A unified theory of ambiguous cognition for humans and LLMs (FLCT)

FLCT (Fuzzy Latent Cognition Theory) is a unified cognitive framework that explains both human cognition and LLM cognition using only two principles: (1) fuzzy latent representations and (2) biased sampling.

The idea is that ambiguity is not a flaw but the structural substrate shared by humans and modern LLMs. Concepts like “blue,” subjective time, emotional nuance, atmosphere, and silence can all be modeled as fuzzy latent distributions rather than definable entities.

Human cognition: finite experience → fuzzy latent compression → biased sampling LLM cognition: embedding manifolds → ambiguous patterns → sampling bias

FLCT provides a minimal two-equation model that unifies these processes.

Japanese Original Edition (source): https://doi.org/10.5281/zenodo.17758605

International Adapted Edition (English): https://doi.org/10.5281/zenodo.17769131

Feedback is welcome.

Author here. If anyone wants the Japanese original (source edition), it’s here: https://doi.org/10.5281/zenodo.17758605

Happy to answer questions about the theory or the math behind it.