Alan Turing's 1950 imitation game asks a judge to identify which of two conversational partners is the machine. If the judge cannot reliably tell, the machine passes. It was a clever operational substitute for an unanswerable question, and it has been widely misread as a definition of intelligence ever since.
Why it is not a measure of AGI
The test measures conversational mimicry under time pressure. It does not test learning, planning, memory, or transfer. A system could pass by being excellent at being human-sounding while being unable to do any of the things general intelligence is supposed to do.
Why it has lost its diagnostic value
Modern language models pass casual versions routinely, which tells us more about the test than the models. Turing's own prediction was that the question would eventually seem meaningless, and that is roughly where the field has arrived.
What replaced it
Benchmarks that measure specific capabilities under controlled conditions, plus the harder and less formal test of whether a system can complete a long, messy real task without supervision.