HOW AI IS KILLING THE PhD! The Collapse of Intellectual Apprenticeship, the Crisis of Originality, and the Future of Academic Knowledge

The Collapse of Intellectual Apprenticeship, the Crisis of Originality, and the Future of Academic Knowledge

Executive summary

Claims that artificial intelligence is killing the PhD require some qualification. The degree is not disappearing, and the evidence does not establish an inevitable collapse of doctoral education. What AI does challenge is the assumption that a completed dissertation reliably demonstrates the intellectual development of its author. When substantial parts of a thesis can be generated or reorganised with assistance, universities must examine more closely how candidates arrive at their conclusions.

There are good reasons to use these tools. They can support programming, help researchers work across languages and make unfamiliar material easier to approach. Problems arise when assistance allows candidates to bypass decisions they need to learn to make. The educational consequences depend on the task, the candidate's understanding and the quality of supervision.

This essay argues that doctoral assessment should give greater weight to intellectual ownership: the ability to explain a research question, justify a method, evaluate evidence and accept responsibility for conclusions. The reforms proposed here follow from that argument. They should be read as recommendations, alongside a careful distinction between established findings, emerging research and plausible risks.

Introduction: the document and the scholar

A dissertation records the results of research, but much of a researcher's education takes place before those results are written down. A candidate learns that an apparently promising question is too broad, discovers that a familiar method rests on unsuitable assumptions, or returns to a source that complicates an attractive interpretation. Such experiences rarely appear in full in the finished thesis. Nevertheless, they help explain what the doctorate is intended to achieve.

Generative AI changes the relationship between this learning and its written expression. An assistant can suggest a chapter structure, explain a difficult concept or supply code before the candidate has worked through the underlying problem. Sometimes that intervention enables understanding. Sometimes it produces a convincing answer that the candidate cannot adequately evaluate. The distinction matters because both situations may yield equally readable pages.

The debate therefore extends beyond whether students should use ChatGPT. A literature review prepared with assistance may be rigorous if its author has read the sources and understood their disagreements. Conversely, an analysis produced without AI may still be poorly reasoned. Neither the presence of a tool nor its absence settles the academic question.

The concern is whether universities can establish what candidates know and how independently they can use that knowledge. It would be unreasonable to imagine scholarship without assistance. It would be equally unreasonable to award a research qualification without examining the decisions made by the person receiving it.

Part I: the historical bargain of the PhD

1. The PhD as an intellectual apprenticeship

The modern research doctorate is closely associated with the university's commitment to producing knowledge. The nineteenth-century German research-university tradition, often discussed through Wilhelm von Humboldt, offers an important reference point, although doctoral systems have developed differently across countries and disciplines. No single historical account captures all of them.

The idea of apprenticeship remains useful because it directs attention to the development of judgement. Candidates learn through reading, experimentation, analysis and sustained discussion. These activities demand more than the acquisition of information: they require decisions about relevance, reliability and the limits of an explanation.

Polanyi's account of tacit knowledge and Schön's work on reflective practice provide ways of thinking about expertise that cannot be fully expressed as a list of instructions. A candidate may understand why an approach fails only after trying it. Supervision helps make that experience available for reflection, rather than leaving it as an unexamined difficulty.

2. What exactly does a PhD certify?

A doctorate is expected to demonstrate advanced knowledge, methodological competence and an original contribution to a field. These expectations involve the candidate's capacity to work with unfamiliar problems. A successful thesis should offer evidence of that capacity, rather than function as a self-contained object whose quality can be detached from its author.

This is why a fluent chapter has never been sufficient on its own. An examiner may reasonably ask how sources were selected, why a procedure was appropriate or whether another interpretation fits the evidence. AI makes those questions more pressing because presentation can improve without a corresponding improvement in understanding.

The immediate difficulty is one of assessment. Universities need to establish how much confidence a dissertation warrants as evidence of competence. Treating every assisted thesis as invalid would avoid that difficult work rather than resolve it.

3. The academic bargain is changing

The candidate, supervisor and university undertake different responsibilities. Candidates carry out and defend the research; supervisors guide its development; institutions set standards and decide whether those standards have been met. The credibility of the qualification depends on those responsibilities being meaningful in practice.

AI introduces assistance that may be extensive yet largely invisible to a supervisor reading a finished chapter. Clear rules are necessary, but rules cannot reveal everything about learning. Supervisors also need regular opportunities to hear candidates explain decisions while the work is still unsettled.

