Xerox and Jixin: On Metonymy and Large Language Models

Current Large Language Models have stirred controversy about the potential of machines to replicate human thought. I argue that such claims are rooted in a highly limited understanding of human cognition, one rooted in a reduction of human language to the electronic replication of English letters. Following the work of Jacques Lacan, I contend that we read LLM generations as meaningful because we human beings are “thrown” into meaning-making. Through our use of symbolic language, we interpret the generation of signs from a machine to be meaningful communication. But metonymic thinking helps us escape this trap by showing that meaning itself is not symbolic but rather imaginary. The untranslatability of Daoist concepts like Dao, jixin and wuwei clarify deeper meaning beyond relations of symbols.