Abstract

This article examines how the emergence of Large Language Models (LLMs) and AI-assisted tools is transforming scientific inquiry through the lens of second-order science and UnCritically Examined Presuppositions (UCEPs). Building on foundational work by Kuhn (1962) and Popper (1959) on how paradigms and methodologies shape scientific knowledge, second-order science provides a critical framework for understanding how underlying assumptions influence scientific inquiry. This paper argues that AI systems—particularly LLMs—serve as both tools that can help identify UCEPs and as epistemic actors that introduce their own UCEPs into scientific practice. By examining the interplay between representations (labels/categories) and compressions (complex models involving ambiguity), we analyze how AI systems influence scientific explanations and risk amplifying errors through misapplied simplifications. Scientific communities and intelligence analysts face particular challenges as AI transforms knowledge practices that have traditionally bracketed ambiguity through enabling constraints. Through comparative analysis across disciplines, this paper demonstrates how reflexive interrogation of AI-human collaborations can strengthen scientific inquiry while avoiding the pitfalls of what has been termed “label causality.” The conclusion proposes that second-order science offers an essential methodological foundation for navigating the transformative impacts of AI on scientific epistemology, ensuring that these technologies enhance rather than undermine scientific rigor and creativity.