Counting hours of unaided labour would be a poor substitute. A more useful account of independence asks whether candidates can justify the claims submitted in their names, including those developed with substantial technical support.

Part II: research, apprenticeship and dependence

4. AI is transforming research tasks

Locating literature, learning a technique and preparing an analysis can occupy substantial portions of doctoral study. AI may help with each, but reliability varies with the system, the question and the sources available. Its usefulness has to be judged at the level of the particular research task.

Literature synthesis illustrates the difficulty. A readable summary can conceal an incomplete search or the omission of contradictory findings. A defensible review requires a search strategy, an explanation of selection and engagement with the original material. References must be checked because an apparently plausible citation may not exist or may fail to support the claim attached to it.

Adel and Alani's examination of generative AI and literature synthesis brings these reliability concerns into focus. Efficiency is relevant, especially where researchers face large bodies of material. It remains necessary to ask what was missed and whether the synthesis fairly represents the scholarship.

5. A polished dissertation is not sufficient evidence

Chapter plans, explanatory passages and analytical code can now be produced with considerable assistance. Their quality may tell an examiner relatively little about how the candidate approached the problem. A convincing result still needs an account of the reasoning that produced it.

Consider two hypothetical candidates who submit similar analyses. One has tested competing explanations and revised the method after encountering contradictory evidence. The other has accepted a suggested procedure without understanding its assumptions. Reading the final results alone may not reveal the difference, although it is central to the educational value of the work.

Manual effort does not settle the issue either. Someone who carefully tests and adapts generated code may understand the analysis better than someone who repeats an unsuitable procedure unaided. Assessment should investigate the quality of those decisions, without confusing independence with an obligation to perform every task by hand.

6. The erosion of intellectual apprenticeship

Research requires a tolerance for uncertainty. Candidates must learn when to persist, when to change direction and when the available evidence cannot sustain a conclusion. Assistance that relieves repetitive work can leave more room for this learning. Assistance that supplies every conceptual decision may leave less.

Several ordinary supervisory questions help expose the distinction. Why was this method chosen? Which competing explanation was taken seriously? What finding would require the argument to change? A candidate who cannot engage with such questions may need further support, however accomplished the chapter appears.

This does not justify making doctoral study unnecessarily difficult. Language support, accessibility measures and efficient technical tools can make research more inclusive. The challenge is to retain the intellectual work through which competence develops while removing obstacles that serve no educational purpose.

7. Cognitive offloading and the illusion of understanding

Scholars have always relied on external resources, from notebooks and reference works to computers and colleagues. Generative AI extends that reliance into explanation and interpretation. A response that is coherent and immediately accessible can create a sense of understanding before the researcher has tested its adequacy.

In their 2024 Nature perspective, Lisa Messeri and M. J. Crockett examine illusions of understanding and the risk of scientific monocultures. Their concern is that researchers may increase production while becoming less able to recognise the limits of what they understand. Shared systems may also encourage a narrower range of questions and approaches.

These are conceptual arguments about possible changes in research practice, not proof that all AI users lose expertise. They suggest a useful discipline, however: researchers should be able to reconstruct an explanation, examine its assumptions and identify circumstances in which it would fail.

8. Assistance versus dependence

The boundary between assistance and dependence depends partly on what a task is intended to teach or assess. Editing a sentence may leave the argument intact. Asking a system to formulate the argument can alter the candidate's relationship to the central intellectual contribution.

Generated code presents similar questions. A researcher who understands the procedure, tests its behaviour and checks its output may use it responsibly. Submitting an unexplained analysis leaves a more serious gap. Confidential material introduces further obligations, even when the assistance itself seems minor.

Supervisory discussions should therefore describe actual uses rather than rely on a simple declaration that AI was involved. What did the system contribute? How was the output checked? Which decisions remained with the candidate? Answers to these questions make the scope of assistance open to academic judgement.

Part III: publishing, supervision and originality

9. Academic publishing and industrialised output

The pressure to publish did not begin with generative AI. Systems that reward publication counts can already favour repetitive studies, fragmented findings and modest variations on familiar questions. Lowering the cost of manuscript production may intensify those incentives without improving the research.

