Fuzzy-Cognitive Integration and LLM-Assisted Interpretation: Toward Traceable and Cognitively Faithful AI Explanations
Year: 2024
Type: conference-paper
Abstract: This article presents a hybrid Fuzzy–LLM framework that bridges numerical modeling and linguistic reasoning to produce traceable, cognitively faithful explanations in sensory and cognitive modeling. Building upon the previous works (Explainable Hybrid Modeling, Robust Explainable Modeling, and From Explainability to Prescription), it introduces a new dataset, FuzzyCog-Sensory v1, combining physico-chemical variables, hedonic responses, and linguistic annotations.The study evaluates how fuzzy rules derived from sensory data can be interpreted by Large Language Models (LLMs) under controlled traceability constraints. Quantitative metrics—Cognitive Fidelity Index (CFI), Align-Score, and Bias Rate—measure the alignment between numerical dependencies and generated linguistic explanations. Results demonstrate that Fuzzy–LLM systems can generate consistent, auditable, and human-centered scientific narratives, advancing explainable AI toward true cognitive interpretability.
Authors: Oettgen, Walther, Saulnier, Benjamin
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7092366020
FWCI:
Cites:
Cited by:
Related to: