AI is Sustainable. Scientific Knowledge Isn't? Reconsidering Human Intellect in the Age of Artificial Intelligence

Abstract

Artificial intelligence (AI) is transforming scientific research by accelerating literature retrieval, enhancing evidence synthesis, generating hypotheses, and supporting scholarly communication. Unlike earlier technological innovations that primarily extended human capabilities, AI increasingly performs cognitive tasks traditionally associated with scientific reasoning and knowledge creation. While these advances improve research efficiency and interdisciplinary discovery, they also raise important questions about the long-term sustainability of scientific knowledge. This perspective explores how the growing integration of AI into scientific workflows may reshape the way knowledge is created, interpreted, and validated. It introduces the concept of recursive epistemic drift, describing the gradual movement of scientific understanding away from direct empirical observation toward successive layers of AI-mediated interpretation. Although AI can identify patterns and synthesize vast amounts of information, it cannot replace the uniquely human capacities for critical inquiry, contextual judgment, conceptual imagination, and reflective skepticism that underpin scientific progress. These epistemic capabilities remain essential for questioning assumptions, recognizing anomalies, and ensuring that scientific knowledge remains self-correcting and continually renewed. From a knowledge management perspective, the paper argues that the future sustainability of scientific knowledge depends not only on advances in AI but also on preserving the human intellectual capabilities that continually reconnect scientific understanding with empirical reality.

Introduction

Artificial intelligence (AI) represents the latest in a long history of technological advances that have transformed scientific inquiry. From the invention of the microscope and telescope to high-throughput sequencing, supercomputing and the internet, each technological innovation has expanded humanity's capacity to observe, measure, analyse and communicate scientific knowledge [11Popper KR. The Logic of Scientific Discovery. London: Routledge; 1959.,22Kuhn TS. The Structure of Scientific Revolutions. 4th ed. Chicago: University of Chicago Press; 2012.]. These technologies accelerated discovery, but they remained fundamentally instruments that extended human capability rather than participants in knowledge creation itself [11Popper KR. The Logic of Scientific Discovery. London: Routledge; 1959.,2020Simon HA. The Sciences of the Artificial. 3rd ed. MIT Press; 1996.]. AI represents a fundamentally different inflection point [66Russell S, Norvig P. Artificial Intelligence: A Modern Approach. 4th ed. Pearson; 2021.,88Kitano H. Nobel Turing Challenge: Creating the engine for scientific discovery. NPJ Systems Biology and Applications. 2021;7:29.,1414Artificial Intelligence in Science: Challenges, Opportunities and the Future of Research. Paris: OECD Publishing; 2023.]. Unlike previous scientific technologies, AI increasingly performs activities traditionally regarded as intellectual work [66Russell S, Norvig P. Artificial Intelligence: A Modern Approach. 4th ed. Pearson; 2021.,1313GPT-4 Technical Report. 2023.,1818Zhang Q, Ding K, Lyv T, et al. Scientific Large Language Models: A Survey on Biological and Chemical Domains. 2024.]. It retrieves and synthesises literature, generates hypotheses, identifies patterns within complex datasets, assists scientific reasoning and contributes to scholarly writing at unprecedented speed and scale [8-118-11Kitano H. Nobel Turing Challenge: Creating the engine for scientific discovery. NPJ Systems Biology and Applications. 2021;7:29.,1313GPT-4 Technical Report. 2023.]. As a result, AI is no longer simply augmenting scientific practice; it is becoming embedded within how scientific knowledge is interpreted, synthesised and communicated [1414Artificial Intelligence in Science: Challenges, Opportunities and the Future of Research. Paris: OECD Publishing; 2023.,2626Dwivedi YK, Hughes L, Baabdullah AM, et al. So what if ChatGPT wrote it? Multidisciplinary perspectives on opportunities, challenges and implications of generative AI. International Journal of Information Management. 2023;71:102642.,2727Nature Editorial. Artificial intelligence can accelerate science if researchers use it responsibly. Nature. 2023;621:451-452.]. This transformation shifts the central challenge facing science [22Kuhn TS. The Structure of Scientific Revolutions. 4th ed. Chicago: University of Chicago Press; 2012.,77Floridi L. The Ethics of Artificial Intelligence: Principles, Challenges, and Opportunities. Oxford University Press; 2023.]. The question is no longer whether technology can improve scientific productivity, as it has throughout history. Rather, it is whether the increasing computational abundance of knowledge fundamentally alters the role of the scientist within the scientific enterprise [77Floridi L. The Ethics of Artificial Intelligence: Principles, Challenges, and Opportunities. Oxford University Press; 2023.,1414Artificial Intelligence in Science: Challenges, Opportunities and the Future of Research. Paris: OECD Publishing; 2023.,2626Dwivedi YK, Hughes L, Baabdullah AM, et al. So what if ChatGPT wrote it? Multidisciplinary perspectives on opportunities, challenges and implications of generative AI. International Journal of Information Management. 2023;71:102642.]. If AI increasingly performs knowledge work, then the defining challenge is no longer technological but epistemological, requiring reconsideration of how scientific knowledge is created, validated, and continually renewed [11Popper KR. The Logic of Scientific Discovery. London: Routledge; 1959.,22Kuhn TS. The Structure of Scientific Revolutions. 4th ed. Chicago: University of Chicago Press; 2012.,77Floridi L. The Ethics of Artificial Intelligence: Principles, Challenges, and Opportunities. Oxford University Press; 2023.,1616Fierro C, Dhar R, Stamatiou F, Garneau N, Søgaard A. Defining Knowledge: Bridging Epistemology and Large Language Models. In: Proceedings of EMNLP 2024; 2024:16096-16111.].

