What does the Turing Test test? (philipball86.substack.com)

At a Royal Society meeting marking the 75th anniversary of Alan Turing’s 1950 paper, experts concluded the Turing Test—Turing’s “Imitation Game” for judging whether a machine’s responses are indistinguishable from a human’s—tests surprisingly little beyond human credulity. Speakers argued the test misreads human psychology: we’re primed to anthropomorphize even simple behaviors (the Kahneman “fast” thinking effect), so convincing imitation—exemplified by 1966’s ELIZA and today’s LLMs—can trick users without demonstrating genuine understanding. The panel also stressed that Turing’s original essay was playful and not intended as a rigorous benchmark, and cautioned against conflating imitation with intelligence. For the AI/ML community this is a timely reminder to stop using indistinguishability as a proxy for cognition or progress toward AGI. Technical implications include rethinking evaluation: conversational fluency or human-like outputs (text, music, art) are insufficient measures of reasoning, grounding, or intentionality. Critics like Gary Marcus and Shannon Vallor urged clearer definitions and more robust metrics that probe model capabilities beyond surface mimicry, and warned that even creators are prone to over-claiming. In short, passing a Turing-style test is impressive rhetorically, but not evidence of true understanding or intelligence.