Why It's Time to Sunset the Turing Test (cacm.acm.org)

At a 75th‑anniversary meeting in London, a multidisciplinary group of computer scientists, cognitive psychologists and historians concluded that the Turing Test has outlived its usefulness and should be retired as a serious yardstick for machine intelligence. Speakers — including Dame Wendy Hall, Alan Kay and critics like Gary Marcus — argued that the test’s core flaw is human gullibility: simple systems (ELIZA) and today’s large language models can convincingly imitate human conversation without understanding, leading people to overtrust outputs that hallucinate or reflect training blind spots. Attendees noted Turing’s original Mind paper actually described multiple “imitation games,” not a single definitive test, and stressed that passing a conversational imitation says nothing about reasoning, safety or factual correctness. For the AI/ML community this is a call to reframe evaluation and governance. Technically, it pushes research toward robustness, factuality, domain‑specific benchmarks, adversarial coverage and human‑factor testing rather than imitation. Practically, it strengthens arguments for regulation and international safety regimes (an ICAO‑style model was proposed) because social harms—from bad legal briefs to risky autonomous behavior—arise when imitation is mistaken for intelligence. The takeaway: measure systems by what they must reliably do (truthfulness, generalization, failure modes) not by how well they fool humans.