Stolterman, E. (2003). The Design Way: Intentional Change in an Unpredictable World. Englewood Cliffs, NJ: Educational Technology Publications. (2nd ed. MIT Press, 2012.)

Vallor develops a technomoral virtue ethics drawing on Aristotelian, Confucian and Buddhist traditions: practical wisdom is a cultivated capacity, and specific technologies either give it exercise or let it atrophy. She names moral deskilling as a live risk of automation.

Vallor's central claim is that virtues — including the executive virtue of practical wisdom — exist only as cultivated habits, and habits require practice. Technologies restructure the practice landscape: what they take over, we stop rehearsing. Her analysis of moral deskilling (in care work, warfare, social interaction) shows the mechanism concretely, without technophobia — the same technologies can be designed and used to extend rather than erode the relevant capacities.

Applied to taste-and-judgement talk, Vallor converts a slogan into a design question: if judgment is a virtue in the technical sense — maintained by exercise — then every delegation decision in an AI-augmented workflow is also a decision about which capacities keep practicing. The discussion 'how do we use these tools effectively' becomes 'which exercises of practical wisdom must the workflow deliberately retain', which is answerable in a way exhortations to have taste are not.

Vallor, S. (2016). Technology and the Virtues: A Philosophical Guide to a Future Worth Wanting. New York: Oxford University Press.

Ryle's attack on the 'ghost in the machine' distinguishes knowing-how from knowing-that and shows by regress that intelligent performance cannot consist in the prior consultation of rules, since applying a rule intelligently is itself an exercise of know-how.

Ryle's regress argument is short and lethal: if intelligent action required first considering a proposition about how to act, then that considering would itself need to be done intelligently, requiring a prior consideration, and so on. Therefore know-how is not reducible to knowledge of propositions; skill is a disposition exhibited in performance, judged by its exercise rather than by an inner process.

This is the logical bedrock under every later account of tacit skill — Polanyi, Dreyfus, Schön all inherit it. For the taste-and-judgement discussion it explains a practical frustration: why writing better and better instructions (prompts, guidelines, checklists) never quite captures what the skilled reviewer does, since following instructions well is itself the skill. The capacity the discourse calls 'judgement' lives on the know-how side of Ryle's line, which is why it transmits through practice and exemplars rather than documentation.

Ryle, G. (1949). The Concept of Mind. London: Hutchinson.

Polanyi's compact statement of tacit knowing: we know more than we can tell. All knowing has a from–to structure — we attend from subsidiary particulars to a focal whole — so the particulars of skilled performance are unspecifiable in principle while being exercised.

Polanyi's account is structural, not mystical: in recognizing a face or riding a bicycle we rely on particulars subsidiarily while attending focally to the whole, and turning attention onto the particulars disrupts the performance. Knowledge is personal — it involves indwelling, the incorporation of tools and clues into the body's extended awareness — and is transmitted by apprenticeship to a master's practice rather than by specification.

Polanyi is the standard citation behind every appeal to inarticulable skill, and he sets the test that taste-claims should face: is this capacity genuinely tacit in Polanyi's structural sense, or merely unarticulated because nobody has done the work? The distinction matters enormously for AI-augmented practice — the first kind of knowledge survives only in practitioners and their apprentices; the second kind can be partially codified, prompted and delegated. Boisot's social learning cycle, already in this timeline, is effectively the organizational dynamics of moving knowledge across that line.

Polanyi, M. (1966). The Tacit Dimension. Garden City, NY: Doubleday.

The Dreyfus brothers' five-stage model of skill acquisition — novice, advanced beginner, competent, proficient, expert — argues that expertise is intuitive, situational coping that leaves rules behind, and on that basis critiques the expert-systems program of 1980s AI.

The model's point is developmental: novices need context-free rules, but each stage of increasing skill involves less rule-following and more holistic, situationally discriminated response, until the expert simply sees what to do. Written against the claim that expertise could be extracted into rule bases, the book argued that asking experts for their rules yields regressions to novice-level articulations — the knowledge is in the doing.

There is a pleasing symmetry in citing this against today's taste-discourse: the canonical phenomenology of expertise was produced in the last argument with AI. Two implications carry over. The 'judgement' that reviews AI output is stage-five perception, not rule application — so it cannot be captured in review checklists. And the model predicts the missing-rungs problem precisely: if AI performs the novice-to-competent work, the rule-following stages where humans previously started, the developmental ladder to intuitive expertise loses its lower rungs.

Dreyfus, H. L., & Dreyfus, S. E. (1986). Mind over Machine: The Power of Human Intuition and Expertise in the Era of the Computer. New York: The Free Press.

The founding paper of deliberate-practice research: expert performance is the product of extended, effortful, individually designed practice with immediate feedback — not innate talent or mere experience. Level of performance correlates with accumulated deliberate practice, not years on the job.

