Fellow Cognizants? A Kantian Perspective on Large Language Models

The linguistic abilities of large language models (LLMs) such as ChatGPT raise important philosophical questions about their potential cognitive abilities. This paper examines these questions within a Kantian framework by asking whether LLMs can possess cognition (Erkenntnis). By introducing a Kantian Thesis of Togetherness, which asserts that cognition requires both sensibility and understanding, I defend two key claims. First, from a strictly Kantian standpoint, we should remain agnostic about the cognition of LLMs, given our lack of phenomenological access to them and our lack of access to their internal way of processing language. Second, a revisionist interpretation of Kantian categories—viewing them metalinguistically rather than as pure concepts—opens up the possibility of attributing cognition to LLMs.