Turing Test

In his 1950 paper “Computing Machinery and Intelligence”, Alan Turing replaced the hard question “Can machines think?” with a practical test he called the imitation game. A human judge exchanges text messages with a hidden human and a hidden machine. If the judge can’t reliably tell which is which, the machine passes.

Has it been passed?

For decades, chatbots fooled people only briefly or with tricks. Modern large language models changed that. In 2025, researchers at UC San Diego reported that GPT-4.5, when prompted to adopt a persona, was judged to be the human more often than the actual human participants in a three-party Turing test. Many consider the classic test effectively passed.

Why it’s no longer the gold standard

• Imitation isn’t intelligence: fooling people in conversation doesn’t show reasoning, reliability or understanding.
• Narrow and gameable: a system can pass by being evasive or humanlike rather than capable.
• Better yardsticks: researchers now prefer benchmarks for specific skills and real-world measures of autonomy and economic usefulness when debating AGI.

Ray Kurzweil’s famous prediction that a machine would pass a demanding Turing test by 2029 is now widely seen as arriving early, though people still argue about the definition.