This Perspective argues that the future sustainability of scientific knowledge will depend not upon producing ever greater volumes of computationally synthesised knowledge but upon preserving the uniquely human epistemic capabilities that continually reconnect scientific understanding with observation, experimentation and the realities it seeks to explain [11Popper KR. The Logic of Scientific Discovery. London: Routledge; 1959.,33Polanyi M. The Tacit Dimension. Chicago: University of Chicago Press; 1966.,77Floridi L. The Ethics of Artificial Intelligence: Principles, Challenges, and Opportunities. Oxford University Press; 2023.]. Sustainability refers not to environmental sustainability or computational efficiency but to epistemic sustainability [77Floridi L. The Ethics of Artificial Intelligence: Principles, Challenges, and Opportunities. Oxford University Press; 2023.,1616Fierro C, Dhar R, Stamatiou F, Garneau N, Søgaard A. Defining Knowledge: Bridging Epistemology and Large Language Models. In: Proceedings of EMNLP 2024; 2024:16096-16111.]. It is the capacity of scientific knowledge to remain continually renewed through observation, experimentation, critical scrutiny and conceptual innovation [11Popper KR. The Logic of Scientific Discovery. London: Routledge; 1959.,22Kuhn TS. The Structure of Scientific Revolutions. 4th ed. Chicago: University of Chicago Press; 2012.,2323Chalmers AF. What Is This Thing Called Science? 4th ed. Hackett Publishing; 2013.]. Scientific knowledge is sustained not through the accumulation or synthesis of information alone but through the continual testing, refinement and reconstruction of ideas in response to new evidence [11Popper KR. The Logic of Scientific Discovery. London: Routledge; 1959.,22Kuhn TS. The Structure of Scientific Revolutions. 4th ed. Chicago: University of Chicago Press; 2012.]. This process depends upon uniquely human epistemic capabilities that include epistemic curiosity, scepticism, conceptual imagination, systems thinking and reflective judgement [33Polanyi M. The Tacit Dimension. Chicago: University of Chicago Press; 1966.,2020Simon HA. The Sciences of the Artificial. 3rd ed. MIT Press; 1996.,2323Chalmers AF. What Is This Thing Called Science? 4th ed. Hackett Publishing; 2013.]. These capabilities enable researchers to question prevailing assumptions, recognise anomalies and generate new scientific understanding [22Kuhn TS. The Structure of Scientific Revolutions. 4th ed. Chicago: University of Chicago Press; 2012.,2020Simon HA. The Sciences of the Artificial. 3rd ed. MIT Press; 1996.]. They continue to anchor scientific inquiry in observable reality and ensure that knowledge remains self-correcting, continually renewable and epistemically sustainable [11Popper KR. The Logic of Scientific Discovery. London: Routledge; 1959.-33Polanyi M. The Tacit Dimension. Chicago: University of Chicago Press; 1966.].

