Human Agency and Epistemic Authority Under Generative Artificial Intelligence

This study examines the epistemic, pedagogical, and institutional ruptures introduced by generative artificial intelligence in educational assessment and evaluation, as well as in scholarly publishing. As large language models (LLMs) achieve increasingly high levels of fluency and coherence, it becomes harder to determine who the relevant agent is behind learning outcomes and academic texts. In response, education systems and peer-reviewed publishing have shown a growing tendency to delegate assessment, evaluation, and oversight to AI-based tools. We argue that this shift is not merely a technical adjustment but a structural transformation that erodes human responsibility and epistemic authority by assigning both production and judgment to closely related algorithmic systems. By discussing the limitations of AI detection tools—especially their false-positive risks—the article highlights the ethical and epistemic problems that arise when academic integrity is reduced to the formal features of text. In educational contexts, LLM-supported assignments and examinations can obscure students’ cognitive effort and weaken the connection between learning and achievement. In scholarly publishing, the same dynamic encourages a surveillance-oriented posture that treats style as evidence of authorship while failing to secure reliability, originality, or conceptual contribution. Against this backdrop, the study argues that the solution lies not in more sophisticated surveillance, but in redesigning assessment and evaluation to re-anchor production and judgment in human oversight. For education, we propose a framework that structurally separates the learning phase from the evaluation phase: LLMs may be used as complementary tools during learning, while evidence of learning is produced under controlled conditions and human supervision. For scholarly publishing, we advocate replacing detection regimes with transparent disclosure of AI use, while keeping final evaluation grounded in human peer review. We conclude that this approach can protect pedagogical validity and academic integrity while restoring human responsibility and epistemic standing in the age of generative artificial intelligence.