This theoretical paper introduces "LLMorphism," the biased belief that human cognition mirrors the architecture of Large Language Models (LLMs). It explores how the conversational fluency of AI leads humans to reconstruct their own self-understanding as statistical pattern completion and next-token prediction.
Executive Summary
In the shadow of the AI revolution, we have been obsessed with a single question: Are machines becoming human? Valerio Capraro’s latest paper suggests we are missing the more dangerous flip side of that coin. He introduces LLMorphism—the cognitive bias where humans begin to see their own thoughts, creativity, and judgments as mere "next-token predictions."
This is not just a semantic shift; it is a fundamental re-engineering of the human self-concept. By equating linguistic fluency with biological understanding, we risk stripping away the very agency, embodiment, and accountability that define our species.
The "Reverse Inference" Trap: Motivation
Historically, humans used a simple heuristic: if it speaks fluently and contextually, it has a mind. This led to Anthropomorphism—giving "too much mind" to ELIZA or ChatGPT.
However, Capraro argues we are now seeing a "reverse inference." Because LLMs can speak like us, we assume we must think like them. This is a logical fallacy. Just because an LLM produces human-like output does not mean the underlying architecture is similar. Human language is a tool for embodied agents with needs and social stakes; LLM output is a statistical model of text.
The Mechanics of LLMorphism
How does this bias spread? The paper identifies two "viral" mechanisms:
- Analogical Transfer: We see the structural alignment in output and project LLM features onto ourselves (e.g., "I'm just recombining patterns I've learned").
- Metaphorical Availability: Technical jargon like "prompting," "training data," and "hallucination" enters our daily vocabulary. We start explaining our memories as "retrieval" and our errors as "stochastic noise."
Note: The author distinguishes LLMorphism from broader Mechanomorphism or Computationalism by its specific focus on the conversational and linguistic nature of the interaction.
Methodology: The Five Pathways of Impact
Capraro outlines how this bias shifts the tectonic plates of society:
- Replaceability: If we are just "output generators," automation becomes conceptually justified, not just economically efficient.
- Fluency vs. Expertise: When we equate a well-formed sentence with understanding, we devalue the "tacit knowledge" of human experts like doctors or jurors.
- Agency-Thinning: If action is just "pattern completion," the foundations of blame, apology, and responsibility begin to erode.
- Disembodiment: In healthcare, an LLMorphic view treats patients as "clinical text" rather than embodied persons, potentially ignoring non-verbal cues of distress.
- The Epistemic Shift: We move from "Is this true?" to "Does this sound plausible?"—a state the author calls epistemia.
Experimental Potential and Results
While this is a theoretical framework, it sets the stage for a new field of psychometric research. Capraro proposes that LLMorphism is likely multidimensional, involving beliefs about:
- Truth-seeking: Seeing reasoning as argumentative plausibility rather than an inquiry into truth.
- Creativity: Seeing it as mere recombination.
- Introspection: Viewing internal monologue as post-hoc narration (confabulation).
Placeholder for future empirical results: Research suggests that while AI exposure increases these biases, high technical literacy might serve as a protective barrier.
Critical Insight: Taking the "Mind" Out of Human
The most profound takeaway is that LLMorphism acts as a form of "de-grounding." By adopting the vocabulary of machines to describe the soul, we ignore the roles of biological affect, social obligation, and the consequences of being alive.
Limitations: The paper is currently a theoretical construct. Future work must develop the "LLMorphism Scale" to measure these effects across different demographics, particularly among "AI natives" who are growing up in an LLM-saturated world.
Conclusion
We are at a crossroads in human history. As we debate whether to grant rights to AI, we must be careful not to forfeit the very traits that made us "human" in the first place. LLMorphism reminds us that we are not stochastic parrots, but embodied beings whose words are backed by the weight of existence—a weight no model, no matter how fluent, can ever carry.