Presentation Information
[U03-01]From Authorship to Knowledge Contribution: Journals in the LLM Era with Implications for Earth and Planetary Science★Invited Papers
*Hajime Naruse1 (1.Department of Geology and Mineralogy, Graduate School of Science, Kyoto University)
Keywords:
Large Language Models (LLMs),Scholarly Publishing,Authorship and Credit,Knowledge Representation
The rapid development of large language models (LLMs) is not merely the introduction of a new writing tool into academic practice; it challenges the conceptual foundations of scholarly journals, which operate at the level of collective knowledge production. Dissertations may legitimately evaluate individual competence, but that separate debate lies beyond the scope of this talk. Here, I adopt the position that the primary purpose of journals is to advance science itself, not to display or reward individual ability.
From this perspective, authorship functions primarily as an institutional mechanism for assigning responsibility and sustaining motivation, rather than as the ultimate marker of intellectual value. If so, there is no principled reason to reject a study largely generated with LLM assistance, provided that its contribution is reliable and advances understanding. The key question is not who produced the prose, but how knowledge is validated and made accountable. Once LLMs are recognized as part of the epistemic environment of science, the issue is no longer whether their use should be permitted, but how reliability and progress can be maintained in a world where their use is unavoidable. Academic publishing cannot realistically be organized around the assumption that generative systems do not exist.
LLMs may surpass human researchers in formal reasoning and large-scale data analysis, removing analytical bottlenecks in many disciplines. However, in Earth and planetary sciences, research remains deeply dependent on data acquisition—fieldwork, sampling, experiments, and instrumentation. While LLMs may transform interpretation and modeling, the physical and logistical constraints on data acquisition will likely remain a primary bottleneck. Our field may therefore evolve differently from domains in which analytical capacity is the primary limitation.
LLMs also reshape the global structure of knowledge production. Non-native English speakers reduce linguistic barriers that have long constrained participation in international publishing. By assisting with drafting and argumentation, LLMs allow researchers to focus more directly on substantive scientific reasoning, potentially making the research community more inclusive.
These developments also invite reconsideration of the form of the scientific paper. Conventional articles are fixed narratives written for heterogeneous audiences, often leading to structural redundancy. LLMs open the possibility of “compressed representations” of knowledge: structured, minimally redundant descriptions of questions, methods, data, and conclusions. Journals could certify this structured core, while LLM systems dynamically generate audience-specific versions. In this model, journals shift from distributing static texts to validating structured knowledge.
Such a transformation carries risks, including homogenization of ideas, increased submission volume, and concerns about attribution and control. Addressing these challenges will be essential to sustaining trust and diversity. Rather than offering definitive answers, this talk invites discussion on how journals should evolve as generative systems become deeply embedded in scientific practice.
From this perspective, authorship functions primarily as an institutional mechanism for assigning responsibility and sustaining motivation, rather than as the ultimate marker of intellectual value. If so, there is no principled reason to reject a study largely generated with LLM assistance, provided that its contribution is reliable and advances understanding. The key question is not who produced the prose, but how knowledge is validated and made accountable. Once LLMs are recognized as part of the epistemic environment of science, the issue is no longer whether their use should be permitted, but how reliability and progress can be maintained in a world where their use is unavoidable. Academic publishing cannot realistically be organized around the assumption that generative systems do not exist.
LLMs may surpass human researchers in formal reasoning and large-scale data analysis, removing analytical bottlenecks in many disciplines. However, in Earth and planetary sciences, research remains deeply dependent on data acquisition—fieldwork, sampling, experiments, and instrumentation. While LLMs may transform interpretation and modeling, the physical and logistical constraints on data acquisition will likely remain a primary bottleneck. Our field may therefore evolve differently from domains in which analytical capacity is the primary limitation.
LLMs also reshape the global structure of knowledge production. Non-native English speakers reduce linguistic barriers that have long constrained participation in international publishing. By assisting with drafting and argumentation, LLMs allow researchers to focus more directly on substantive scientific reasoning, potentially making the research community more inclusive.
These developments also invite reconsideration of the form of the scientific paper. Conventional articles are fixed narratives written for heterogeneous audiences, often leading to structural redundancy. LLMs open the possibility of “compressed representations” of knowledge: structured, minimally redundant descriptions of questions, methods, data, and conclusions. Journals could certify this structured core, while LLM systems dynamically generate audience-specific versions. In this model, journals shift from distributing static texts to validating structured knowledge.
Such a transformation carries risks, including homogenization of ideas, increased submission volume, and concerns about attribution and control. Addressing these challenges will be essential to sustaining trust and diversity. Rather than offering definitive answers, this talk invites discussion on how journals should evolve as generative systems become deeply embedded in scientific practice.