ChatGPT, extended: large language models and the extended mind
Synthese June 1, 2025 Paul Smart, Robert Clowes, A. Clark
The paper presents Retrieval-Augmented Generation (RAG) as an alternative to fine-tuning for incorporating philosophical knowledge into Large Language Models (LLMs). Unlike fine-tuning, which adjusts internal parameters, RAG retrieves information from an external store. The authors demonstrate this by creating Digital Andy, an LLM that answers questions about the extended mind using the works of philosopher Andy Clark. This approach highlights how philosophical knowledge can be integrated without additional machine learning. The reliance on external resources also prompts reflection on active externalist theorizing and Digital Andy's status as an extended cognitive system, revealing convergences between philosophical understanding of the extended mind and technological LLM development.