Studying violinists and pianists, Ericsson and colleagues distinguished deliberate practice — designed activities targeting specific weaknesses, with feedback and repetition, effortful and not inherently enjoyable — from both work and play. The distinction explains why sheer experience plateaus: routine performance does not improve what it does not strain. Expertise is manufactured by a specific practice regime, sustained over roughly a decade.

The uncomfortable question this poses to taste-and-judgement discourse: deliberate practice requires performing the task, at the edge of one's ability, with feedback — and AI-augmented workflows systematically remove exactly those conditions, converting production tasks into review tasks. If discrimination and production are trained together (as the craft literature also insists), a generation that reviews without producing accumulates no deliberate practice in the capacity everyone says will matter most. Naming the mechanism makes the training-design problem tractable.

Ericsson, K. A., Krampe, R. T., & Tesch-Römer, C. (1993). The role of deliberate practice in the acquisition of expert performance. Psychological Review, 100(3), 363–406.

Klein's field studies of fireground commanders, nurses and military officers produced the recognition-primed decision model: experts rarely compare options; they recognize situations as typical, retrieve a workable course of action, and test it by mental simulation.

Against decision-analytic models, Klein's naturalistic decision making program studied experts under time pressure, uncertainty and high stakes, and found almost no option comparison. Expertise works through a large repertoire of patterns: the situation is recognized, expectations and cues come with the recognition, and a single candidate action is evaluated by mentally simulating it. Deviations from expectation — the sense that something is off — are data.

Sources of Power gives the empirical content of the 'judgement' that reviews AI output: pattern recognition against a repertoire, plus mental simulation of consequences. Both components depend on a history of real cases with real feedback — which is exactly what a review-only role accumulates slowly, if at all. Klein's account also identifies what to protect in workflow design: the encounters that build the pattern repertoire, and the expectation-violations that make experts sensitive to subtle wrongness.

Klein, G. (1998). Sources of Power: How People Make Decisions. Cambridge, MA: MIT Press.

Collins and Evans build a periodic table of expertises, centrally distinguishing contributory expertise (the ability to do a practice) from interactional expertise (fluent mastery of a practice's language without the ability to perform it), acquired through deep linguistic immersion in the expert community.

The book's normative project — who should legitimately weigh in on technical questions — required an analytical one: a taxonomy of what expertise is. Its lasting contribution is interactional expertise: managers, interpreters, critics and peer reviewers can acquire real, testable competence in a domain's language and judgments through immersion in its community, without contributory skill. Collins demonstrated this empirically with imitation-game experiments in gravitational-wave physics.

Interactional expertise is arguably the single best-fitting concept for AI-augmented work: directing, briefing and assessing work one does not perform oneself is precisely the interactional stance. Two consequences follow. The capacity the discourse calls 'taste' may be sustainable as interactional expertise — but Collins and Evans insist it is parasitic on a living community of contributory experts to be immersed in; if AI hollows out the contributory community, the interactional layer loses its host. That dependency is the sharpest version of the reproduction problem.

Collins, H., & Evans, R. (2007). Rethinking Expertise. Chicago: University of Chicago Press.

The adversarial collaboration between the heuristics-and-biases and naturalistic decision-making traditions converges on a joint verdict: expert intuition is trustworthy only where the environment offers valid, stable cue–outcome regularities and the practitioner gets adequate opportunity to learn them through feedback.

Kahneman (whose tradition catalogued intuition's failures) and Klein (whose tradition documented its successes) set out to find their real disagreement and found little: both accept that skilled intuition is recognition, and that it develops only under two conditions — a high-validity environment, and prolonged practice with timely, unambiguous feedback. Firefighters and chess players meet the conditions; stock pickers and political pundits do not, and their confident intuitions are pseudo-expertise.

This paper supplies the discriminating question taste-talk never asks: is judging AI output a high-validity, feedback-rich environment? Where quality verdicts get tested against outcomes (does the code run, does the design ship and perform), intuitive 'taste' can become real expertise; where verdicts are never checked — fluent text approved into a void — confidence will grow without validity. The paper turns 'trust your taste' from a slogan into an empirical claim that is sometimes false, and tells you how to make it true: build the feedback loop.

Kahneman, D., & Klein, G. (2009). Conditions for intuitive expertise: A failure to disagree. American Psychologist, 64(6), 515–526.

Todd, Gigerenzer and the ABC Research Group's statement of ecological rationality: simple heuristics succeed not despite ignoring information but because their structure matches the structure of the environment. Rationality is a fit between mind and world, not calculation.

Against the view of heuristics as biased shortcuts, the ecological program shows formally when fast-and-frugal strategies outperform optimization: in environments with scarce data, redundancy among cues, or instability, less can be more. The normative question shifts from 'did the reasoner follow logic' to 'does this strategy fit this environment' — which makes the environment, not the mind alone, the unit of analysis.

