Turing test
Alan Turing's 1950 thought experiment: if a machine can fool a human into thinking it's human, does it think?
Think of it like
A courtroom where the judge questions both a human and a hidden machine and must guess which is which.
Example
Early chatbot ELIZA (1960s) often fooled people in text because they read human intent into minimal responses; people wanted to believe it understood them.
How it actually works
Turing proposed the test as a practical alternative to answering "can machines think?" — sidestep philosophy, just check: if it behaves indistinguishably from a human, that's what matters. The test is surprisingly easy to game (you don't need real understanding, just good mimicry), and modern LLMs pass it trivially — yet we don't say they "think" the way we do. The test reveals a gap between behavioral equivalence and actual understanding.
For product teams
A famous benchmark that sounded profound but turned out to measure mimicry more than intelligence.
For engineers
A benchmark test where a model must sustain human-like conversation to convince a human judge; easily passed by modern LLMs through pattern matching, not reasoning.
Related
- AI Safety — A foundational question in AI philosophy.
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome