Beyond the Single Discipline: Why Multidisciplinarity Has Become a Structural Necessity in the Age of Artificial Intelligence

Abstract

This paper argues that multidisciplinarity has shifted from being an intellectual preference to becoming a structural condition for producing significant knowledge in the age of artificial intelligence. Drawing on three independent bodies of evidence — the disciplinary composition of landmark AI achievements, large-scale labour-market survey data, and bibliometric research on collaboration outcomes — the paper advances a two-part claim. First, artificial intelligence is not an autonomous discipline but a convergence point, and its most consequential results have emerged where computational methods met domain-specific knowledge that computation alone could not supply. Second, and equally important, disciplinary diversity does not by itself generate impact: bibliometric evidence from approximately 15,000 AI–COVID-19 papers demonstrates that the composition of a research team predicts impact far less reliably than the depth of knowledge integration achieved within the work. The paper concludes that the operative distinction is not between narrow and broad training, but between superficial juxtaposition of fields and genuine epistemic integration — a distinction with direct consequences for how universities design curricula and how students structure their own intellectual formation.

Keywords:

artificial intelligence, multidisciplinarity, interdisciplinary research, knowledge integration, higher education, skills economy, AlphaFold

1. Introduction

In October 2024, the Royal Swedish Academy of Sciences awarded the Nobel Prize in Chemistry to Demis Hassabis and John Jumper of Google DeepMind, alongside David Baker of the University of Washington, for work on protein structure prediction and computational protein design (“Chemistry Nobel Goes to Developers of AlphaFold,” 2024). Two of the three laureates were, by training and professional identity, computer scientists. The prize was awarded in chemistry. This was the first occasion on which a scientific breakthrough enabled by artificial intelligence received a Nobel Prize, and commentators noted at the time that it was unlikely to be the last (“Chemistry Nobel Goes to Developers of AlphaFold,” 2024).

The episode is worth pausing on, because the disciplinary confusion it produced is precisely the point. AlphaFold could not be located cleanly within chemistry, biology, or computer science, because it was not produced by any of them acting alone. This paper takes that observation as its starting point and develops it into a general argument: in the era of artificial intelligence, the boundaries that organise academic disciplines have become poor predictors of where significant knowledge is produced.

The argument proceeds in four movements. Section 2 establishes that artificial intelligence is constitutively a convergence field rather than an autonomous discipline. Section 3 examines what this convergence means economically, drawing on large-scale employer survey data concerning the reconfiguration of skill demand. Section 4 confronts a serious counterargument — that interdisciplinary collaboration frequently underperforms — and argues that this evidence, correctly interpreted, refines rather than refutes the thesis. Section 5 considers the institutional implications for higher education.

2. Artificial Intelligence as a Convergence Field

A discipline, in the conventional sense, possesses its own objects of study, its own methods, and its own criteria for what counts as a valid result. By this standard, artificial intelligence has never been a discipline in the ordinary sense. Its mathematical foundations are drawn from probability theory and optimisation; its architectural metaphors from neuroscience; its evaluation methods from statistics; its most productive problem domains from linguistics, biology, and vision science. Artificial intelligence is better understood as a set of computational methods that acquire meaning only in contact with a substantive domain.

Kusters et al. (2020), writing in Frontiers in Big Data, formalise this observation. They argue that the relationship between artificial intelligence and interdisciplinary research must be understood as bidirectional. The more familiar direction — applying AI to quantitative science, healthcare, biology, economics, and finance — has been pursued extensively, in their assessment perhaps even excessively. The less examined direction, and the one they consider decisive for the field’s future, runs the other way: applying knowledge from other fields to the development of AI itself. Their argument implies that the frontier of AI research is not primarily an engineering frontier but an epistemic one, located at points where computational methods encounter phenomena they were not designed to describe.

Dignum et al. (2023) extend this reasoning from the production of AI to the understanding of its consequences. As the impact of AI across scientific fields intensifies, they argue, interdisciplinary knowledge becomes essential for understanding what technology does to society. This is not a call for ethical supervision added after the technical work is complete. It is a claim about the object of study: the interdependencies between technology and how humans relate, form institutions, and structure decision-making cannot be observed from within computer science, because computer science does not possess the conceptual vocabulary in which such questions are posed.

The AlphaFold case demonstrates both directions concretely. For roughly six decades, protein structures were determined experimentally, through X-ray crystallography and cryo-electron microscopy, yielding structures for over 190,000 proteins (EMBL, 2024). That accumulated body of work was not incidental to AlphaFold’s success — it was its precondition. As John Jumper of DeepMind observed, the careful curation of these large data resources, representing the collective output of an entire subfield of biology, is precisely what enabled the machine learning models to generalise across such an enormous range of proteins (EMBL, 2024). The system subsequently produced predictions for over 200 million protein structures from more than a million organisms (EMBL, 2024). The asymmetry is instructive: sixty years of experimental biology produced the training substrate; deep learning produced the generalisation. Neither field contained both halves.

