Technohybridism: Recursive Semantic Normalization in AI-Mediated Societies

Charalampos M. Liapis, Epameinondas PanagopoulosarXiv 2026 · arXiv

TL;DR

Generative AI is no longer just a tool we use; its outputs are actively re-entering and reshaping the human linguistic environment. This paper introduces Recursive Semantic Normalization (RSN), a framework explaining how machine-generated language patterns become social norms through human adoption, potentially "laundering" their machine origins and narrowing the diversity of human thought and expression.

What Problem Does This Paper Address?

Current debates about Large Language Models (LLMs) often get stuck on two questions: "Do they actually understand anything?" and "Will they technically collapse if trained on their own data?"

The authors argue we are missing a critical third dimension: the social loop. AI-generated text is now everywhere—in scientific abstracts, peer reviews, news releases, and everyday emails. When humans accept, edit, and reuse these machine-generated formulations, those formulations become part of the "linguistic environment." This creates a feedback loop where the statistical regularities of a machine become the new baseline for human communication and, eventually, the training data for the next generation of AI. The paper seeks to define this mechanism and its long-term consequences for how societies organize meaning and knowledge.

Methods

This is a conceptual and theoretical paper that synthesizes social theory, social epistemology, and the philosophy of language. The authors develop an analytical model of the "hybrid semantic system"—a configuration where human practices, computational generation, and institutional uptake are recursively coupled.

Key components of their approach include:

  • Distinguishing RSN: They systematically differentiate their concept from related ideas like "model collapse" (a technical failure) and "looping effects" (how classifications change the people classified).
  • The Analytical Model: They break down the process into six stages: domain of practice, candidate space (what the AI proposes), selection (human judgment), uptake share, persistence (how long the change lasts), and a counterfactual baseline (what would have happened without AI).
  • The Evidential Ladder: They propose a methodological framework for empirical researchers to measure RSN, moving from simple word frequency changes to deep conceptual shifts in how communities form beliefs.
Figure 1: The hybrid semantic system and its two orders of recursion. Solid return arrow: first-order (social) recursion, through which the altered environment conditions later practice. Dashed return arrow: second-order (computational) recursion, through which it also enters the corpora of later models. Components correspond to the analytical model in Sect. 3.4

Datasets

Not discussed in the paper. As a conceptual work, it does not introduce new datasets. However, it cites several recent empirical studies as motivation, such as:

  • Kobak et al. (2025): Estimates that 13.5% of biomedical abstracts in 2024 were LLM-processed.
  • Liang et al. (2024/2025): Research on LLM-assisted peer reviews and UN press releases.
  • Yakura et al. (2026): Evidence of LLM-characteristic words entering unscripted human speech.

Key Insights and Ideas

  • Technohybridism: The authors argue that the "unit of analysis" for language today is neither the human nor the machine alone, but the hybrid system where both interact recursively.
  • Provenance Laundering: When a human adopts an AI-generated phrase into their own writing, the machine origin is "laundered." Hearers attribute the statement to the human's authority, bypassing the "epistemic vigilance" we usually apply to suspicious sources.
  • From Regularity to Norm: A machine-generated pattern becomes a norm when it meets three criteria: Correction (deviations are treated as errors), Expectation (people expect others to use the form), and Codification (it enters style guides or templates).
  • Erosion of Cognitive Diversity: Because LLMs tend to favor "bland," "Western-style," or "statistically likely" phrasing, RSN can lead to a narrowing of the "hermeneutical resources" (the tools we use to make sense of our experience) available to a community.
  • Social Stratification: RSN isn't uniform. It may help marginalized groups access "legitimate" institutional language, but it does so by forcing their expression into a narrow, machine-standardized mold.
Table 1: Recursive semantic normalization (RSN) and neighbouring concepts. The comparison concerns the analytical focus of each concept, not the full scope of the literatures cited.

Key Results

The paper provides a rigorous theoretical framework rather than quantitative benchmarks. Its main "results" are the definitions and the Evidential Ladder:

  1. Lexical Rung: Changes in word frequency (e.g., the sudden rise of the word "delve").
  2. Phraseological Rung: Convergence in templates and genre structures.
  3. Framing Rung: Shifts in the perspectives and argument types used in a field.
  4. Conceptual Rung: Changes in how key terms are defined and how inferences are licensed.

The authors note that while we have plenty of evidence for the "Lexical" rung, the higher rungs—where the real semantic danger lies—require more longitudinal and ethnographic research.

Limitations and Future Work

  • Empirical Gap: The paper is conceptual; the authors acknowledge that the "higher rungs" of the evidential ladder currently lack systematic evidence.
  • Language Bias: Most cited studies focus on English and professional/academic contexts. How RSN functions in low-resource languages or different cultural settings remains an open question.
  • Political Economy: The framework doesn't deeply address the power dynamics of the few companies that control the "candidate space" (the models themselves).
  • Future Directions: The authors call for "epistemic reflexivity"—institutions like universities and courts must become aware of how their adoption of AI-generated "boilerplate" might be permanently altering the linguistic field they operate within.

Summary updated

Abstract

Language-generating AI systems are no longer only objects of linguistic interaction: their outputs increasingly re-enter the communicative and institutional practices from which their training material was drawn. Drawing on social theory, social epistemology, and the philosophy of language, this article develops a sociotechnical account of that re-entry. We introduce recursive semantic normalization: the process through which machine-generated formulations, once recurrently taken up, evaluated, reused, and institutionally retained, participate in the reproduction and transformation of semantic regularities. Under the heading of technohybridism, we argue that the relevant unit of analysis is the hybrid semantic system, a recursively coupled configuration of human practices, computational generation, and institutional uptake. First, we distinguish the concept from neighbouring accounts, including model collapse, looping effects, machine habitus, machine culture, and the epistemology of AI testimony. Second, we develop an analytical model that separates generation, selection, uptake, and persistence and distinguishes machine-attributable change from change that would have occurred anyway. Third, we analyse uptake as routinized, institutionally diffused, and socially stratified practice and state conditions under which machine-derived regularities become norms. Fourth, we identify social-epistemic consequences: the laundering of provenance that weakens epistemic vigilance, the erosion of the cognitive diversity on which collective inquiry depends, and the narrowing of shared hermeneutical resources. An evidential ladder for empirical research follows. The account attributes no normative authority to machine outputs; what changes is the field within which human and institutional uptake operates.
BibTeX
@misc{liapis2026technohybridism,
  author = {Charalampos M. Liapis and Epameinondas Panagopoulos},
  title = {{Technohybridism: Recursive Semantic Normalization in AI-Mediated Societies}},
  year = {2026},
  eprint = {2611.07984},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2611.07984},
}
Figure