Why the Turing Test No Longer Reveals Whether AI Truly Thinks

A new analysis argues that the Turing Test has lost its power to diagnose machine intelligence now that large language models can convincingly mimic human conversation. It cites research showing that GPT-4.5 is often judged indistinguishable from a human in conversation. The debate is therefore shifting from whether AI is intelligent to how its outputs are produced. The author proposes replacing single-test evaluation with scattering experiments, drawing on ideas from Douglas Hofstadter and the FARG program. These experiments would examine how a system responds across diverse and atypical situations, separating superficial imitation from genuine underlying competence.