3. The Economic Signature of Convergence

If the convergence described above were confined to elite research institutions, its practical significance would be limited. The evidence suggests otherwise. The World Economic Forum’s Future of Jobs Report 2025 provides the most systematic available measurement, drawing on responses from more than 1,000 leading global employers who collectively represent over 14 million workers across 22 industry clusters and 55 economies (World Economic Forum, 2025).

Three findings from that survey bear directly on this paper’s argument. First, employers expect that 39% of workers’ core skills will change by 2030 (World Economic Forum, 2025). Notably, this figure represents a decline from 44% reported in 2023, which the report attributes partly to expanded institutional investment in continuous learning and reskilling — an important qualification, since it suggests that the pace of disruption, while high, is being partially absorbed rather than accelerating without limit.

Second, technological skills, with AI and big data at the top of the list, are projected to grow in importance more rapidly than any other skill category over the coming five years (World Economic Forum, 2025). Across the ten leading industries, more than 90% of respondents expect the use of AI and big data to increase, with the lowest figures appearing in agriculture, forestry and fishing at 70% and in accommodation, food and leisure at 69% (World Economic Forum, 2025). Even the least affected sectors report supermajority expectations of change.

Third, and most significant for the present argument, the skills rising alongside AI are not exclusively technical. Creative thinking, resilience, flexibility and agility, curiosity and lifelong learning all appear among the fastest-growing competencies (World Economic Forum, 2025). The same survey reports declining projected demand for reading, writing and mathematics as discrete skills, alongside manual dexterity and attention to detail (World Economic Forum, 2025). Read together, these two movements describe a labour market in which the capacity to acquire and recombine knowledge is appreciating in value while the possession of any fixed body of knowledge is depreciating. The structural transition underlying these figures involves roughly 92 million jobs displaced by 2030 against approximately 170 million created, yielding a net increase of 78 million positions (World Economic Forum, 2025). The critical observation is that displaced and created roles are not interchangeable. The new positions cluster disproportionately at intersections — between computation and medicine, computation and agriculture, computation and finance — where the required competence is not mastery of either field alone but the ability to operate across their boundary.

4. The Central Qualification: Composition Is Not Integration

An argument for multidisciplinarity that stopped here would be seriously incomplete, because the strongest available evidence against naive interdisciplinarity is also the most methodologically rigorous. Abbonato, Bianchini, Gargiulo and Venturini (2024), publishing in Quantitative Science Studies, examined approximately 15,000 papers at the intersection of artificial intelligence and COVID-19 — a domain in which interdisciplinary collaboration was urgently needed, heavily funded, and institutionally encouraged. Their finding was that collaborations between medical professionals and AI specialists largely produced publications with low visibility and low impact. The detail of their results is more instructive than the headline. Teams containing collaborators experienced in AI showed no significant effect on impact. Teams with a high proportion of researchers holding established AI publication records received, all else equal, fewer citations, achieved less online visibility, and proved less able to reach distant disciplines. Only the presence of a top-tier AI researcher produced a positive effect on citations, and even that effect was weak (Abbonato et al., 2024).

Their conclusion is precise and should be quoted in substance rather than paraphrased loosely: impactful research depends less on the overall interdisciplinarity of author teams than on the diversity of knowledge those teams actually harness in their research (Abbonato et al., 2024). The distinction is between the demographic composition of a collaboration and the epistemic content of its output. A team may be maximally diverse in its members’ affiliations while producing work that draws on a narrow and conventional body of prior knowledge.

This finding does not weaken the argument of Sections 2 and 3; it specifies it. What distinguishes AlphaFold from the median AI–COVID-19 collaboration is not that one crossed disciplinary boundaries and the other did not — both did. The difference is the depth at which the crossing occurred. AlphaFold required its developers to absorb the structural logic of protein folding sufficiently to encode it architecturally, and required structural biologists to have organised their accumulated data in forms a learning system could exploit. This is integration. By contrast, a collaboration in which clinicians supply a dataset and computer scientists apply a standard classifier to it is a division of labour across fields, not an integration of them. The evidence suggests that only the former reliably generates impact. The practical inference for a student is direct and somewhat demanding. Accumulating exposure to multiple fields is not the objective, and may produce nothing of value. The objective is to understand a second field deeply enough that its questions, methods, and standards of evidence become genuinely available for use — deeply enough to see what a specialist in one’s primary field would not think to ask.

5. Institutional Implications and the Incentive Problem

If integration rather than exposure is what generates value, the question becomes whether universities are structured to produce it. Sejdiu, Sejdiu, Bllaca and Alhasani (2025) examine what they characterise as a quiet transformation of higher education under generative AI, arguing that its implications for teaching, learning, assessment and institutional policy require an explicitly interdisciplinary lens, and that the traditional conception of digital literacy must be redefined as algorithmic literacy — the capacity to reason about systems whose behaviour is statistically rather than deterministically specified.

