The LLM as variety transducer: Ashby's intelligence-amplifier architecture and Beer's viable system model as a diagnostic and predictive framework
Richard S O'Rourke (Independent Researcher, London, UK)
Purpose — Ashby's intelligence-amplifier framework (1956) and Beer's viable system model (1984), taken together, offer a more precise and more diagnostically powerful account of large language model (LLM) systems than the four dominant AI framings.
Approach — Structure-mapping methods apply Beer's variety-transducer as the base model for LLMs and Ashby's formal specification for intelligence-amplifiers to modern LLM deployment via chat interaction, with reflexive case studies.
Findings — The coupled human–LLM–environment system constitutes the true intelligence-amplifier; the human supplies meta-systemic functions absent in the model alone, and autonomous agent failure reflects the structural absence of these functions.
Originality — It generates testable predictions regarding hallucination, prompt sensitivity, session variability, and autonomous agent failure.