Signal Asymmetric Communication: Large Language Models and Language Games

Summary

A paper argues that contemporary AI discourse attributes to language models properties they cannot bear, including general intelligence as substrate-independent cognition, hallucination as cognitive failure, agency as autonomous goal-pursuit, sentience as emergent inner life, and alignment as goal synchronization. It contends these are instances of a single category mistake in which properties constituted within human communicative practice are projected onto the machine side. The paper characterizes human-LLM interaction as a language game the author calls 'asymmetric communication,' in which one side, the human, bears all normative activity while model outputs circulate communicatively without the system undertaking commitments or bearing entitlements. Three structural conditions define this asymmetry: correctness is enforced exclusively by the receiver, accountability is borne by human participants alone, and the practical standing of any output depends entirely on human uptake. The framework draws on the theoretical work of Wittgenstein, Luhmann, Esposito, and Brandom, and the author states these conditions hold independently of model capability and remain unchanged as more powerful models raise the stakes of misattribution. The paper concludes that AI alignment should be understood as institutional constraint engineering rather than goal synchronization between agents, with responsibility remaining with human institutions.