For the taste-and-judgement question, ecological rationality reframes what generative tools do: they change the environment's statistical structure — the base rates of fluency, the cue validities that old heuristics relied on. A reviewer's trained heuristic 'fluent and well-structured implies competent author' was ecologically rational for decades and is now invalid, because fluency no longer correlates with the qualities it used to signal. 'Judgement' failing around AI output is often not a character deficit but a heuristic–environment mismatch — a diagnosis that points to re-training and environment design rather than exhortation.

Todd, P. M., Gigerenzer, G., & the ABC Research Group (2012). Ecological Rationality: Intelligence in the World. New York: Oxford University Press.

Aesthetic experience is not a special category separate from everyday life but the fullest expression of lived experience. By dissolving the boundary between fine and applied arts, Dewey provided a philosophical basis for design's claim to aesthetic seriousness. An antecedent to Langer's 1953 work.

Dewey’s Art as Experience reframes aesthetics from being about special objects to being about a distinctive quality of experience that can arise in any human activity. He rejects the “museum conception” of art, where value is locked into rarefied objects isolated from everyday life, and instead claims that aesthetic experience is continuous with ordinary doing and undergoing—cooking, conversation, work, sport, tool use.

What marks something as art, on this view, is not the category of object but the character of the experience: a unified, intense, consummatory episode in which means and ends are integrated and the person is fully engaged—what he calls an experience. Art is thus paradigmatic of a way of experiencing, not a separate realm of things.

For design, this implies that the designer’s primary responsibility is not to produce beautiful artifacts but to shape conditions for meaningful, coherent, and fulfilling experiences. It shifts emphasis from object to experience, from passive contemplation to active engagement, and from isolated works to situated interactions. This anticipates later developments such as service design, user experience, interaction design, experience design, and co-creation.

Dewey’s impact on design has been both conceptual and historical. Charles and Ray Eames exemplify his pragmatist aesthetics in their fusion of learning, pleasure, communication, and everyday life. Richard Buchanan’s claim that design is a liberal art draws explicitly on Dewey. Donald Schön’s notion of the reflective practitioner is Deweyan in treating professional knowledge as emergent in practice rather than applied from detached theory. Interaction design’s focus on temporally unfolding use, engagement, and participation operates within the experiential framework Dewey articulated.

Dewey also criticizes formalism in aesthetics: he denies that form can be abstracted from content, insists that perception is historically and situationally conditioned, and treats meaning as an emergent property of the transaction between work and audience. These ideas prefigure reception theory, reader-response criticism, and participatory design, all of which understand meaning as co-created rather than simply transmitted from maker to passive receiver.

Dewey, John. Art as Experience. New York: Minton, Balch & Company, 1934.

Pye distinguishes the workmanship of risk (outcome depends on the maker's skill at every moment) from the workmanship of certainty (outcome predetermined by tooling), dismantling sentimental craft rhetoric in the process.

Pye provides the precise conceptual key to the quality-vs-scale flip in consumer electronics: the historical moment when certainty-workmanship at extreme volume began to exceed risk-workmanship in achievable quality. Luxury had always marketed risk-workmanship; Apple demonstrated that the best-made objects most people own could be made in the hundreds of millions, and Vertu's collapse proved the corollary. A design-theory classic that this collection reads as economics.

Pye, D. (1968). The Nature and Art of Workmanship. Cambridge: Cambridge University Press.

Sennett's study of craft as embodied skill and intrinsic motivation — the desire to do a job well for its own sake.

Grounds the craftsmanship facet of luxury and frames the open question this collection keeps returning to: if software is workmanship of certainty by definition, what is digital craft? Sennett's answer — craft as a relation between maker and work, not a production mode — suggests the question is answerable.

Sennett, R. (2008). The Craftsman. New Haven: Yale University Press.

Stiegler's compact statement of proletarianization: grammatization — the discretization of gesture, speech and thought into reproducible technical systems — progressively strips workers, consumers and now knowledge professionals of their savoir-faire and savoir-vivre.

Extending Marx's observation that the machine, not the worker, becomes the bearer of skill, Stiegler traces successive waves of grammatization: writing exteriorized memory, industrial machinery exteriorized the worker's gesture, and digital systems exteriorize cognition itself. Each wave is pharmacological — both poison and remedy — but unmanaged, it produces proletarianization: the loss of knowledge from the humans who once carried it.

Stiegler is the pessimistic pole of the taste-and-judgement debate, and worth stating at full strength: on this reading, generative AI is the grammatization of design judgment itself, and the discourse's anxious insistence on 'taste' is the sound of a profession sensing its proletarianization. The pharmacological framing prevents mere doom: the same externalization that de-skills can be the support of new capability, but only under deliberate arrangements — which is exactly what the 'how do we use these tools' discussion should be designing.

Stiegler, B. (2010). For a New Critique of Political Economy (D. Ross, Trans.). Cambridge: Polity. (Original French, Galilée, 2009.)

