Sellars, Kant and Hegel: Hegelian Reflections on the Large Language Models

Abstract

The pragmatist philosopher W. Sellars, in his essay "Philosophy and the Scientific Image of Man", proposed that it would be science, through its theoretical work, that could account for the place of human beings in the world by integrating the conceptual framework of human beings and their relations into a common ontology. We propose to use large language models (LLMs), such as ChatGPT 3.5, as the most explicit attempt at a theory that could integrate such a conceptual framework of human beings. Given Sellars's interest in the philosophies of Kant and Hegel, we will critique the ambitions of LLMs in Sellars's terms, and the possibility of them moving from models of language to models of knowledge. Our critique will focus on the possibility of a criterion of unity (such as the use of tools that aid inferential reasoning) to avoid hallucinations, as well as the main driver behind the possible transition to models of knowledge. We will point out that this effort lies in the position that Hegel calls subjective idealism, which shows that a principle of explanation cannot be predictive. This will lead to the impossibility of integrating the common ontology proposed by Sellars.