Artificial intelligence (AI) is widely recognised as one of the most significant technological advancements in the history of scientific research [66Russell S, Norvig P. Artificial Intelligence: A Modern Approach. 4th ed. Pearson; 2021.,88Kitano H. Nobel Turing Challenge: Creating the engine for scientific discovery. NPJ Systems Biology and Applications. 2021;7:29.,1414Artificial Intelligence in Science: Challenges, Opportunities and the Future of Research. Paris: OECD Publishing; 2023.]. It has accelerated literature retrieval, enhanced systematic evidence synthesis, generated novel hypotheses from increasingly complex datasets, improved research efficiency and strengthened reproducibility through automated analytical workflows [8-118-11Kitano H. Nobel Turing Challenge: Creating the engine for scientific discovery. NPJ Systems Biology and Applications. 2021;7:29.,1818Zhang Q, Ding K, Lyv T, et al. Scientific Large Language Models: A Survey on Biological and Chemical Domains. 2024.]. These advances have substantially expanded researchers' capacity to interrogate complex scientific problems and created unprecedented opportunities for interdisciplinary discovery [88Kitano H. Nobel Turing Challenge: Creating the engine for scientific discovery. NPJ Systems Biology and Applications. 2021;7:29.,1010Jumper J, Evans R, Pritzel A, et al. Highly accurate protein structure prediction with AlphaFold. Nature. 2021;596:583-589.,1111Stokes JM, Yang K, Swanson K, et al. A deep learning approach to antibiotic discovery. Cell. 2020;180(4):688-702.e13.]. The argument presented here is therefore not that AI threatens science. On the contrary, AI can identify patterns and associations that may escape human attention and substantially augment scientific enquiry [99Gil Y, Greaves M, Hendler J, Hirsh H. Amplify scientific discovery with artificial intelligence. Science. 2014;346(6206):171-172.-1111Stokes JM, Yang K, Swanson K, et al. A deep learning approach to antibiotic discovery. Cell. 2020;180(4):688-702.e13.]. However, it cannot independently determine when prevailing assumptions should be abandoned, whether unexpected observations warrant the development of new conceptual frameworks, or whether an apparent anomaly signals the beginning of a scientific transformation [22Kuhn TS. The Structure of Scientific Revolutions. 4th ed. Chicago: University of Chicago Press; 2012.,33Polanyi M. The Tacit Dimension. Chicago: University of Chicago Press; 1966.,77Floridi L. The Ethics of Artificial Intelligence: Principles, Challenges, and Opportunities. Oxford University Press; 2023.,1616Fierro C, Dhar R, Stamatiou F, Garneau N, Søgaard A. Defining Knowledge: Bridging Epistemology and Large Language Models. In: Proceedings of EMNLP 2024; 2024:16096-16111.]. These remain fundamentally human epistemic responsibilities because they require contextual judgement, conceptual imagination, empirical scepticism and the capacity to continually reconnect scientific understanding with observed reality [33Polanyi M. The Tacit Dimension. Chicago: University of Chicago Press; 1966.,77Floridi L. The Ethics of Artificial Intelligence: Principles, Challenges, and Opportunities. Oxford University Press; 2023.,2020Simon HA. The Sciences of the Artificial. 3rd ed. MIT Press; 1996.,2323Chalmers AF. What Is This Thing Called Science? 4th ed. Hackett Publishing; 2013.]. AI therefore does not diminish the importance of human epistemic capabilities; rather, it reveals them as the defining resource upon which the creation, renewal and sustainability of scientific knowledge ultimately depend [77Floridi L. The Ethics of Artificial Intelligence: Principles, Challenges, and Opportunities. Oxford University Press; 2023.,1414Artificial Intelligence in Science: Challenges, Opportunities and the Future of Research. Paris: OECD Publishing; 2023.,2626Dwivedi YK, Hughes L, Baabdullah AM, et al. So what if ChatGPT wrote it? Multidisciplinary perspectives on opportunities, challenges and implications of generative AI. International Journal of Information Management. 2023;71:102642.].

