Expert Comment: Has a computer passed The Turing Test?Thursday 12 June 2014
Following claims last week that the Turing Test has been passed for the first time Dr David Reid, Senior Lecturer in Computer Science, looks at the truth behind the headlines.
Just over 60 years ago (7th June 1954) Alan Turing, considered to be the father of theoretical Computer Science and AI, committed suicide. Persecuted for his homosexuality by society he was stripped of his security clearance from GCHQ and threatened with prison unless he accepted injections of oestrogen for a year in order to “cure” him. Many speculated that Turing, a quiet and nervous man (he had a pronounced stammer), could take the strain no longer and committed suicide by biting an apple laced with cyanide, others maintain it was an effect of working with dangerous laboratory apparatus that killed him. What is undisputable is that he left behind a vast and glittering array of academic work that was sublime in its genius.
Perhaps the most widely known work is the “Turing Test”. His 1950 paper "Computing Machinery and Intelligence," which opens with the words: "I propose to consider the question, 'Can machines think?'". Because "thinking" is difficult to both define and measure Turing decided to err on the pragmatic and question whether a computer could be created that would respond to general purpose questions in such a way as it would be indistinguishable to a human response. In other words to ask whether can a computer imitate a human? Turing believed that this test would be the first stage of proving that the brain is essentially a computer and that a machine can have a mind. A reverse form of this test (to test if you are talking to human rather than a software bot) is widely used on the internet (CAPITA).
Early this week in the Royal Society in London the University of Reading organised a Turing Test competition. For the first time ever a chatterbot called “Eugine Goostman” passed this test. It imitated a Ukrainian teenager effectively enough to convince judges that they were talking to a real person in a series of online chats. This program works in a fairly traditional way, it analyses questions it receives, by searching a tree like knowledge base for material before compiling a response. Occasionally it will ask a clarifying question, or draw on a response from memory. A similar system that took part in the competition can be accessed here.
However, this achievement has already been criticised: the judges only had five minutes to ask questions; imitating a 13 Ukrainian boy who learned English as a second language and has limited knowledge of the world is a cheat; that the press have reported the winner as a “supercomputer”- it is not; the test itself is flawed as a test of artificial intelligence.
Certainly some of these objections are valid, it could be argued that a more impressive Turing Test like winner was in 2011 in the USA when the Watson supercomputer won “Jeopardy!” playing against previous winners of the TV show; the AI in this computer has recently been repurposed to for management decisions in lung cancer treatment, to diagnose diseases, to produce new drugs, to make financial decisions, to help individual healthcare and perform fundamental research in . It has even learnt to cook, make cocktails, and debate ethical issues.
However, despite the backlash, let’s be clear, this undeniably a great technical achievement. Each time a milestone is reached in AI the goalposts are moved. A few years ago it was said that a computer could never beat a Grand Master in chess (the Deep Blue computer beat Gary Kasparov in 1997, that computers would never understand natural language- Microsoft have bought Skype and just announced a real time translation service, more recently many said intelligence should be defined not by rigid calculation or playing well defined games but by the ability to simply cross the road; Google have now developed a car that can not only cross the road but can drive down it, and Boston Dynamics (recently bought by Google) have developed battlefield robots that are far more lifelike in the way they interact with the environment.
Both the US and Europe governments are currently launching in multi-billion dollar research in trying to completely mimic in precise cellular detail the components of a mammalian brain.
What is equally interesting is not the news about the Turing Test being passed but the reaction from the general public/press about this news and advances in AI in general.
Some see AI as a potential global threat to humanity, as an indifferent primordial force of nature, some feel that AI isn’t possible and that thinking is a uniquely human activity, in a real sense that it defines what we are as a species. These worries are expressed in the new Hollywood blockbuster “Transcendence” out in cinemas at the moment. The film considers the notion of what that other great pioneer of Computer Science, John Von Neuman, called the technological singularity: a theoretical point when AI surpasses human intelligence. It also ponders the notion of Transhumanism (or H+) a movement which encourages technologies to greatly enhance human intellectual, physical, and psychological capacities (the robotic exoskeleton in the World Cup opening ceremony is one aspect of this). Professor Steven Hawking has recently expressed his concerns in the Independent newspaper stating that: "Success in creating AI would be the biggest event in human history. Unfortunately, it might also be the last."
Whether or not it is correct to fear or welcome AI, whether or not you believe a machine can ever think or not what is clear is that AI if/when it comes will be: different, it may be a gradual event rather than a “singularity” event, it may be achieved on a number of individual fronts and it may be for both good and bad.
What is certain whatever happens AI will ultimately open up just as many moral questions about our society as Turing’s death did in 1954.