cyberspaceФилософские проблемы информационных технологий и киберпространстваPhilosophical Problems of IT & Cyberspace (PhilIT&C)2305-3763Пятигорский государственный университет10.17726/philIT.2024.1.6cyberspace-311Research ArticleСтатьиЧем является научное знание, произведенное методами Больших языковых моделей ?What is scientific knowledge produced by Large Language Models?БарышниковП. Н.BaryshnikovP. N.

Барышников Павел Николаевич - доктор философских наук, доцент, профессор кафедры исторических и социально-философских дисциплин, востоковедения и теологии.

Пятигорск

Pavel N. Baryshnikov - Doctor of science (in Philosophy), Assistant professor, Professor of the Department of Historical and Socio-Philosophical Disciplines, Oriental Studies and Theology, Pyatigorsk State University.

Pyatigorsk

pnbaryshnikov@pgu.ruПятигорский государственный университетРоссияPyatigorsk State UniversityRussian Federation2024120720240189103Copyright © Барышников П.Н., 20242024Барышников П.Н.Baryshnikov P.N.Данная работа распространяется под лицензией Creative Commons Attribution 4.0.This work is licensed under a Creative Commons Attribution 4.0 License.https://cyberspace.pgu.ru/jour/article/view/311

В данной статье исследуется природа научного знания, созданного с помощью больших языковых моделей (LLM), и оценивается их влияние на научные открытия и философию науки. LLM, такие как GPT‑4 и прочие версии генеративных предобученных трансформеров, представляют собой продвинутые алгоритмы глубокого обучения, способные выполнять различные задачи обработки естественного языка, включая генерацию текста, перевод и анализ данных. Цель исследования заключается в изучении того, как эти технологии влияют на процесс научных исследований, ставя под вопрос квалификацию и достоверность научных открытий, созданных с участием ИИ. Методология включает всесторонний обзор существующей литературы по применению LLM в различных научных областях, а также анализ их этических последствий. Основные выводы подчеркивают преимущества LLM, такие как ускорение научных процессов, повышение точности и возможность интеграции междисциплинарных знаний. Однако обсуждаются и проблемы, такие как вопросы надежности, этическая ответственность за контент, созданный ИИ, и экологические аспекты. В статье делается вывод о том, что, хотя LLM значительно способствуют научным достижениям, их использование требует пересмотра традиционных понятий в философии науки и установления новых этических норм для обеспечения прозрачности, подотчетности и добросовестности в исследованиях с участием ИИ.

This article examines the nature of scientific knowledge generated by Large Language Models (LLMs) and assesses their impact on scientific discoveries and the philosophy of science. LLMs, such as GPT‑4, are advanced deep learning algorithms capable of performing various natural language processing tasks, including text generation, translation, and data analysis. The study aims to explore how these technologies influence the scientific research process, questioning the classification and validity of AI‑assisted scientific discoveries. The methodology involves a comprehensive review of existing literature on the application of LLMs in various scientific fields, coupled with an analysis of their ethical implications. Key findings highlight the benefits of LLMs, including accelerated research processes, enhanced accuracy, and the ability to integrate interdisciplinary knowledge. However, challenges such as issues of reliability, the ethical responsibility of AI‑generated content, and environmental concerns are also discussed. The paper concludes that while LLMs significantly contribute to scientific advancements, their use necessitates a reevaluation of traditional concepts in the philosophy of science and the establishment of new ethical guidelines to ensure transparency, accountability, and integrity in AI‑assisted research. This balanced approach aims to harness the potential of LLMs while addressing the ethical and practical challenges they present.

