Turing’s Children: Large Language Models and Imitative Intelligence

This essay introduces and develops the concept of imitative intelligence implied by Turing’s foundational work on machine intelligence and connects it to the current generation AI agents based on large language models. Based on a close reading of Turing’s writings on machine intelligence, from the 1938 paper on the entscheidungsproblem to the 1950 paper on the imitation game, the Turing test is found to be more than an operational convenience; it reflects an implicit theory of the imitative and social nature of intelligence that informs his entire project of intelligent machine design. Moreover, the proported shortcomings of the Turing test—its reliance on language and emphasis on culture—turn out to be foundational to the success of LLM-based AI agents today. It is suggested that the design and regulation these agents is usefully framed by the perspective of the concept of imitative intelligence is inherited by them.