The Turing test: Can a computer pass for a human? - Alex Gendler

Summary

Alan Turing proposed judging machine intelligence by whether a computer could hold a text conversation that a human evaluator could not readily distinguish from one with a person. In his 1950 paper Computing Machinery and Intelligence, he described a test in which a judge converses with unseen participants. A machine passes if replacing one participant with it does not substantially change the judge’s results.

Turing expected machines to pass by 2000, but increased memory and computing power have not been enough. Some early chat programs instead steered conversations in ways that encouraged people to mistake them for intelligent:

• ELIZA imitated a psychologist, prompting people to talk and reflecting their questions back to them. It was not put through a formal test.
• PARRY imitated a paranoid person by repeatedly returning to programmed concerns.
• In the Loebner Prize competitions, programs such as Catherine and Eugene Goostman used narrow conversational personas that could lead judges to interpret limited or awkward responses favorably.
• Cleverbot used statistical analysis of large collections of conversations and could retain memories of past chats, but it lacked a consistent personality and struggled with new topics.

These examples reveal a weakness in conversational testing: people can attribute intelligence to responses that are scripted or limited. Human conversation also depends on complex language understanding, contextual knowledge, and intuition. Small hesitations, ambiguous questions, and ordinary sentences can challenge chatbots, even though computers can perform other demanding tasks.

The video concludes that passing as human requires more than increasing memory and processing power—and that doing so may bring questions about consciousness back into focus.

Reviews, Guides, or Tutorials

None presented; the video is an educational explanation of the Turing test and chatbot examples.

Main Speaker and Source

Alex Gendler, in a TED-Ed lesson.