As AI-generated reviews, evidence syntheses and scientific interpretations become increasingly embedded within the scholarly literature, subsequent AI systems will progressively learn from knowledge that has already been mediated through earlier computational interpretations [1313GPT-4 Technical Report. 2023.,1414Artificial Intelligence in Science: Challenges, Opportunities and the Future of Research. Paris: OECD Publishing; 2023.,1818Zhang Q, Ding K, Lyv T, et al. Scientific Large Language Models: A Survey on Biological and Chemical Domains. 2024.,2525Bender EM, Gebru T, McMillan-Major A, Shmitchell S. On the dangers of stochastic parrots: Can language models be too big? Proceedings of FAccT. 2021:610-623.]. I define this phenomenon as recursive epistemic drift. It is the progressive displacement of scientific understanding from continual empirical engagement towards successive layers of computational interpretation. Rather than being continually renewed through observation, experimentation and critical scholarly scrutiny, knowledge increasingly evolves through the recursive synthesis of prior computationally mediated interpretations [11Popper KR. The Logic of Scientific Discovery. London: Routledge; 1959.,33Polanyi M. The Tacit Dimension. Chicago: University of Chicago Press; 1966.,1616Fierro C, Dhar R, Stamatiou F, Garneau N, Søgaard A. Defining Knowledge: Bridging Epistemology and Large Language Models. In: Proceedings of EMNLP 2024; 2024:16096-16111.]. The concern is not that AI necessarily generates inaccurate science. Rather, successive layers of computational abstraction risk distancing scientific understanding from the empirical observations, contextual judgement and tacit knowledge that originally established its validity [33Polanyi M. The Tacit Dimension. Chicago: University of Chicago Press; 1966.,2525Bender EM, Gebru T, McMillan-Major A, Shmitchell S. On the dangers of stochastic parrots: Can language models be too big? Proceedings of FAccT. 2021:610-623.,2626Dwivedi YK, Hughes L, Baabdullah AM, et al. So what if ChatGPT wrote it? Multidisciplinary perspectives on opportunities, challenges and implications of generative AI. International Journal of Information Management. 2023;71:102642.]. Consequently, the epistemic richness of scientific enquiry may become compressed into statistically derived representations of evidence rather than the reflective, iterative and empirically grounded processes through which scientific discovery occurs [33Polanyi M. The Tacit Dimension. Chicago: University of Chicago Press; 1966.,77Floridi L. The Ethics of Artificial Intelligence: Principles, Challenges, and Opportunities. Oxford University Press; 2023.,1616Fierro C, Dhar R, Stamatiou F, Garneau N, Søgaard A. Defining Knowledge: Bridging Epistemology and Large Language Models. In: Proceedings of EMNLP 2024; 2024:16096-16111.].

This recursive epistemic drift may already be emerging across multiple scientific disciplines. For instance, in biomedical research, AI-assisted systematic reviews may increasingly synthesise conclusions drawn from earlier AI-generated reviews rather than returning to primary clinical studies [1010Jumper J, Evans R, Pritzel A, et al. Highly accurate protein structure prediction with AlphaFold. Nature. 2021;596:583-589.,1111Stokes JM, Yang K, Swanson K, et al. A deep learning approach to antibiotic discovery. Cell. 2020;180(4):688-702.e13.,1818Zhang Q, Ding K, Lyv T, et al. Scientific Large Language Models: A Survey on Biological and Chemical Domains. 2024.]. Similarly, in engineering research, AI-generated design recommendations may increasingly be derived from prior simulation outputs rather than validated through physical testing [66Russell S, Norvig P. Artificial Intelligence: A Modern Approach. 4th ed. Pearson; 2021.,1818Zhang Q, Ding K, Lyv T, et al. Scientific Large Language Models: A Survey on Biological and Chemical Domains. 2024.]. In both cases, scientific knowledge may remain internally coherent while progressively drifting from the empirical reality that originally informed it [11Popper KR. The Logic of Scientific Discovery. London: Routledge; 1959.-33Polanyi M. The Tacit Dimension. Chicago: University of Chicago Press; 1966.]. The greater risk, therefore, is not the automation of scientific writing or knowledge synthesis itself, but the gradual displacement of the uniquely human epistemic capabilities through which scientific knowledge is questioned, interpreted, challenged and ultimately renewed through empirical enquiry [33Polanyi M. The Tacit Dimension. Chicago: University of Chicago Press; 1966.,77Floridi L. The Ethics of Artificial Intelligence: Principles, Challenges, and Opportunities. Oxford University Press; 2023.,2525Bender EM, Gebru T, McMillan-Major A, Shmitchell S. On the dangers of stochastic parrots: Can language models be too big? Proceedings of FAccT. 2021:610-623.].