Yet a structural obstacle stands in the way, and Dignum et al. (2023) name it directly: research beyond disciplinary boundaries, though essential for addressing complex societal problems and generating positive impact, is notoriously difficult to evaluate and frequently goes unrecognised within existing academic career progression systems. This produces a durable misalignment. Institutions publicly celebrate interdisciplinary ambition while their promotion committees, departmental structures, and journal hierarchies continue to reward depth within a single field. A researcher who invests years in acquiring genuine competence in a second discipline pays an immediate career cost for a benefit that existing evaluation instruments are poorly designed to detect.

Two considerations qualify this pessimism. The first is that the evaluation problem is partly a measurement problem, and measurement instruments change — the bibliometric literature examined in Section 4 is itself part of an emerging effort to measure integration rather than mere composition. The second is that recognition at the highest level has already shifted. AlphaFold has been used by more than two million researchers across enzyme design, drug discovery, and related domains (AI Magazine, 2024), and the Nobel Committee’s 2024 decision represents an institutional acknowledgement from one of the most conservative bodies in science. Formal incentive structures typically follow such recognitions rather than anticipating them.

6. Conclusion

This paper has argued that multidisciplinarity in the age of artificial intelligence is a structural condition rather than an intellectual preference, and has attempted to state that claim with the precision the evidence warrants. Artificial intelligence is constitutively a convergence field, dependent on domain knowledge it cannot itself generate (Kusters et al., 2020; EMBL, 2024). The labour market is reorganising around this convergence, with 39% of core skills expected to change by 2030 and the fastest-growing competencies including both technical fluency and the adaptive capacities that permit movement between domains (World Economic Forum, 2025).

The decisive qualification is that crossing disciplinary boundaries guarantees nothing. The bibliometric evidence establishes that impact tracks the diversity of knowledge actually integrated into research rather than the diversity of those conducting it (Abbonato et al., 2024). The relevant distinction is therefore not between narrow and broad training but between superficial juxtaposition and genuine epistemic integration — a considerably more demanding standard than the word “interdisciplinary” usually implies.

For students entering higher education at the moment when artificial intelligence is simultaneously reshaping scientific practice, labour markets, and the university itself, the implication is neither that specialisation is obsolete nor that breadth is automatically valuable. It is that the scarce and appreciating capacity is the ability to hold two fields at sufficient depth that each can genuinely inform the other — and that this capacity, unlike the accumulation of credentials, cannot be acquired quickly.

References

Abbonato, D., Bianchini, S., Gargiulo, F., & Venturini, T. (2024). Interdisciplinary research in artificial intelligence: Lessons from COVID-19. Quantitative Science Studies, 5(4), 922–935. https://doi.org/10.1162/qss_a_00329

AI Magazine. (2024, October 10). Alphafold 2: The AI system that won Google a Nobel Prize. https://aimagazine.com/articles/alphafold-2-the-ai-system-that-won-google-a-nobel-prize

Chemistry Nobel goes to developers of AlphaFold AI that predicts protein structures. (2024, October 9). Nature. https://www.nature.com/articles/d41586-024-03214-7

Dignum, V., Casey, D., Cerratto-Pargman, T., Dignum, F., Fantasia, V., Formark, B., Hammarfelt, B., Holmberg, G., Holzapfel, A., Larsson, S., Lagerkvist, A., Lakemond, N., Lindgren, H., Lorig, F., Marusic, A., Rahm, L., Razmetaeva, Y., Sikström, S., Tatar, K., & Tucker, J. (2023). On the importance of AI research beyond disciplines. arXiv. https://arxiv.org/abs/2302.06655

EMBL. (2024, October 18). Computational protein design and protein structure prediction win Nobel Prize in Chemistry. https://www.embl.org/news/science-technology/alphafold-wins-nobel-prize-chemistry-2024/

Kusters, R., Misevic, D., Berry, H., Cully, A., Le Cunff, Y., Dandoy, L., Díaz-Rodríguez, N., Ficher, M., Grizou, J., Othmani, A., Palpanas, T., Komorowski, M., Loiseau, P., Moulin Frier, C., Nanini, S., Quercia, D., Sebag, M., Soulié Fogelman, F., Taleb, S., Tupikina, L., Sahu, V., Vie, J.-J., & Wehbi, F. (2020). Interdisciplinary research in artificial intelligence: Challenges and opportunities. Frontiers in Big Data, 3, Article 577974. https://doi.org/10.3389/fdata.2020.577974

Sejdiu, N. P., Sejdiu, S., Bllaca, N., & Alhasani, M. (2025). The quiet transformation of higher education in the AI era. Open Research Europe, 5, Article 249. https://doi.org/10.12688/openreseurope.20715.1

World Economic Forum. (2025). The future of jobs report 2025. https://www.weforum.org/publications/the-future-of-jobs-report-2025/

test