The Turing test is a proposal, made by mathematician Alan Turing in 1950, for sidestepping the question of whether a machine can think: instead of debating that directly, Turing asked whether a computer could do well enough in a text-based imitation game that a human judge, conversing with both a machine and another human, could not reliably tell which was which.
Turing’s own framing, as the Stanford Encyclopedia of Philosophy’s entry on the test records it, was that the deeper question — whether machines can think — was itself “too meaningless”, and that the imitation game gave a more precise, debatable stand-in for it. That source is an academic philosophy reference, not a Canadian or governmental one, and is cited here for the history of the idea, not as law or policy.
No Canadian statute, regulator or court decision defines or relies on the Turing test, and there is no reason to expect one to: passing a conversational imitation game says nothing about whether a system is safe, accurate, fair, or fit for a given use, which is what Canadian AI oversight actually asks about. That gap is visible in the machinery Canada’s own privacy regulator points to: a footnote in the federal, provincial and territorial privacy commissioners’ generative-AI principles states, “For more information on validity and reliability in AI systems, see the NIST AI Risk Management Framework” — a named, measurable standard, not a pass/fail conversation.
Suppose a company wants to know whether its new customer-service chatbot is “good enough.” A Turing-style test would only ask whether callers can tell it apart from a human — which a scripted, shallow bot can sometimes achieve on narrow small talk without being remotely reliable on an actual account problem.
The framework NIST publishes instead asks a different, more useful set of questions, naming “valid and reliable, safe, secure and resilient, accountable and transparent, explainable and interpretable, privacy-enhanced, and fair with harmful bias managed” among the properties a system should have. A chatbot can fail every one of those while still fooling a caller for thirty seconds, which is precisely why a passed imitation game and a trustworthy system are not the same finding.
See also: machine learning, artificial general intelligence, narrow AI.
Deciding what “good enough” actually means for a specific AI use case — before a single line of it is built — is a strategy question, not an engineering one; ai-strategy-roadmapping covers how that gets defined up front.
