AI-Extended Agents and the Transformation of Human Communication: A Game-Theoretic Model of Norm Shift in Populations with AI-Mediated Communicators

This paper proposes a formal social-theoretic model of a phenomenon increasingly reported in anecdotal and preliminary empirical accounts: individuals who interact extensively with large language models (LLMs) develop elevated expectations for the depth and structure of interpersonal communication. Rather than producing social withdrawal, this shift generates systemic pressure that raises communication norms within social environments. Drawing on Granovetter's (1978) threshold models of collective behavior and evolutionary game theory, the paper introduces the concept of the AI-extended agent A_φ, defined by the application of an AI-mediation operator φ to a biological agent A. The operator is shown to be bidirectional: AI amplifies human cognitive capacity by providing access to vast knowledge repositories, real-time logical verification, and iterative refinement, while humans reciprocally improve the coupled system by supplying embodied experience, situated context, cultural nuance, and evaluative judgment that AI cannot independently access. The coupled system A_φ is therefore not A plus a static tool but a co-evolving cognitive dyad whose capabilities exceed those of either component operating independently. The model formalizes how AI mediation expands the agent's communicative strategy space, reduces interaction friction, and raises the agent's minimum acceptable dialogue depth. Using a threshold dynamics framework, conditions are derived under which a critical proportion p* of AI-extended agents triggers a phase transition in population-level communication norms. A dedicated section analyzes reputational resistance to AI adoption — the systematic underestimation of AI capability caused by early exposure to primitive models, drawing a structural parallel to the trajectory of Wikipedia from dismissed novelty to indispensable reference. Three mechanisms of social reinforcement are identified: stigmatization of AI use as intellectual dishonesty, Dunning–Kruger dynamics in AI evaluation (those who would benefit most are least able to assess current capability), and authority-based prohibition by institutions. Four mechanisms by which resistance is overcome are formalized: undeniable output asymmetry, generational replacement, institutional capitulation, and normalization through ubiquity. The communication norm shift is situated within the formal framework for a unified civilization of autonomous agents (Kriger, 2026b), showing that it constitutes the empirically observable precursor phase to the civilizational transition. The AI-extended agent A_φ is formally a transitional state between the unbounded executor and the fully autonomous subject, and the communication friction reduced by φ is substantially composed of the biological distortion D demonstrated to be irreducible in living organisms (Kriger, 2026a). Six falsifiable hypotheses with specific study designs are presented, including a discriminating longitudinal prediction (H6) that separates the causal account from the selection-effect alternative. Appendix A provides a three-version dialogue demonstration contrasting unmediated human conversation (destructive outcome), AI-mediated asynchronous writing (productive outcome with qualitatively deeper analysis), and live conversation with AI as a third participant (productive outcome demonstrating bidirectional learning, where humans correct AI's omissions and AI introduces theoretical frameworks inaccessible to the humans alone). Keywords: AI-mediated communication, game theory, threshold models, communication norms, human–AI interaction, phase transitions, strategy space, cognitive amplification, bidirectional learning, autonomous agents, biological distortion, unified civilization, reputational resistance Related identifiers Kriger, B. (2026). The Stimulus Problem: A Formal Theory of Goal Generation in Post-Scarcity Information Environments. Information Physics Institute. https://doi.org/10.5281/zenodo.18511908 Kriger, B. (2026). The Inevitability of a Unified Civilization of Autonomous Agents: Why the Biological Basis of Subjecthood Becomes Irrelevant. Information Physics Institute. https://doi.org/10.5281/zenodo.18512941