Ingold argues against the hylomorphic model — form imposed by mind on passive matter — for making as correspondence: the maker joins forces and flows of materials, following where they lead. Skill is attentive responsiveness in this ongoing exchange, not executed intention.

Through carpentry, flint-knapping, basket-weaving and drawing, Ingold shows that no maker simply stamps a prior design onto the world: the grain resists, the clay slumps, and competence is the ability to keep corresponding — to modulate intention through what the material does back. Thinking happens in the making, not before it; the anthropology of skill here converges with 4E cognition from the craft side.

Ingold undercuts the picture implicit in taste-talk of the designer as taste-haver who evaluates outputs from outside. If quality emerges in correspondence, then the question about generative tools is what kind of material they are — what resistances, invitations and back-talk a model offers — and whether prompt-response cycles sustain the attentive correspondence in which judgment forms, or reduce making to selecting among finished offerings. That reframing turns a debate about personal virtue into one about the texture of the new medium.

Ingold, T. (2013). Making: Anthropology, Archaeology, Art and Architecture. London: Routledge.

The classic statement of automation's central paradox: the more you automate a process, the more crucial — and the more difficult — the human's residual role becomes. Automating the easy parts leaves operators with the hardest tasks (monitoring, and taking over when the system fails) while eroding the very skills and situation-awareness needed to do them. Foundational to human factors, supervisory control, and every later 'human-in-the-loop' debate.

Lisanne Bainbridge's 'Ironies of Automation' (1983), first presented at a 1982 IFAC conference on man-machine systems, is among the most cited papers in human factors. Its argument is deceptively simple: designers automate the parts of a task that are easy to automate, leaving the human with whatever is left over. But what's left over is precisely the hardest part — monitoring a system that runs itself, and intervening in the rare, high-stakes moments when it fails. The automation meant to remove human error instead concentrates responsibility at exactly the points where humans are least equipped to act.

Two ironies do the work. First, the operator is asked to monitor for abnormalities, yet humans are poor at sustained passive vigilance — the design that sidelines them also undermines their attention. Second, the manual and cognitive skills needed to take over in an emergency decay without regular practice, so automating routine operation deskills the very operator the system relies on in a crisis. Bainbridge's conclusion is not anti-automation but a design challenge: keep the human meaningfully engaged, preserve their skills, and design human and automation as a single system.

For the AI-as-participant debate the paper is uncannily current. Generative tools automate the fluent middle of design and writing and hand back the framing, judgment and error-catching — the residual hard parts — to a human whose skills may atrophy precisely because the tool did the practice. Read alongside Weizenbaum and Suchman, Bainbridge supplies the operational version of 'thinking partner, not replacement': partnership has to be designed, or automation quietly removes the human's capacity to participate at all.

Bainbridge, L. (1983). Ironies of Automation. Automatica, 19(6), 775-779. Originally presented at the IFAC/IFIP/IFORS/IEA Conference on Analysis, Design and Evaluation of Man-Machine Systems, Baden-Baden, 1982.

Demolished the cognitivist assumption that people follow internal plans. Action is situated and improvised; plans are retrospective accounts. Based on ethnographic study of Xerox photocopier use. Directly challenges Simon's rational planning model.

Lucy Suchman’s Plans and Situated Actions: The Problem of Human-Machine Communication (1987) sits at a pivotal moment in the history of human–computer interaction (HCI), artificial intelligence (AI), and cognitive science. Written from within Xerox PARC but grounded in sociology and anthropology, the book stages a direct confrontation between two very different ways of understanding human action: the cognitivist, plan-based model and the ethnomethodological, situated-action perspective. Its impact was to unsettle the taken-for-granted assumptions of AI and HCI and to reorient design practice toward the empirical study of real-world activity.

Suchman, L. A. (1987). Plans and Situated Actions: The Problem of Human-Machine Communication. Cambridge: Cambridge University Press.

Parasuraman and Riley organise human-automation problems into use, misuse (overreliance), disuse (underreliance after false alarms), and abuse (designer or management deployment without regard for human consequences), showing that reliance tracks trust rather than actual reliability.

The fourfold scheme moves the analysis of automation failure up a level: from single interactions to patterns of reliance shaped by experience, workload, and — crucially — by what the automation reveals about its own competence. Misuse and disuse are both calibration failures, and calibration requires evidence the interface may never supply.

The abuse category quietly introduces the political dimension the cluster's policy cases make explicit: reliance patterns are also chosen for operators by those who deploy automation. The line from automation abuse to Horizon and Robodebt — institutions insisting on the machine's word against human testimony — is short.

Parasuraman, R., & Riley, V. (1997). Humans and automation: Use, misuse, disuse, abuse. Human Factors, 39(2), 230–253.

Lee and See define the goal of trust design as appropriate reliance: trust calibrated to actual trustworthiness, with sufficient resolution and specificity that operators trust the right functions in the right contexts rather than the system wholesale. Miscalibration — overtrust and distrust alike — stems from evidence the interface fails to provide.

