The Turing test was proposed by Alan Turing, a mathematician and pioneer of computer science, in the mid-twentieth century. Instead of asking "Can a machine think?", he suggested a more practical question: can a machine converse so well that it cannot be told apart from a human?
How the test works#
A human judge chats in text with two parties, one a human and the other a machine, without knowing which is which. If after the conversation the judge cannot reliably identify the machine, the machine is said to have passed the test.
Why it became famous#
The test was the first practical benchmark for discussing machine intelligence and has been present in popular culture and science fiction for years. With the arrival of chatbots that produce very fluent text, the debate about it heated up again.
Major criticisms#
- Fooling is not the same as being intelligent: a machine can deceive the judge with tricks, canned answers and imitated tone, without understanding anything.
- It measures only language: it does not test abilities like creativity, knowledge of the physical world and emotion.
- It depends on the judge: the outcome changes with the judge's experience and care.
Its place today#
Today researchers mostly use specialized benchmarks and multimodal tests to evaluate models, and the Turing test has more historical and philosophical value.
Common mistake#
Passing the Turing test does not mean the machine is conscious or truly understands; it only means that, in the judge's eyes, its conversation resembled a human's.