In May 2025, Science reported a surge of poor-quality papers exploiting public datasets, including repeated analyses of the National Health and Nutrition Examination Survey. Related reporting in Nature examined formulaic biomedical papers and misleading claims. Together, these accounts draw attention to the relationship between accessible data, automated assistance and weaknesses in editorial scrutiny.

They do not establish AI as the cause of every weak paper. Paper mills, institutional incentives and inadequate review form part of the same problem. For doctoral education, the implication is that publication volume cannot safely stand in for the significance or reliability of a candidate's contribution.

10. Peer review remains a responsibility

Peer review gives other researchers an opportunity to challenge a manuscript's evidence, methods and interpretation. It is fallible, but its value rests on the exercise of informed judgement. A well-written review offers little reassurance if it overlooks the study's decisive limitation.

AI may assist with some review tasks, subject to the journal's policies. Reviewers still have to decide whether a criticism is relevant and whether the evidence supports it. They must also respect confidentiality rather than assume that unpublished manuscripts can be uploaded to an external service.

The same principle applies to a supervisor's response to a chapter: responsibility cannot be delegated simply because assistance produces persuasive language. An assessment needs to withstand scrutiny of its own reasoning.

11. The supervisor's changing role

Rapid access to explanations may change what candidates ask of supervisors. It does not remove the need for someone who understands the discipline, knows the developing project and can challenge an apparently satisfactory answer. Much supervision concerns the significance of a question rather than access to information.

A supervisor can ask why a promising result matters, recognise a recurring weakness in the candidate's approach or help distinguish a temporary setback from a fundamental problem. Such judgements arise within a continuing academic relationship. Automated suggestions do not assume the supervisor's institutional responsibility.

More attention to work in progress would help preserve this relationship. Notes, provisional analyses and unresolved questions give supervisors opportunities to assess development before polished writing conceals the difficulties through which it occurred.

12. The crisis of originality

An unfamiliar idea is not necessarily an original contribution. It may repeat work the researcher has not encountered, raise a question of little consequence or offer an explanation unsupported by evidence. Originality has to be established in relation to a field and its existing scholarship.

AI-generated suggestions make this obligation especially clear. Candidates need to explain where a proposed contribution differs from prior work and why that difference matters. They also need to examine whether the suggestion can be supported, rather than treating its novelty to them as evidence of novelty to the discipline.

Kuhn's account of scientific change offers a broader context for thinking about how research acquires significance within established ways of investigating problems. It does not provide a simple test of originality, but it helps explain why an isolated surprising statement is insufficient.

13. Democratisation versus homogenisation

Researchers working in a second language or with limited technical support may benefit from accessible explanations and writing assistance. Those benefits deserve serious consideration. They should not be confused with the removal of wider inequalities in funding, supervision, laboratories or access to reliable data.

The OECD's Digital Education Outlook 2026 distinguishes improvement in task performance from genuine learning. Although its scope is broader than doctoral education, the distinction is relevant here: finishing an assignment more successfully does not by itself establish what the learner can subsequently do.

There is also a possible tension between wider participation and greater uniformity. Researchers using similar systems may receive similar suggestions. Universities should make room for local knowledge, disciplinary disagreement and approaches that depart from the most readily available answer. Diversity of participation should extend to the questions considered worth asking.

Part IV: economic value and academic authority

14. The changing economic value of the PhD

A doctorate involves a substantial commitment by candidates and institutions. If some familiar research tasks become easier to automate, the case for that investment must give greater attention to the capabilities developed through training. These include identifying important problems, interpreting uncertain evidence and taking responsibility for consequential decisions.

Gross, Shvadron and Zhang's 2026 NBER working paper examined approximately 226,000 United States STEM doctoral graduates' dissertations from 2019 through May 2026. The researchers used an AI-detection service to classify abstracts and checked a sample of complete dissertations. They reported AI-identified writing in 29.4% of abstracts in the portion of the 2026 cohort covered by their data.

The paper also reported associations with subsequent employment. Among United States-origin graduates, AI-classified dissertations were associated with lower academic employment and higher industry employment. Those associations do not establish that AI caused the career differences or that the candidates had weaker research competence.

Several qualifications matter. This is a working paper rather than a peer-reviewed journal article, and a detector's classification is a measurement, not a direct history of the writing process. Nor should a move into industry be treated as an academic or personal failure. The findings raise questions worth investigating; they do not support a simple verdict about the worth of individual graduates.