Calibration, resolution, and specificity give the essay measurable form for its trust claims: an interface that shows only outcomes produces coarse, brittle trust, while one that exposes its competence boundaries lets trust vary function by function. The AF447 crew's undifferentiated trust in 'the automation' is a resolution failure the flight deck design entailed.

Together with Hoff and Bashir, this paper lets the essay state its thesis in trust vocabulary: visibility is what makes appropriate reliance possible, and invisible design decisions are, among their other harms, calibration sabotage.

Lee, J. D., & See, K. A. (2004). Trust in automation: Designing for appropriate reliance. Human Factors, 46(1), 50–80.

Hollnagel and Woods consolidate cognitive systems engineering around the joint cognitive system: human and machine analysed as one unit whose performance depends on coordination. Effective joint systems require mutual observability and directability — each agent able to see and shape what the other is doing.

The book reframes the cluster's unit of analysis: the question is not whether the human's model matches the machine, but whether the joint system can keep itself coordinated. Observability — the machine making its state and intent available — becomes a first-class engineering requirement, the cybernetic demand of Conant–Ashby translated into design language.

For the essay it supplies the mature theoretical frame in which the aviation, platform, and policy cases can be compared: each is a joint cognitive system in which one side has been made unobservable and undirectable to the other. Its Ashby lineage also ties the cluster back to the timeline's systems-thinking spine.

Hollnagel, E., & Woods, D. D. (2005). Joint Cognitive Systems: Foundations of Cognitive Systems Engineering. Boca Raton, FL: CRC Press.

Risko and Gilbert's framework paper defines cognitive offloading — using physical action or external tools to reduce internal processing demands — and reviews when people offload, what governs the decision (often miscalibrated metacognitive beliefs), and what offloading does to internal capacity.

The paper organizes a scattered literature: tilting the head instead of rotating the image, writing intentions instead of remembering them, trusting the search engine instead of encoding the fact. Two findings matter most. The decision to offload is driven by metacognitive evaluations that are systematically biased — people offload more than optimal when tools feel effortless — and offloading has downstream costs: what is reliably externalized is less well encoded, and monitoring of the external store is often lax.

This is the lab-science counterpart to the deskilling worry, at the scale of minutes rather than careers, and it gives the taste-and-judgement discussion measurable variables: offloading propensity, metacognitive calibration, monitoring accuracy. The 4E tradition celebrates the extended mind; Risko and Gilbert supply its audit — extension is a trade, and the terms of the trade with generative tools (what internal capacity is given up, how well the external resource is monitored) can be studied rather than asserted.

Risko, E. F., & Gilbert, S. J. (2016). Cognitive offloading. Trends in Cognitive Sciences, 20(9), 676–688.

Elish coins the moral crumple zone: in highly automated systems, the human operator — often with limited actual control — absorbs the moral and legal impact when the system fails, as a crumple zone absorbs a crash. Cases from Three Mile Island to autonomous-vehicle safety drivers show responsibility misattributed to whoever was nearest.

Elish closes the loop the cluster opens: after the invisible decisions have produced the fail state, the attribution machinery assigns it to the operator whose model the system starved. The crumple-zone metaphor explains why the backwards trace the essay performs is both necessary and resisted — blame that stops at the human is cheaper for everyone upstream.

Read with Reason's latent conditions and Sarter and Woods's mode analysis, the paper turns a design observation into an accountability principle: control and responsibility must be allocated together, and interfaces that hide the former while assigning the latter are doing moral work, not just informational work.

Elish, M. C. (2019). Moral crumple zones: Cautionary tales in human-robot interaction. Engaging Science, Technology, and Society, 5, 40–60.

Engeström's extension of Vygotskian activity theory: the unit of analysis is the collective, tool-mediated activity system — subject, object, instruments, community, rules, division of labor — whose internal contradictions drive expansive learning, the collective construction of a qualitatively new form of the activity.

Where individual learning theories ask how a person acquires existing competence, Engeström asks how work communities learn things that do not yet exist. His answer runs through contradictions: tensions within and between elements of an activity system (a new instrument against old rules, a changed object against an old division of labor) surface as disturbances and dilemmas, and their collective resolution — expansive learning — reorganizes the whole system.

Activity theory gives the taste-and-judgement discussion its proper unit of analysis. A generative model is a new mediating instrument inserted into an activity system, and the current turbulence — juniors without rungs, reviewers without feedback, quality disputes without shared criteria — is the predictable pattern of secondary contradictions between the new instrument and the old rules, community and division of labor. On this reading the discourse's 'taste' is a symptom; the expansive-learning question is what qualitatively new form of design activity resolves the contradictions, and who gets to construct it.

Engeström, Y. (1987). Learning by Expanding: An Activity-Theoretical Approach to Developmental Research. Helsinki: Orienta-Konsultit.

