Hallucination Under Epistemic Incompleteness: A Comparative Framework for Human and Artificial Intelligence

This article examines hallucination as a possible outcome of inference under conditions of epistemic incompleteness. While hallucinations in generative artificial intelligence are commonly treated as technical failures resulting from missing or inaccurate data, we argue that analogous epistemic constraints also characterize human cognition. The article proposes a comparative epistemological framework explaining how human and artificial systems construct knowledge under incomplete conditions. It identifies four interrelated constraints—limited perception, representational and translational loss, bounded processing, and material constraints—that make inference both necessary and inherently fallible. First, neither humans nor artificial intelligence (AI) systems can access the totality of available information. Second, knowledge is constructed through successive layers of representation and translation, introducing compression, transformation, and inevitable informational loss. Third, inference operates under finite processing capacities constrained by memory, time, energy, and cognitive or computational architecture. Finally, cognition and computation remain embedded in material conditions that restrict what can be known and processed. Under these constraints, systems rely on approximation, prediction, and pattern completion. Hallucination is therefore interpreted not merely as an error but as a byproduct of meaning-making when complete certainty is unavailable. Reality functions as the ultimate corrective process through which hypotheses are continuously tested and revised. By situating hallucination within a broader epistemological framework, this study contributes to discussions on AI, cognition, uncertainty, and the limits of knowledge, arguing that neither increasing computational power nor expanding data availability can eliminate the epistemic conditions that make hallucination possible.