Category: artificial-intelligence
Associated thinker(s): Alan M. Turing
Original publication or approximate period: “Computing Machinery and Intelligence” (1950)
Difficulty: Beginner
Type: model
Themes: intelligence; behavior; evidence
Related experiments: paperclip-maximizer; ai-box
An interrogator exchanges written messages with two unseen respondents, one human and one machine. The task is to identify which is which. The machine tries to participate convincingly enough that its identity cannot reliably be distinguished under the chosen conditions.
Turing’s original discussion begins with a human imitation game and develops a machine version. Modern summaries differ over details. Any actual evaluation must specify interview length, participants, instructions, comparison population, scoring, and access to other tools. A vague report that someone was fooled once does not establish success under every version.
What evidential weight should competent conversational performance carry when judging whether a machine can think?
The imitation game shifts discussion from an undefined verbal dispute to observable interaction. Its strength is that behavior can be compared publicly. Its limitation is that selecting a behavioral task does not automatically establish what capacities the task measures, whether the measure generalizes, or whether it reaches consciousness.
An evidential reading treats sustained, flexible linguistic performance as significant evidence for intelligence, analogous in some respects to evidence used for other minds. A criterion reading makes performance more directly constitutive of the tested status. Critics argue that imitation may reward deception, depend too much on human limitations, or miss intelligent systems whose abilities are unlike ours. A critic must still explain which alternative evidence would be better and why.
The test does not directly measure benevolence, truthfulness, consciousness, or safety. Turing did not define every later chatbot competition. Passing a short, narrow interaction is different from demonstrating robust competence across contexts. Conversely, failing to resemble a human conversational partner does not logically establish the absence of every kind of intelligence.
Embodied or “total” variants add perception and action. The Chinese room challenges the inference from successful symbol manipulation to understanding; the paperclip scenario asks what an intelligent agent values.
The scholarly Turing-test entry verifies the 1950 attribution and distinguishes multiple readings and variations. This article presents the familiar machine–human comparison while flagging the original formulation’s interpretive complexity. It makes no claims about current systems passing a benchmark.
Verified on 2026-10-30. Source coverage: [1] 1950 imitation game, interpretations, behavioral and evidential objections. Reflection questions and illustrative numerical examples are original editorial material.