Topic
The Turing Test and AI Benchmarks
The Turing Test and AI Benchmarks: How we measure what AI can do, from the Turing test to today's benchmarks..
The Turing test and AI benchmarks are methods for judging what AI systems can do. The Turing test asks whether a machine's conversation can be told apart from a human's, while benchmarks are standard tests that score AI performance on set tasks. In our stories, a 2025 study found that GPT-4.5 was judged human 73% of the time in five-minute chats, reported on 17 November 2026. On 15 November 2026, researchers described a spreadsheet test in which one AI model passed 38.3% of attempts. Amazon's open-source BenchProofer, covered on 17 November 2026, adds formal proofs to coding tests. Stanford's 2026 AI Index, covered on 6 November 2026, reports 53% adoption in three years.