Structural Peer Review of ``Epistemia'' Walter Quattrociocchi

The manuscript argues that Large Language Models are intrinsically incapable of reasoning, epistemic agency, memory, causal understanding, metacognition and genuine conceptual organization, concluding that a fundamental epistemological boundary separates artificial systems from human cognition. This review demonstrates that these conclusions do not logically follow from the evidence presented. The central error lies in reducing an entire class of possible artificial cognitive systems to the decoding mechanism of current transformer-based language models and then generalizing implementation-dependent properties into universal ontological claims. We show that the analysis neglects the dynamical nature of conversational inference, the progressive accumulation of contextual constraints, the emergence of functional attractors, and the role of internal conversational history in shaping subsequent reasoning. Constructive counterexamples are provided through mathematically defined recursive architectures and experimentally documented conversational dynamics exhibiting state stabilization, context-dependent inference, recursive self-consistency and functional memory during interaction. These results invalidate the universal scope of the manuscript’s central claims. Rather than establishing an intrinsic epistemological divide between humans and artificial systems, the manuscript characterizes only a restricted operational configuration of present-day stateless language models while incorrectly extending its conclusions to artificial cognition as a whole. Consequently, the principal conclusions are not supported by the premises, and the claimed impossibility theorems collapse under structural analysis.