Writing toward more rational AI.

Abstract

Observation is essential to the production of knowledge, yet it is also limited by perception, instruments, and prior assumptions. This paper examines whether these limitations weaken knowledge, with reference to the natural sciences and history.

In the natural sciences, observational flaws are often treated as errors that can be identified and reduced. Replication, falsification, peer review, and improved technology allow scientific claims to be tested over time. The 1919 eclipse observations used to support Einstein’s theory of relativity show that observation does not simply produce knowledge. It also tests claims that already exist. The thalidomide case, however, shows that limited observation can produce harmful conclusions before those limits are recognised.

In history, observation works differently. Historians cannot observe past events directly. They rely on testimony, records, and surviving artefacts. These sources are incomplete and shaped by perspective. Historical knowledge therefore depends on interpretation. Colonial accounts of Indigenous peoples show how one perspective can dominate the record. The Vinland Map also shows how later methods can expose earlier errors, though they cannot remove the basic limits of historical evidence.

This paper argues that flawed observation does matter, but not in the same way across different fields. In the natural sciences, reliability depends on systems of testing and correction. In history, it depends on critical interpretation and the comparison of sources. Knowledge does not require perfect observation. It requires methods that recognise and respond to its limits.

Abstract

Large language models are increasingly used in sustained interactions rather than isolated prompts. Existing research has examined failures such as hallucination, sycophancy, and factual error, but less attention has been given to how such failures can become connected through changes in the epistemic state of context over time.

We develop AI Epistemic Dynamics, a conceptual framework for analyzing this temporal dimension of LLM interaction. Central to the framework is the Epistemic Context State (ECS), which characterizes the epistemic state of the context on which subsequent generation depends. We distinguish four idealized states: Stable, Uncertain, Contaminated, and Self-Reinforcing. We then introduce Recursive Epistemic Contamination (REC) to explain how a low-quality claim (LQC) may contribute to further LQCs that become part of the context and alter subsequent generation. At the scale of sustained interaction, we call this process Micro-REC. We extend the same abstract recursive logic to Macro-REC, in which contaminated epistemic outputs may circulate through wider informational environments and alter the conditions of later knowledge production.

This framework recasts AI reliability as a dynamic rather than purely static epistemic problem. It explains how otherwise distinct local epistemic failures can become connected through path-dependent and recursively reinforced trajectories of degradation. AI Epistemic Dynamics thereby provides a basis for empirical research on epistemic-state transitions, recursive degradation, corrigibility, and the long-term reliability of AI systems.

Abstract

We argue that rational AI is not AI that removes uncertainty. It is AI that can handle uncertainty responsibly. This claim begins from a simple fact about human knowledge. Science and history do not work because they reach perfect certainty. They work because they expose doubt, test claims, give reasons, and correct mistakes. Error is not the enemy of reason. Unmanaged error is.

This point changes how we should understand large language models. The main problem is not simply that they hallucinate. The deeper problem is that their claims often sound authoritative even when their grounds are weak. These claims can enter classrooms, research, policy, search systems, and training data as if they were knowledge. Yet they often lack the responsibility structures that make knowledge claims trustworthy.

We develop this view through the H-A-R-R model. Humility requires AI to show its limits rather than speak with unwarranted authority. Accountability requires its claims to be open to scrutiny, challenge, and attribution within a responsible epistemic process. Responsiveness requires it to revise or qualify its answers when faced with counterevidence. Resilience names the capacity to maintain or restore epistemically stable conditions despite recursive contamination. Governance remains a further institutional requirement. Once AI-generated claims circulate through institutions, platforms, and data environments, responsibility cannot remain inside a single answer. We need mechanisms that prevent weak claims from being copied, cited, and fed back into future models.

We connect this account of epistemic responsibility to two forms of Recursive Epistemic Contamination (REC): Micro-REC, in which low-quality claims can alter the Epistemic Context State (ECS) of an ongoing dialogue and recursively shape later generation, and Macro-REC, in which such claims circulate through the wider knowledge ecosystem and alter the conditions of future knowledge production. A responsible AI system must therefore do more than avoid individual errors: it must be able to interrupt, resist, and recover from recursive contamination, while governance provides safeguards against the same dynamics at the institutional level. Rational AI, we conclude, should be understood not as an oracle, but as a corrigible cognitive partner.