You are the interrogator. One answer is a person, one is a machine. Spot the machine.

Turing's 1950 game, compressed: each round shows one interrogator question and the two answers it got. Click the answer you believe came from the machine. Turing predicted that after five minutes of this, the average interrogator would be right at most 70 % of the time.

All dialogues are constructed teaching material, written for this course — they illustrate answer styles reported in published Turing-test studies, and are not transcripts of any real person or system.

Round 1 of 5

Interrogator

Your result as interrogator

0/5

≤ 70 %

Turing's 1950 prediction: the ceiling for an average interrogator's accuracy after five minutes of questioning (Mind, 1950)

73 %

How often one 2025 system was judged to be the human — more often than the real human participants (Jones & Bergen, preprint, three-party test, 5 min)

21 %

A plain chat assistant in the same study, judged human far below chance: its default style gives it away

What the game did and did not measure

Every cue you used — tone, detail, hedging, structure — is a property of style, and style is trainable. Treat it as a hint, never as proof, in either direction.
The test scores imitation of conversation. It says nothing about whether an answer was true, or useful for a decision.
The question to carry into section 4 and the whole module: not "does it pass?" but "does it help — and on which data was it trained?"

Back to the labs

Sources: Turing, Computing Machinery and Intelligence, Mind 59, 433–460 (1950), doi:10.1093/mind/LIX.236.433 · Jones & Bergen, Large Language Models Pass the Turing Test, arXiv:2503.23674 (2025, preprint). Dialogues: constructed teaching material, HTW M6 Block 01 (2026) — not transcripts from either study. Paul Manski
HTW · M6 · AI in Life Sciences HTW Berlin