AI looks beyond the Turing test
Add Axios as your preferred source to
see more of our stories on Google.
Alan Turing and colleagues working on the Ferranti Mark I Computer, 1951. Photo: SSPL/Getty Images
In 1950, Alan Turing, whose 106th birthday would have been yesterday, laid out a test that is known in popular culture today as the gold standard for evaluating AI.
What's happening now: As AI researchers notch successes in processing language, playing games, and recognizing images, some of the field's heavyweights are calling for a wider approach to creating intelligent machines and expanding how it is tested.
Several chatbots have claimed to "pass" Turing's test by convincing some proportion of human judges that they are human, too.
- In 2014, for example, a chatbot named Eugene Goostman managed to persuade 10 out of 30 judges that it was a 13-year-old Ukranian boy.
- But Goostman isn't truly intelligent by any stretch; it just managed to play a few tricks to convince judges that its conversational shortcomings were due to age and linguistic ability.
The Turing test captures some dimensions of human intelligence, like emotional fluency and natural language. But how else can intelligence be measured? It's a complicated question, says Jon Crowcroft from the Alan Turing Institute in London. Human intelligence encompasses intuition and common sense, which is informed by our social norms, laws — and the list goes on.
"The Turing test is reductionist — it removes consciousness, emotions and also our rich, constantly changing social structure. It leaves out a bunch of things that make humans human."— Jon Crowcroft
Yesterday's tomorrow: Turing's experiments set off wild optimism as well as fear in the 1940s, writes Diane Proudfoot, a professor at the University of Canterbury in New Zealand and co-director of the Turing Center at ETH Zurich.
- Turing made fun of both camps, Proudfoot writes. Mocking the fearmongers, Turing said: "A similar danger and humiliation threatens us from the possibility that we might be superseded by the pig or the rat."
- She argues that voices promising AI-driven utopias or warning of the existential dangers of super-intelligent AI today are repeating age-old hysterias. "Turing was a voice of sanity" in the 1940s, she writes, but she says similarly moderate opinions risk being drowned out today.
Go deeper:
- Proudfoot's blog post on Turing has great details about the popular reaction to AI in the 1940s and 1950s.
- AI researcher Filip Piekniewski explains the limits of Turing's test. (h/t Azeem Azhar)
