Murray Shanahan

Murray Shanahan arrives at the language-model debate with a credential that almost no other voice in the conversation possesses: he has spent decades building reasoning machines, proved theorems about the limits of logic, written a dense monograph on consciousness and embodiment, and served as a Principal Scientist at Google DeepMind, all while refusing the easy verdicts that the debate rewards. His distinctive move, practiced in his landmark essays Talking About Large Language Models (2022) and Role-Play with Large Language Models (2023, with Kyle McDonell and Laria Reynolds), is a discipline of repeated stepping back—holding the engineering mechanism in view precisely so that it deepens rather than dissolves the genuinely hard questions. He will tell you with mathematical precision that a large language model is a generative model of the statistical distribution of tokens in a vast text corpus, and then insist that this fact settles almost nothing about whether the words