Lave and Wenger relocate learning from individual heads to communities of practice: newcomers become competent through legitimate peripheral participation — doing real but low-stakes parts of the practice at its edges, moving inward as their participation deepens.

Studying apprenticeships — tailors, midwives, quartermasters, butchers, nondrinking alcoholics — Lave and Wenger found that learning is not teaching's product but participation's: the curriculum is the practice itself, seen from a legitimate position at its periphery. Skill, identity and membership develop together; the apprentice becomes a tailor by tailoring peripherally, absorbing the field of mature practice from within it.

This names the reproduction problem in the taste-and-judgement debate with precision: judgment is apprenticed through exactly the peripheral tasks — the first drafts, the production work, the low-stakes real contributions — that AI now absorbs. Remove legitimate peripheral participation and the community keeps its masters but loses its route to making new ones; the missing-rungs problem is structural, not motivational. Any serious answer to 'how do we use these tools' has to redesign the periphery, not exhort juniors to develop taste.

Lave, J., & Wenger, É. (1991). Situated Learning: Legitimate Peripheral Participation. Cambridge: Cambridge University Press.

Weick's canonical statement of organizational sensemaking: retrospective, social, identity-grounded, enactive of its environment, and driven by plausibility rather than accuracy. Organizations do not discover situations; they construct workable accounts to act from.

Weick's seven properties give sensemaking its analytical teeth: it is grounded in identity construction, retrospective, enactive, social, ongoing, cue-focused and plausibility-driven. People act, then look back at what they did to discover what the situation must have been; the accounts that survive are the ones that let action continue, not the ones that are most accurate.

Applied reflexively, Weick describes what organizations are doing when they reach for 'taste and judgement': an occupational community whose identity is disturbed by generative AI is retrospectively constructing a plausible account of what still requires humans — and the phrase's plausibility, not its precision, is what makes it spread. That is not a cynical dismissal but a prediction: the account will stabilize practice whether or not it is analytically sound, which is exactly why replacing it with sounder concepts (Vickers, Nelson & Stolterman, phronesis) matters now, while the sensemaking is still in flux.

Weick, K. E. (1995). Sensemaking in Organizations. Thousand Oaks, CA: Sage.

Bourdieu's empirical monument: status is signaled not just by spending but by taste — cultural capital that cannot simply be bought. English translation 1984.

Distinction is why quiet luxury exists, why knowing the obscure brand outranks buying the obvious expensive one, and why heritage-by-acquisition may fail once the decoder ring updates: the value of a signal depends on who can read it. For designers, Bourdieu is the reminder that every aesthetic choice is also a legibility choice about which audience can decode it.

Bourdieu, P. (1979). La Distinction. Paris: Minuit. English: Distinction (1984).

Gieryn's classic paper reframes the demarcation of science from non-science as boundary-work: a practical, rhetorical accomplishment in which practitioners attribute selected characteristics to their enterprise to defend autonomy, authority and resources — drawing the line differently depending on the opponent.

Where philosophers sought the criterion separating science from pseudo-science, Gieryn showed scientists themselves drawing the boundary opportunistically: against religion, science was empirical; against mechanics, it was theoretical; the line moved as the rhetorical need changed. Boundary-work is not deception — it is how occupations under pressure defend their jurisdiction with whatever contrast is locally effective.

'AI can generate, but it takes taste to judge' is boundary-work in Gieryn's exact sense: a line drawn between the real work and its mechanical imitation, positioned so the speaker lands on the valuable side. The tell is the line's mobility — as models improve, the boundary redraws (from making to curating to briefing to 'strategy') while the rhetorical structure stays constant. Naming this does not settle whether the claims are true; it explains why the phrase is doing so much work while meaning so little, and predicts where it moves next.

Gieryn, T. F. (1983). Boundary-work and the demarcation of science from non-science: Strains and interests in professional ideologies of scientists. American Sociological Review, 48(6), 781–795.

Abbott models professions as an ecology of competitors for jurisdiction over tasks: claims are staked through abstract knowledge, contested at the workplace, in public opinion and in law, and periodically upended by technological and organizational shocks that create, destroy and redistribute task areas.

Abbott's move was to stop studying professions one at a time and study the system: tasks are the territory, jurisdiction is the tie between a profession and its tasks, and abstraction is the weapon — a profession holds work by redescribing it in terms only its own knowledge system can treat. Shocks (new technology, new organizational forms) open jurisdictional vacancies and set off chains of competition down the status ladder.

Abbott predicts the current move with uncanny precision: when a technology absorbs a profession's routine tasks, the profession redescribes its work at a higher level of abstraction to relocate jurisdiction beyond the machine's reach. 'Taste and judgement' is exactly such a redescription — abstract enough that no benchmark can falsify it, positioned to retain the client relationship and the fee. The analysis neither vindicates nor debunks the claim; it identifies the contest the phrase is a move in, and suggests watching workplace jurisdiction (who actually signs off, who gets hired) rather than the rhetoric.

