The drive towards artificial intelligence continues apace, with everything from smart cars to server malware protection implementing the technology in some form or other – but has any computer managed to pass Alan Turing’s famous test?
Developed by famed mathematician Alan Turing back in 1951, the “Turing test” is essentially an evaluation of a machine’s ability to exhibit intelligent behaviour equivalent to, or indistinguishable from, that of a human. As it remains stubbornly difficult to naturally synthesize audio to sound accurately like human speech, the test has come to be a challenge for chatbot developers that work via text – so can bots mimic humans via text chat to convince human judges they might in fact be real people?
In 2014, a chatbot named “Eugene Goostman”, posing as a 13-year-old Ukrainian boy, managed to convince a third of the judges that he was human after a brief two-way conversation and sparked a flurry of press coverage that the Turing test had been finally beaten.
However, the claim certainly had some detractors and Goostman was not even the first chatbot to “pass” the test. Back in 2011, Cleverbot managed to convince 59.3% of judges that it was real by automatically searching through previous conversations it had had with real people and trying to re-use the closest answers.
The question that comes from these claims of success, whether you can trust the non-peer-reviewed results or not, has more to do with how our ideas about AI and machine learning have evolved over the last 70 years, from a time when it was all theoretical to the current technological situation where the cloud offers huge computing power and number crunching on-demand. The idea that a machine could convince anyone it could be a person would have been thrilling to Alan Turing, but now that they can (or nearly can), do we still think that this test is sufficient to imply “intelligence”? And anyway, with all the talk of Russian bots on social media, could you pass the Turing test? And does it matter?
Technologists are right when they say: “artificial intelligence is changing the playing field for businesses and they must remain current in technology or be left behind”, but as the technology develops we need to redefine what we think of as intelligence. Is convincing a person you are real a sign of intelligence, or does it just show a developer has cleverly mapped out how most conversations tend to go? Maybe the Turing test needs an update for the internet age.
Photograph by RoboCup