Discussion

Viewed through a Knowledge Management lens, this represents a fundamental transition in the scientific enterprise rather than simply another technological advance [44Nonaka I, Takeuchi H. The Knowledge-Creating Company. New York: Oxford University Press; 1995.,55Davenport TH, Prusak L. Working Knowledge: How Organizations Manage What They Know. Boston: Harvard Business School Press; 1998.]. If Knowledge Management has traditionally focused on the creation, transfer and application of knowledge, the AI era requires equal attention to its continual epistemic renewal [44Nonaka I, Takeuchi H. The Knowledge-Creating Company. New York: Oxford University Press; 1995.,55Davenport TH, Prusak L. Working Knowledge: How Organizations Manage What They Know. Boston: Harvard Business School Press; 1998.,1414Artificial Intelligence in Science: Challenges, Opportunities and the Future of Research. Paris: OECD Publishing; 2023.]. Preventing recursive epistemic drift therefore depends less on producing greater volumes of information than on cultivating the human epistemic capabilities required to question assumptions, recognise anomalies, exercise contextual judgement and integrate tacit and explicit knowledge [33Polanyi M. The Tacit Dimension. Chicago: University of Chicago Press; 1966.-55Davenport TH, Prusak L. Working Knowledge: How Organizations Manage What They Know. Boston: Harvard Business School Press; 1998.]. Although AI can accelerate discovery, enhance synthesis and expand analytical capability [8-108-10Kitano H. Nobel Turing Challenge: Creating the engine for scientific discovery. NPJ Systems Biology and Applications. 2021;7:29.,1313GPT-4 Technical Report. 2023.], the enduring strength of science lies in its capacity for continual questioning, experimentation, falsification, conceptual refinement and scholarly critique [11Popper KR. The Logic of Scientific Discovery. London: Routledge; 1959.,22Kuhn TS. The Structure of Scientific Revolutions. 4th ed. Chicago: University of Chicago Press; 2012.,2323Chalmers AF. What Is This Thing Called Science? 4th ed. Hackett Publishing; 2013.]. As AI increasingly performs knowledge work, these uniquely human epistemic capabilities become the scarce resource upon which the long-term sustainability of scientific knowledge depends [33Polanyi M. The Tacit Dimension. Chicago: University of Chicago Press; 1966.,77Floridi L. The Ethics of Artificial Intelligence: Principles, Challenges, and Opportunities. Oxford University Press; 2023.,1414Artificial Intelligence in Science: Challenges, Opportunities and the Future of Research. Paris: OECD Publishing; 2023.,2626Dwivedi YK, Hughes L, Baabdullah AM, et al. So what if ChatGPT wrote it? Multidisciplinary perspectives on opportunities, challenges and implications of generative AI. International Journal of Information Management. 2023;71:102642.].

AI will continue to transform the way scientific knowledge is generated, synthesised and disseminated. However, the long-term sustainability of scientific knowledge will depend not on the continued advancement of AI alone but on preserving the uniquely human epistemic capabilities that continually reconnect scientific understanding with observation, critical inquiry and scientific reasoning. The larger question confronting science is therefore not what AI can produce, but which human epistemic capabilities must be preserved if scientific knowledge is to remain self-correcting, continually renewable and epistemically sustainable.

Conclusion

Artificial intelligence is transforming scientific research by enhancing knowledge generation, synthesis, and dissemination. However, the long-term sustainability of scientific knowledge depends not only on technological advancement but also on preserving the uniquely human epistemic capabilities that underpin scientific inquiry. As AI becomes increasingly embedded within research, maintaining critical judgement, conceptual imagination, empirical validation, and reflective inquiry will be essential to prevent recursive epistemic drift and ensure that scientific knowledge remains self-correcting and continually renewable. The future of science will therefore depend on a balanced partnership in which AI augments, rather than replaces, human intellectual responsibility and scientific reasoning.