Abbott, A. (1988). The System of Professions: An Essay on the Division of Expert Labor. Chicago: University of Chicago Press.

Introduced 'boundary objects' — artefacts shared across communities while holding different meanings for each. One of the most cited concepts in design studies and CSCW.

Susan Leigh Star and James Griesemer’s 1989 article, “Institutional Ecology, Translations, and Boundary Objects: Amateurs and Professionals in Berkeley’s Museum of Vertebrate Zoology, 1907–39,” has become a foundational text across science and technology studies (STS), information science, organizational theory, and design research. Its enduring influence stems from the way it reframes a basic sociological problem: how heterogeneous actors manage to cooperate productively without first resolving their differences in worldview, expertise, and interest.

Historical situation and empirical setting

Star and Griesemer situate their argument in a rich historical case: the early decades of the Museum of Vertebrate Zoology (MVZ) at the University of California, Berkeley. Founded by Joseph Grinnell in the early twentieth century, the MVZ was not simply a collection of specimens; it was an institutional experiment in coordinating multiple social worlds around a shared enterprise of zoological research.

The museum’s success depended on the cooperation of at least four distinct groups:

Professional zoologists – university-based scientists pursuing evolutionary and ecological research, for whom specimens were data points in broader theoretical arguments.

Amateur naturalists and collectors – enthusiasts and local experts who contributed specimens and observations, often motivated by personal interest, prestige, or the pleasures of collecting.

University administrators – actors concerned with budgets, institutional prestige, and organizational stability, for whom the museum’s holdings were assets and indicators of success.

Trappers, hunters, and field workers – people who collected specimens as part of other economic activities, for whom animals were commodities or by-products rather than scientific objects.

The same physical object—a bird skin, for instance—circulated through these worlds with different meanings. To the professional zoologist it was a standardized specimen embedded in a system of classification and ecological inference; to the amateur collector, a trophy or token of field experience; to the trapper, a saleable good; to the administrator, an inventory item that could be counted, reported, and justified. The museum’s institutional ecology thus consisted of overlapping but non-identical practices, values, and interpretive frameworks.

Rather than assuming that cooperation requires shared beliefs or a unified culture, Star and Griesemer ask how the MVZ managed to function as a coherent institution despite these differences. Their answer is that the museum developed particular kinds of artifacts and organizational arrangements—what they call boundary objects—that enabled coordination without consensus.

Central claim: cooperation without consensus

The article’s central claim is that cooperation across social worlds is often achieved not by translating everyone into a single worldview, but by constructing shared objects that are simultaneously robust and flexible. These objects are robust in that they maintain a recognizable identity across contexts; they are flexible (or “plastic”) in that they can be interpreted and used differently by different groups.

This argument is explicitly positioned against the then-dominant translation model in STS, associated with Bruno Latour and Michel Callon. In the translation model, successful cooperation is described as the result of one actor (or network of actors) enrolling others into their project, gradually aligning interests and interpretations so that heterogeneous actors come to speak, act, and think in ways compatible with a central program.

Star and Griesemer do not deny that translation and enrollment occur, but they argue that this model underplays the persistence of difference. In many real-world settings, actors never fully converge on a single set of meanings or interests. Instead, they work together through partial, negotiated, and materially mediated alignments. Boundary objects are the key analytical tool for understanding these alignments.

Key concept: boundary objects

Star and Griesemer define boundary objects as artifacts that:

Are shared across different social worlds.

Are weakly structured in common use—there is enough structure to support coordination.

Are strongly structured in individual use—each group can impose its own local meaning and practice on the object.

Are plastic enough to adapt to local needs and constraints.

Are robust enough to maintain a recognizable identity across contexts.

Crucially, boundary objects do not require consensus about meaning. They work because they can be embedded in different interpretive schemes while still serving as points of connection. This allows heterogeneous actors to cooperate on concrete tasks—collecting specimens, building databases, designing services—without having to resolve deep epistemic or value disagreements.

Four types of boundary objects

Star and Griesemer identify four ideal-typical forms of boundary objects, all illustrated in the MVZ case but generalizable across domains:

Repositories

Ideal types

Coincident boundaries

Standardized forms

These four types are not exhaustive, but they illustrate the range of ways in which boundary objects can be designed and maintained to support cross-world coordination.

Reception and influence

The concept of boundary objects has been extraordinarily influential. It has been cited thousands of times and has traveled far beyond the original historical and institutional context.

In design research and practice

Design fields—especially participatory design, co-design, interaction design, and service design—have embraced boundary objects as a way to theorize the artifacts that mediate collaboration among diverse stakeholders.