15. The crisis of epistemic authority

Academic authority rests partly on the expectation that claims have been tested through appropriate methods and exposed to criticism. Generative AI can reproduce the language associated with that authority without supplying the corresponding justification. Readers therefore need ways to assess what supports a statement beyond its confident presentation.

The distinction between retrieving information and understanding its relevance is familiar to researchers. A fact may be accurate yet misapplied; an explanation may work under conditions absent from the case at hand. Expertise includes recognising those limits and knowing when further evidence is needed.

Universities should make these standards visible in assessment. Candidates need to show how evidence supports their conclusions and address credible alternatives. The institutional title carries more weight when the reasoning behind it can be examined.

16. Productivity is not the same as intellectual achievement

Publications, citations and grant income provide information about academic activity, but their interpretation requires care. A publication count cannot establish whether a finding is reliable or a question consequential. Increasing the speed of production may make such measures even less satisfactory as proxies for research quality.

Merton's work on the organisation and norms of science supplies a useful institutional context. The practices through which knowledge is produced are shaped by expectations and rewards. It would be inconsistent to criticise candidates for pursuing volume while leaving the incentives that encourage it unchanged.

Assessment should recognise careful replication, informative negative findings and the correction of errors. Such contributions may be less conspicuous than claims of novelty, yet they can strengthen a field's understanding. Their value should be considered on the evidence rather than inferred from how readily they attract attention.

Part V: reinventing doctoral education

17. Five capabilities worth certifying

The preceding discussion suggests five capabilities that deserve explicit attention in doctoral assessment. They are proposed priorities, not a universal checklist already adopted by universities. Their practical expression would vary with the discipline and the nature of the project.

Problem formulation: candidates should explain why a question warrants investigation, how it relates to existing work and what an answer might contribute. Generating an unanswered question is only the beginning of that justification.

Epistemological judgement: candidates need to distinguish observation from interpretation and recognise the difference between a plausible account and a well-supported conclusion. This includes understanding why an association may be insufficient evidence of causation.

Methodological rigour: the choice of method should fit the question. Candidates should understand how evidence was produced, identify likely sources of bias and explain the consequences of uncertainty for their conclusions.

Intellectual originality: a contribution requires an account of its relationship to prior scholarship. Candidates should be able to defend both its significance and the limits of their claim to novelty.

AI literacy: responsible use requires an understanding of system limitations, verification, confidentiality and disclosure. It also requires the judgement to recognise when a task should not be delegated.

18. Rebuilding doctoral assessment

A dissertation could be assessed alongside selected evidence of the research process. A proportionate portfolio might include changes to the question, reasons for methodological choices and examples of revisions made after encountering contradictory evidence. Its purpose would be to illuminate development without turning every working note into an administrative obligation.

Statements of significant AI use would be more useful if they described contributions relevant to the discipline. Assistance with language, the production of code and suggestions about interpretation involve different questions. Disclosure should make those differences available for assessment rather than operate as an automatic admission of misconduct.

Research records, discussion and oral examination can then provide complementary evidence. None offers a complete view on its own. Together, they may give examiners a stronger basis for judging the candidate's competence than the final document alone.

19. Why AI detection is not enough

AI detectors classify patterns in text. They do not directly establish who chose a method, whether assistance was permitted or whether a candidate understands the analysis. Their outputs have to be interpreted within the limitations of the measurement, particularly when an individual faces a serious allegation.

A method used to estimate patterns across a large population is not automatically sufficient to decide an individual misconduct case. The consequences of error differ, as do the questions being asked. Fair procedures need to recognise that distinction rather than treat a classification as a complete account of authorship.

Clear institutional rules and direct evidence remain essential. Examiners can investigate understanding through research records and academic discussion. A detector may contribute information in a particular setting, but it should not displace the judgement needed to interpret that information.

20. The viva as a test of ownership

The oral examination gives candidates an opportunity to explain decisions that a written thesis may leave implicit. An examiner can pursue a methodological choice, introduce an alternative interpretation or ask how a conclusion would change under different assumptions. The exchange tests the candidate's ability to reason beyond a prepared account.

Discussion of limitations is particularly informative. Recognising where an argument becomes uncertain can demonstrate a more developed understanding than repeating its strongest claims. A candidate should be able to acknowledge a weakness without that acknowledgement being mistaken for a failure of scholarship.

The viva must remain fair, rigorous and accessible, with appropriate accommodations. Turning it into an exercise in catching AI users would narrow its academic purpose. The relevant question is whether the candidate understands and can defend the research, including any assistance used to produce it.

