Large language models as cognitive shortcuts: a systems-theoretic reframing beyond bullshit

Area of Science:

• Techno-semiotics
• Cognitive Science
• Systems Theory

Background:

• Current Large Language Models (LLMs) discourse often uses deficit metaphors (e.g., "bullshit," "stochastic parrots"), focusing on perceived lacks in grounding, belief, or intention.
• These framings obscure the practical utility of LLMs in sense-making, ideation, and communication processes.
• A need exists for a non-anthropomorphic framework to understand LLM functionality within human-computer interaction.

Purpose of the Study:

• To reframe Large Language Models (LLMs) not as entities with agency or understanding, but as functional operators within techno-semiotic assemblages.
• To develop a theoretical model, GECOS (General Cognitive Shortcuts), that explains LLM utility in terms of communicative complexity reduction and interactional continuity.
• To shift focus from ontological debates about LLM "understanding" to their operational role in distributed sense-making.

Main Methods:

• Integration of concepts from Luhmannian systems theory, Deleuzian ontology, and Husserlian phenomenology.
• Application of conceptual analysis as a functional-comparative synthesis method.
• Analysis of human-LLM interaction without attributing anthropomorphic qualities like agency, belief, or understanding to the LLM.

Main Results:

• LLMs, reframed as GECOS, reduce communicative complexity by approximating "second-order expectations," facilitating interactional continuity.
• The GECOS framework explains LLM usefulness through procedural plausibility and coherence, helping users navigate uncertainty via Luhmann's contingency formula.
• LLMs enable workable continuations by generating connectable text, independent of truth or intentionality.

Conclusions:

• The GECOS framework provides a non-anthropomorphic model for understanding LLMs as pragmatic operators in complex socio-technical environments.
• This reframing shifts attention from LLM "understanding" to their operational role in distributed sense-making and communication.
• Potential risks include overreliance, emotional projection, and normative flattening when connectability replaces justification.