Design processes typically involve designers, engineers, managers, clients, users, and other stakeholders, each with distinct vocabularies, priorities, and forms of expertise. Common design artifacts—such as:

Prototypes (physical or digital models),

Personas (fictional user profiles),

Customer journey maps,

Service blueprints,

Wireframes and mockups, and

Scenarios or storyboards

function as boundary objects. They are shared across groups, but each group reads them differently: engineers see technical constraints and implementation tasks; managers see risks, costs, and opportunities; users see potential experiences; designers see form, interaction, and coherence.

The effectiveness of these artifacts often depends on how well they balance robustness (clear enough to anchor discussion and decision-making) with plasticity (open enough to invite multiple interpretations and contributions). Star and Griesemer’s framework thus provides both a descriptive vocabulary and a normative guide for designing better collaborative tools.

In information science, classification, and infrastructure studies

Star’s later work with Geoffrey Bowker extends the boundary objects concept to large-scale information infrastructures—classification systems, coding schemes, diagnostic categories, and administrative taxonomies.

Here, boundary objects are not just local artifacts but components of global systems that coordinate action across vast organizational and geographical distances. Examples include:

International disease classifications (e.g., ICD codes),

Library cataloguing standards,

Database schemas in large organizations,

Administrative categories used in census-taking or social services.

These systems enable coordination—of research, funding, policy, and care—without requiring that all participants share the same understanding of what categories “really” mean. A diagnostic code, for instance, can be simultaneously a clinical tool, a billing category, a research variable, and a policy lever. The code is a boundary object that holds these uses together while allowing them to diverge.

Bowker and Star emphasize the politics of classification: boundary objects and infrastructures are not neutral. The way categories are defined, standardized, and enforced shapes whose perspectives are privileged, whose work is made visible or invisible, and whose lives are made legible to institutions. The boundary object concept thus becomes a way to analyze how social order is produced and maintained through material and informational arrangements.

Broader connections and implications

Star and Griesemer’s article connects several important themes in social theory and STS:

Multiplicity of social worlds: Rather than assuming a single, coherent “society” or “culture,” the article foregrounds partially overlapping social worlds with their own norms, practices, and ontologies.

Materiality and practice: Cooperation is not explained primarily through discourse or belief, but through the design and maintenance of concrete artifacts, forms, and infrastructures.

Non-consensual coordination: The article offers a powerful alternative to models that equate order with consensus. It shows how stable patterns of action can emerge from ongoing negotiation, partial alignment, and the strategic use of ambiguity.

Design of institutions: By focusing on the MVZ’s institutional ecology, Star and Griesemer highlight that successful organizations often depend on carefully crafted boundary objects that mediate between constituencies.

In contemporary contexts—distributed software development, global supply chains, interdisciplinary research, platform ecosystems—the problem Star and Griesemer identified has only intensified. Heterogeneous actors must cooperate across greater distances and deeper differences. Boundary objects remain a central tool for understanding how this is possible, and for critically examining the infrastructures that make some forms of cooperation easy while rendering others difficult or impossible.

In sum, “Institutional Ecology, Translations, and Boundary Objects” is not only a historical study of a particular museum; it is a general theory of how shared artifacts can hold together complex, heterogeneous social arrangements. Its enduring appeal lies in showing that coordination does not require erasing difference. Instead, it depends on the careful construction and maintenance of objects that are at once common and multiple, stable and flexible—objects that sit at the boundaries and make collective work possible.

Star, S. L., & Griesemer, J. R. (1989). Institutional ecology, 'translations' and boundary objects: Amateurs and professionals in Berkeley's Museum of Vertebrate Zoology, 1907–39. Social Studies of Science, 19(3), 387–420.

Boltanski and Thévenot's pragmatic sociology of critique: in disputes, people justify themselves by appeal to distinct orders of worth — inspired, domestic, fame, civic, market and industrial — each with its own test of greatness. Conflicts are often clashes between orders, resolved by tests and compromises.

Against both Bourdieu's unmasking of interests and rational-choice reduction, Boltanski and Thévenot took disputes seriously as argument: ordinary people competently justify and criticize by reference to a small plurality of worth regimes, each internally coherent — the inspired order prizes creative singularity, the industrial order efficiency and method, the market order price, the domestic order tradition and trust, fame prizes recognition, the civic order the collective good. What counts as a valid test of quality differs by order.

Team fights over AI-assisted work are textbook clashes of orders: 'it lacks soul' (inspired), 'it shipped in half the time' (industrial), 'clients can't tell the difference' (market), 'that's not how we do things here' (domestic). Calling all of this 'taste and judgement' flattens a structured plurality into one word and makes the disputes look irrational. The framework restores tractability: identify which orders are in play, design the tests each would accept, and construct the compromises — which is what effective tool-adoption discussions actually do.

Boltanski, L., & Thévenot, L. (1991). De la justification: Les économies de la grandeur. Paris: Gallimard. English: On Justification: Economies of Worth (C. Porter, Trans.). Princeton, NJ: Princeton University Press, 2006.