21. Universities must reform their own incentives

Calls for deeper scholarship will have limited effect if assessment continues to reward output volume above everything else. Candidates and supervisors respond to the standards by which their work is judged. Technological assistance may make those responses faster without making the standards better.

Institutions could give greater recognition to transparent methods, reproducibility and well-supported contributions. Supervisors would also need time and training to examine the research process. Adding requirements without providing that support risks producing more paperwork while leaving the underlying assessment unchanged.

Measurement still has a place. The difficulty lies in treating a convenient measure as though it fully captures the purpose of doctoral education. A reform worth pursuing would make the relationship between the measure and that purpose open to scrutiny.

22. A new doctoral contract

A revised doctoral contract would clarify obligations on both sides. Candidates should remain responsible for the accuracy and defensibility of their work. Universities, in turn, should provide meaningful supervision, understandable rules and opportunities to learn responsible ways of using research tools.

Transparency, accountability, demonstrated competence and proportionality offer a basis for such an arrangement. Rules should reflect the task being assessed and the intellectual contribution expected. A language-editing aid and assistance with a central analytical decision should not be treated as though they raise identical issues.

The aim is to adapt research education to changed conditions while retaining its educational substance. Universities cannot restore an earlier environment by declaration. They can decide which forms of understanding must still be demonstrated and organise supervision accordingly.

Part VI: three possible futures

23. Decline, adaptation or intellectual renewal

The declining PhD: universities retain familiar output-based assessment while heavily assisted documents become harder to interpret as evidence of competence. The qualification continues, but uncertainty grows about what it certifies. This is a possible institutional failure, not a prediction established by the research reviewed here.

The AI-assisted PhD: institutions incorporate tools into existing programmes, establish disclosure requirements and adjust supervision. Some work becomes more efficient. Whether learning improves would depend on how candidates use the assistance and how closely assessment examines their understanding.

The intellectual PhD: universities use the disruption to reconsider assessment more thoroughly. The dissertation remains important, but sits alongside evidence of how candidates formulate questions, evaluate methods and respond to criticism. Intellectual development becomes a more explicit object of assessment.

Different institutions may move in different directions, and elements of all three futures may coexist. Their value as scenarios is to identify choices that universities can examine, rather than to imply that one outcome is inevitable.

24. Five questions universities can no longer avoid

1. What does the dissertation establish that cannot be established more directly?
2. Which parts of literature work develop essential understanding, and which can be assisted without losing that learning?
3. What makes an original contribution meaningful rather than merely novel?
4. How should candidates demonstrate ownership of tool-assisted analysis?
5. What evidence shows that a graduate can investigate unfamiliar problems responsibly?

Answering these questions would require attention to disciplinary differences and the practical conditions of supervision. They nevertheless provide a more useful starting point than treating human and machine writing as opposing categories. The educational issue concerns what a candidate has learned to do.

Frequently asked questions

Is AI making the PhD obsolete?

The evidence discussed here does not establish that conclusion. AI challenges some familiar ways of assessing research, especially reliance on the finished document. A doctorate can retain its value if it offers credible evidence of original inquiry, methodological competence and independent judgement.

Can doctoral students use AI responsibly?

Yes, provided the use is permitted under the relevant rules, significant assistance is disclosed where required and outputs are checked. Candidates must understand their methods and accept responsibility for submitted claims. The appropriate boundaries depend on the discipline, task and institutional research policies.

Is AI-assisted writing automatically plagiarism or misconduct?

No. Permitted editing or accessibility support may be legitimate, while prohibited or undisclosed assistance can breach institutional rules. Fabricated citations, inaccurate statements and unacknowledged borrowing remain academic problems regardless of the technology involved. Each case requires attention to what was done and which rules applied.

What is the difference between assistance and intellectual dependence?

Assistance supports research the candidate can understand, evaluate and defend. Dependence becomes a concern when a candidate delegates important decisions without acquiring that understanding. Examining the contribution made by the tool is more informative than asking only whether it was used.

Does the NBER study prove AI harms academic careers?

No. The 2026 working paper reports associations between AI-classified dissertation writing and employment outcomes. It does not establish that AI caused those outcomes or that industry employment indicates failure. Its detector-based measurements also need to be interpreted within their methodological limits.

Can an AI detector prove who wrote a thesis?

