Mehmet Ali Dombaycı

The integration of Large Language Models (LLMs) into human knowledge production processes has given rise to new problems within contemporary epistemology. This study argues that the factual inconsistencies frequently observed in the outputs of Large Language Models—commonly referred to as “hallucinations”—should be reconsidered through the conceptual framework of Harry Frankfurt’s notion of “bullshit”. The term “hallucination” is examined as a misleading metaphor that obscures the true nature of factual deviations in these model outputs. Such deviations are not merely technical errors; instead, they are the result of the models’ text-generation mechanisms, which are based on statistical probability and their structural indifference toward factual accuracy. This indifference parallels the disregard for truth emphasised in Frankfurt’s definition of bullshit, thereby positioning Large Language Models as systems inherently predisposed to producing bullshit continuously. Furthermore, techniques based on recalibration, such as Reinforcement Learning from Human Feedback (RLHF), are shown to reinforce this epistemic indifference by privileging persuasiveness over truth. As illustrated in John Searle’s Chinese Room argument, the lack of semantic grounding and the limitations of large language models in normative reasoning further complicate their epistemic status. From this perspective, the proliferation of model-based production weakens individual epistemic autonomy, thereby increasing the risk of “epistemic passivation” which poses a serious threat to collective knowledge structures and democratic processes.

With the growing prevalence of generative artificial intelligence, debates surrounding the value of art have gained renewed momentum. This study aims to deepen these debates by examining generative AI outputs through the lens of the concept of “genius,” which occupies a central place in Arthur Schopenhauer’s aesthetics. In Schopenhauer’s framework, the essential criteria of genius are summarised as the capacity to transcend individual desires and purposes to become a “pure knowing subject independent of the will,” to apprehend the Ideas beyond phenomena, and to produce an “interest-free” objectivity grounded in such apprehension. These criteria are compared with the production logic and operational structure of generative artificial intelligence. In doing so, the study provides a perspective on the productive conditions of the artwork within the ongoing discussion of whether the products of generative AI can be considered works of art. The conclusion argues that, although generative AI can produce outputs of remarkable sophistication, its success—viewed from a Schopenhauerian perspective—stems not from an act of genius that transcends rules and grasps the universal, but from its capacity to imitate and recombine existing patterns within data sets in a complex manner.