Development of a Dialog-Based Agent Equipped with an L-FTM-Based Mood Evaluation Mechanism for a Living Neuronal Networkt

Abstract

Recent research in artificial life aims to mimic biological principles in semi-artificial systems. This study interprets biological signal fluctuations as psychological phenomena using a living neuronal network, composed of dissociated and cultured neurons. A voice input/output interface allows interaction with the external environment. The system converts voice input to text and performs sentiment analysis. Based on the sentiment, it applies either a pleasant or unpleasant constant current stimulation to the neuronal network. The resulting electrical activity patterns are analyzed using a learning-based fuzzy template matching (FTM) algorithm. The FTM output represents the “ mood ” of the network on a pleasant-unpleasant scale. An emotional label corresponding to the mood is assigned, and a conversational response is generated through a large language model (LLM), forming a dialog-based agent. The mood is determined by comparing input patterns to predefined templates of pleasant and unpleasant responses. Experimental results revealed individual differences in the biological neuronal network, with emotional tendencies based on neuronal activity. However, inconsistencies in learning suggest potential variability in responses.