большие языковые моделинаучное знаниенаучное открытиестратегия научного поискаэтика искусственного интеллектаlarge language modelsscientific knowledgescientific discoveryscientific research strategyethics of artificial intelligenceИсследование выполнено за счет гранта Российского научного фонда № 24-28-00540, https://rscf.ru/project/24-28-00540.ReferencesAddis M. et al. Computational Scientific Discovery and Cognitive Science Theories // Computing and Philosophy Synthese Library. Cham: Springer International Publishing, 2016. P. 83‑97.Addis M. et al. Computational Scientific Discovery and Cognitive Science Theories // Computing and Philosophy Synthese Library. Cham: Springer International Publishing, 2016. P. 83‑97.AI4Science. Microsoft Research, Quantum, Microsoft Azure. The Impact of Large Language Models on Scientific Discovery: a Preliminary Study using GPT‑4. URL: https://arxiv.org/abs/2311.07361 (дата обращения: 16.05.2024).AI4Science. Microsoft Research, Quantum, Microsoft Azure. The Impact of Large Language Models on Scientific Discovery: a Preliminary Study using GPT‑4. URL: https://arxiv.org/abs/2311.07361 (дата обращения: 16.05.2024).Birhane A. et al. Science in the age of large language models // Nature Reviews Physics. 2023. Т. 5. № 5. P. 277‑280.Birhane A. et al. Science in the age of large language models // Nature Reviews Physics. 2023. Т. 5. № 5. P. 277‑280.Boden M.A. The creative mind: myths and mechanisms. London; New York: Routledge, 2004. Вып. 2nd ed. 344 p.Boden M.A. The creative mind: myths and mechanisms. London; New York: Routledge, 2004. Вып. 2nd ed. 344 p.Castelvecchi D. DeepMind AI outdoes human mathematicians on unsolved problem // Nature. 2024. Т. 625. № 7993. P. 12‑13.Castelvecchi D. DeepMind AI outdoes human mathematicians on unsolved problem // Nature. 2024. Т. 625. № 7993. P. 12‑13.Fawzi A. et al. Discovering faster matrix multiplication algorithms with reinforcement learning // Nature. 2022. Т. 610. № 7930. P. 47‑53.Fawzi A. et al. Discovering faster matrix multiplication algorithms with reinforcement learning // Nature. 2022. Т. 610. № 7930. P. 47‑53.Hinsen K. Computational science: shifting the focus from tools to models // F1000Res. 2014. Т. 3. P. 101.Hinsen K. Computational science: shifting the focus from tools to models // F1000Res. 2014. Т. 3. P. 101.Hope T. et al. A Computational Inflection for Scientific Discovery // Commun. ACM. 2023. Т. 66. № 8. P. 62‑73.Hope T. et al. A Computational Inflection for Scientific Discovery // Commun. ACM. 2023. Т. 66. № 8. P. 62‑73.Muntean I. Computation and Scientific Discovery? A Bio‑Inspired Approach // Artificial Life 14. Proceedings of the Fourteenth International Conference on the Synthesis and Simulation of Living Systems, 2014.Muntean I. Computation and Scientific Discovery? A Bio‑Inspired Approach // Artificial Life 14. Proceedings of the Fourteenth International Conference on the Synthesis and Simulation of Living Systems, 2014.Nickles T. Truth or Consequences? Generative versus Consequential Justification in Science // PSA: Proceedings of the Biennial Meeting of the Philosophy of Science Association. 1988. Т. 1988. P. 393‑405.Nickles T. Truth or Consequences? Generative versus Consequential Justification in Science // PSA: Proceedings of the Biennial Meeting of the Philosophy of Science Association. 1988. Т. 1988. P. 393‑405.Park J. S. et al.Generative Agents: Interactive Simulacra of Human Behavior // 2023.Park J. S. et al.Generative Agents: Interactive Simulacra of Human Behavior // 2023.Romera‑Paredes B. et al. Mathematical discoveries from program search with large language models // Nature. 2023.Romera‑Paredes B. et al. Mathematical discoveries from program search with large language models // Nature. 2023.Schick T. et al. Toolformer: Language Models Can Teach Themselves to Use Tools // 2023.Schick T. et al. Toolformer: Language Models Can Teach Themselves to Use Tools // 2023.Schickore J. Scientific Discovery // The Stanford Encyclopedia of Philosophy; edited by E. N. Zalta, U. Nodelman: Metaphysics Research Lab, Stanford University, 2020. Вып. Fall 2020.Schickore J. Scientific Discovery // The Stanford Encyclopedia of Philosophy; edited by E. N. Zalta, U. Nodelman: Metaphysics Research Lab, Stanford University, 2020. Вып. Fall 2020.Thagard P. Computing in the Philosophy of Science // The Blackwell guide to the philosophy of computing and information Blackwell philosophy guides; edited by L. Floridi. Malden, Mass.: Blackwell, 2004. Вып. 1. Publ. P. 307‑317.Thagard P. Computing in the Philosophy of Science // The Blackwell guide to the philosophy of computing and information Blackwell philosophy guides; edited by L. Floridi. Malden, Mass.: Blackwell, 2004. Вып. 1. Publ. P. 307‑317.FunSearch: Making new discoveries in mathematical sciences using Large Language Models. URL: https://deepmind.google/discover/blog/funsearch-making-new-discoveries-in-mathematical-sciences-using-large-language-models (дата обращения: 28.05.2024).FunSearch: Making new discoveries in mathematical sciences using Large Language Models. URL: https://deepmind.google/discover/blog/funsearch-making-new-discoveries-in-mathematical-sciences-using-large-language-models (дата обращения: 28.05.2024).AI for Scientific Discovery: Proceedings of a Workshop; edited by R. Pool. Washington, D.C.: National Academies Press, 2024. Вып. Science and Engineering Capacity Development Unit, Computer Science and Telecommunications Board, Policy and Global Affairs, Division on Engineering and Physical Sciences, National Academies of Sciences, Engineering, and Medicine.AI for Scientific Discovery: Proceedings of a Workshop; edited by R. Pool. Washington, D.C.: National Academies Press, 2024. Вып. Science and Engineering Capacity Development Unit, Computer Science and Telecommunications Board, Policy and Global Affairs, Division on Engineering and Physical Sciences, National Academies of Sciences, Engineering, and Medicine.What is a large language model (LLM)? URL: https://www.elastic.co/what-is/large-language-models (дата обращения: 28.05.2024).What is a large language model (LLM)? URL: https://www.elastic.co/what-is/large-language-models (дата обращения: 28.05.2024).

The authors declare that there are no conflicts of interest present.