References

1. Popper KR. The Logic of Scientific Discovery. London: Routledge; 1959.

2. Kuhn TS. The Structure of Scientific Revolutions. 4th ed. Chicago: University of Chicago Press; 2012.

3. Polanyi M. The Tacit Dimension. Chicago: University of Chicago Press; 1966.

4. Nonaka I, Takeuchi H. The Knowledge-Creating Company. New York: Oxford University Press; 1995.

5. Davenport TH, Prusak L. Working Knowledge: How Organizations Manage What They Know. Boston: Harvard Business School Press; 1998.

6. Russell S, Norvig P. Artificial Intelligence: A Modern Approach. 4th ed. Pearson; 2021.

7. Floridi L. The Ethics of Artificial Intelligence: Principles, Challenges, and Opportunities. Oxford University Press; 2023.

8. Kitano H. Nobel Turing Challenge: Creating the engine for scientific discovery. NPJ Systems Biology and Applications. 2021;7:29.

9. Gil Y, Greaves M, Hendler J, Hirsh H. Amplify scientific discovery with artificial intelligence. Science. 2014;346(6206):171-172.

10. Jumper J, Evans R, Pritzel A, et al. Highly accurate protein structure prediction with AlphaFold. Nature. 2021;596:583-589.

11. Stokes JM, Yang K, Swanson K, et al. A deep learning approach to antibiotic discovery. Cell. 2020;180(4):688-702.e13.

12. Bommasani R, Hudson DA, Adeli E, et al. On the Opportunities and Risks of Foundation Models. Stanford Center for Research on Foundation Models; 2021.

13. GPT-4 Technical Report. 2023.

14. Artificial Intelligence in Science: Challenges, Opportunities and the Future of Research. Paris: OECD Publishing; 2023.

15. Forbus KD. The importance of knowledge bases for artificial intelligence in science. In: Artificial Intelligence in Science. OECD Publishing; 2023.

16. Fierro C, Dhar R, Stamatiou F, Garneau N, Søgaard A. Defining Knowledge: Bridging Epistemology and Large Language Models. In: Proceedings of EMNLP 2024; 2024:16096-16111.

17. Kim M, Thorne J. Epistemology of Language Models: Do Language Models Have Holistic Knowledge? 2024.

18. Zhang Q, Ding K, Lyv T, et al. Scientific Large Language Models: A Survey on Biological and Chemical Domains. 2024.

19. Collins H, Thorne S. Large language models and scientific discourse: Where's the intelligence? Synthese. 2026;207:160.

20. Simon HA. The Sciences of the Artificial. 3rd ed. MIT Press; 1996.

21. Latour B, Woolgar S. Laboratory Life: The Construction of Scientific Facts. Princeton University Press; 1986.

22. Merton RK. The Sociology of Science: Theoretical and Empirical Investigations. University of Chicago Press; 1973.

23. Chalmers AF. What Is This Thing Called Science? 4th ed. Hackett Publishing; 2013.

24. Searle JR. Minds, brains, and programs. Behavioral and Brain Sciences. 1980;3(3):417-457.

25. Bender EM, Gebru T, McMillan-Major A, Shmitchell S. On the dangers of stochastic parrots: Can language models be too big? Proceedings of FAccT. 2021:610-623.

26. Dwivedi YK, Hughes L, Baabdullah AM, et al. So what if ChatGPT wrote it? Multidisciplinary perspectives on opportunities, challenges and implications of generative AI. International Journal of Information Management. 2023;71:102642.

27. Nature Editorial. Artificial intelligence can accelerate science if researchers use it responsibly. Nature. 2023;621:451-452.

28. Recommendation on the Ethics of Artificial Intelligence. Paris: UNESCO; 2021.

29. World Economic Forum. The Future of AI for Science. Geneva: World Economic Forum; 2024.

30. National Academies of Sciences, Engineering, and Medicine. Fostering Integrity in Research. Washington, DC: National Academies Press; 2017.

記事について

この記事を引用する

Kannan S. AI is Sustainable. Scientific Knowledge Isn't? Reconsidering Human Intellect in the Age of Artificial Intelligence. IgMin Res. July 31, 2026; 4(7): 306-308. IgMin ID: igmin354; DOI:10.61927/igmin354; Available at: igmin.link/p354

• DOI10.61927/igmin354

13 Jul, 2026

受け取った

29 Jul, 2026

受け入れられた

31 Jul, 2026

発行された

この記事を共有する

次のリンクを共有した人は、このコンテンツを読むことができます:

共有リンクを取得する

トピックス

Artificial Intelligence