A detector's classification cannot reconstruct the full authorship or intellectual history of a thesis. It may provide information that requires further interpretation. Research records, academic discussion and fair assessment procedures are needed to address questions that a text classification alone cannot resolve.

What should universities assess instead of writing quality alone?

Examiners should consider the research question, engagement with prior work, choice of methods, interpretation of evidence and understanding of limitations. A thesis, a proportionate research portfolio and an oral examination can offer complementary evidence of those capabilities.

Conclusion: preserve the scholar, not merely the document

The claim that AI is killing the PhD is too broad to serve as an empirical conclusion. It does, however, draw attention to a genuine problem: universities may find it increasingly difficult to infer a candidate's competence from the quality of a submitted text. That difficulty calls for better assessment, not an assumption that every assisted thesis is compromised.

The doctorate has an educational purpose that extends beyond producing a dissertation. It should prepare someone to recognise a worthwhile problem, investigate it carefully and revise an argument when the evidence demands it. Those abilities develop through practice, criticism and sustained engagement with uncertainty. They need to be visible in the decision to award the degree.

AI can contribute to that development, particularly when assistance makes difficult material more accessible or allows more attention to interpretation. Its use becomes educationally troubling when the candidate can present conclusions without understanding how they were reached. Institutional policies should be judged by how well they address that distinction.

Universities do not need to preserve every inherited practice to preserve the value of doctoral education. They do need credible grounds for saying that a graduate can conduct and defend research. The dissertation remains part of that judgement; the person behind it must remain its central concern.

Sources & References

The bibliography brings together empirical research, conceptual analysis, educational reviews, reporting and historical works. These sources inform different parts of the argument; news accounts and philosophical discussions should not be read as equivalent to empirical findings about doctoral outcomes.

Research and education

1. Messeri, L., & Crockett, M. J. (2024). Artificial intelligence and illusions of understanding in scientific research. Nature, 627, 49–58. A perspective on understanding and scientific monocultures.
2. Gross, D. P., Shvadron, D., & Zhang, H. (2026). How Heavily Are PhD Students Offloading Their Scientific Writing to AI? Evidence from US STEM PhD Dissertations. NBER Working Paper 35859. Observational research, not a peer-reviewed journal article.
3. Basu, M. (2026). PhD students who write dissertations with AI less likely to stay in academia. Nature, 16 October 2026. Science-news reporting on the NBER study.
4. OECD (2026). OECD Digital Education Outlook 2026: Exploring Effective Uses of Generative AI in Education. OECD Publishing, Paris. Broader educational context, not a doctoral-outcomes study.

Academic publishing and research integrity

5. O'Grady, C. (2025). Low-quality papers surge thanks to public data and AI. Science, 388(6749), 807–808. Reporting also published online under the headline “Low-quality papers are surging by exploiting public data sets and AI.”
6. Naddaf, M. (2025). AI linked to explosion of low-quality biomedical research papers. Nature, 641, 1080–1081.
7. Arar, K. H., Özen, H., Polat, G., & Turan, S. (2025). Artificial intelligence, generative artificial intelligence and research integrity: a hybrid systemic review. Smart Learning Environments, 12, 44.
8. Adel, A., & Alani, N. (2025). Can generative AI reliably synthesise literature? Exploring hallucination issues in ChatGPT. AI & Society, 40, 6799–6812.
9. Balalle, H., & Pannilage, S. (2025). Reassessing academic integrity in the age of AI: A systematic literature review on AI and academic integrity. Social Sciences & Humanities Open, 11, 101299.

Historical and philosophical foundations

10. Humboldt, W. von (1810). On the Internal and External Organization of the Higher Scientific Institutions in Berlin. Historical context for the relationship between research, teaching and universities.
11. Kuhn, T. S. (1962). The Structure of Scientific Revolutions. University of Chicago Press. Scientific paradigms and intellectual change.
12. Merton, R. K. (1973). The Sociology of Science. University of Chicago Press. Scientific norms and institutional organisation.
13. Polanyi, M. (1966). The Tacit Dimension. University of Chicago Press. Tacit knowledge and the limits of reducing expertise to explicit information.
14. Schön, D. A. (1983). The Reflective Practitioner: How Professionals Think in Action. Basic Books. Professional judgement and reflective practice.

Explore More from KIMLUD

Continue Reading About Technology

Explore the KIMLUD Technology blog for more technology articles.