How It Works — Step by Step
The Turing Test is not a complex machine — it is a protocol. Here is how it works, broken down exactly.
Step 1: Set up the game
Three participants are required. The interrogator sits alone in one room. In separate rooms (or behind screens) are one human and one machine. The interrogator cannot see or hear them — all communication happens through typed text only. Voice, face, and handwriting are all excluded to prevent easy cheating.
The human’s job is to convince the interrogator they are human. The machine’s job is also to convince the interrogator it is human — by producing typed responses that sound natural, intelligent, and human-like.
Step 2: The interrogator asks questions
The interrogator can ask anything. Turing gives examples:
“Please write me a sonnet on the subject of the Forth Bridge.”
The machine might reply: “Count me out on this one. I never could write poetry.”
This response is actually brilliant. A human might say the same thing. A machine that tried too hard to write a perfect sonnet might give itself away by being too good. Pretending to have limitations is itself a human behaviour.
Other questions might probe:
- Arithmetic: “What is 34957 plus 70764?” (A machine might deliberately introduce a small error to seem more human)
- Opinion: “Do you like cricket?”
- Memory: “What did you say three questions ago?”
- Creativity: “Make up a story that starts with: ‘It was a dark and stormy night…’”
The interrogator can be as clever or as probing as they like. There are no restrictions on questions.
Step 3: The machine responds
The machine’s responses must be typed and sent through a neutral interface. It has time to construct each response. It can choose to be slow or fast in responding (a machine that responds instantly might seem suspicious on complex questions).
The machine can:
- Pretend not to know things it actually knows
- Introduce deliberate errors in arithmetic
- Claim to have emotions, feelings, and opinions
- Express uncertainty or hesitation
- Tell jokes or make small talk
Anything a human could type, the machine is permitted to type.
Step 4: The interrogator decides
After a fixed time — Turing suggested about five minutes of conversation — the interrogator must guess: which one is the human? Which is the machine?
This is the critical moment. The Turing Test is passed if the machine fools the interrogator at least as often as the human fooled the interrogator in a control condition (where both participants are human).
Step 5: Measure the result
Turing asked: “What will happen when a machine takes the part of A [the deceiver] in this game? Will the interrogator decide wrongly as often when the game is played like this as he does when the game is played between a man and a woman?”
In other words: is the machine as good at fooling humans as humans are at fooling each other?
If yes — the machine passes. Not because we have proven it thinks. But because we cannot distinguish it from something that does.
Step 6: Turing’s prediction
Turing made a specific prediction in the paper. He said that by the year 2000, machines would be good enough at conversation that an average interrogator would have no more than a 70% chance of correctly identifying the machine after five minutes of conversation.
This prediction was remarkably close. In 2014, a programme called Eugene Goostman — pretending to be a 13-year-old Ukrainian boy (a clever choice, since a non-native English speaker can “explain” odd responses) — was judged by 33% of interrogators to be human in a competition at the Royal Society in London. Whether this counts as “passing” the Turing Test is still debated. But the scale of Turing’s prediction was correct.
Modern large language models like GPT-4 and Claude would almost certainly fool most interrogators in a five-minute conversation, if they were trying to.
Next: The Mathematics →