A sculpture of Alan Turing at Bletchley Park by Stephen Kettle. Photo: Steve Parker / Flickr
Show Hide image

Supercomputer passes Turing Test by convincing judges it’s a 13-year-old Ukrainian boy

A computer programme has succeeded in passing the Turing Test, 65 years after it was first conceived of by the father of artificial intelligence, Alan Turing.

Wartime cryptographer Alan Turing’s iconic question – “can machines think?” – was put to the test once again at the Royal Society’s Turing Test 2014 competition in London. Supercomputer “Eugene Goostman” managed to fool 33 per cent of judges into thinking it’s a human.

The test, proposed by mathematician Alan Turing in his 1950 paper Computing Machinery and Intelligence, understands its limitations. Defining “think” is not an easy task. He instead replaced the question with something more tangible – can a computer successfully convince an observer that it’s human?

Eugene can do just that. In a five-minute question-and-answer text chat – with no limitations of topic – a third of judges believed the program to be a real human. Until now, no computer has managed to reach the 30 per cent benchmark set by Turing.

The chatbot, brainchild of Russian computer scientist Vladimir Veselov, has an important advantage over its (/his?) competitors – his “personality” is a 13-year-old Ukrainian boy. After winning the competition Veselov explained his team’s intentions. “Eugene was ‘born’ in 2001,” he said. “Our main idea was that he can claim that he knows anything, but his age also makes it perfectly reasonable that he doesn’t know everything. We spent a lot of time developing a character with a believable personality.”

Subject-specific knowledge is rarely a strong point for 13-year-olds, and Eugene has just enough for a brief chat about a wide range of topics. In addition, the bot’s grammatical errors can be put down to speaking English as a second language. With these caveats in mind, it’s perhaps unsurprising that it managed to convince the judges.

The achievement was hailed as a landmark. Roboticist and cybernetics researcher Kevin Warwick, of the University of Reading, which organises the competition, said: “There is no more iconic and controversial milestone than the Turing Test... This milestone will go down in history as one of the most exciting.”

The practical implications of this are ominous. Face-to-face conversations are being progressively replaced by social media – Channel 4 found that the average Briton will text friends and family more regularly than see them face-to-face. If the move to digital media is accompanied by increasingly sophisticated computers, then we need to be sure of who we’re talking to.

Warwick warned: “Having a computer that can trick a human into thinking that someone, or even something, is a person we trust is a wake-up call to cybercrime. The Turing Test is a vital tool for combating that threat.”

Though Matrix-style scenarios of machine domination are still a long way off, it doesn’t take much to imagine the potential for misuse of such machines. Already our online presences give away a huge part of our personalities. A future version of Eugene could plausibly analyse our social media profiles en masse and conjure up a passable imitation of a loved one – enough, say, to start sending requests for pin numbers and passwords.

Getty
Show Hide image

Don’t shoot the messenger: are social media giants really “consciously failing” to tackle extremism?

MPs today accused social media companies of failing to combat terrorism, but just how accurate is this claim? 

Today’s home affairs committee report, which said that internet giants such as Twitter, Facebook, and YouTube are “consciously failing” to combat extremism, was criticised by terrorism experts almost immediately.

“Blaming Facebook, Google or Twitter for this phenomenon is quite simplistic, and I'd even say misleading,” Professor Peter Neumann, an expert on radicalisation from Kings College London, told the BBC.

“Social media companies are doing a lot more now than they used to - no doubt because of public pressure,” he went on. The report, however, labels the 14 million videos Google have removed in the last two years, and the 125,000 accounts Twitter has suspended in the last one, a “drop in the ocean”.

It didn’t take long for the sites involved to refute the claims, which follow a 12-month inquiry on radicalisation. A Facebook spokesperson said they deal “swiftly and robustly with reports of terrorism-related content”, whilst YouTube said they take their role in combating the spread of extremism “very seriously”. This time last week, Twitter announced that they’d suspended 235,000 accounts for promoting terrorism in the last six months, which is incidentally after the committee stopped counting in February.

When it comes to numbers, it’s difficult to determine what is and isn’t enough. There is no magical number of Terrorists On The Internet that experts can compare the number of deletions to. But it’s also important to judge the companies’ efforts within the realm of what is actually possible.

“The argument is that because Facebook and Twitter are very good at taking down copyright claims they should be better at tackling extremism,” says Jamie Bartlett, Director of the Centre for the Analysis of Social Media at Demos.

“But in those cases you are given a hashed file by the copyright holder and they say: ‘Find this file on your database and remove it please’. This is very different from extremism. You’re talking about complicated nuanced linguistic patterns each of which are usually unique, and are very hard for an algorithm to determine.”

Bartlett explains that a large team of people would have to work on building this algorithm by trawling through cases of extremist language, which, as Thangam Debonnaire learned this month, even humans can struggle to identify.  

“The problem is when you’re dealing with linguistic patterns even the best algorithms work at 70 per cent accuracy. You’d have so many false positives, and you’d end up needing to have another huge team of people that would be checking all of it. It’s such a much harder task than people think.”

Finding and deleting terrorist content is also only half of the battle. When it comes to videos and images, thousands of people could have downloaded them before they were deleted. During his research, Bartlett has also discovered that when one extremist account is deleted, another inevitably pops up in its place.

“Censorship is close to impossible,” he wrote in a Medium post in February. “I’ve been taking a look at how ISIL are using Twitter. I found one user name, @xcxcx162, who had no less than twenty-one versions of his name, all lined up and ready to use (@xcxcx1627; @xcxcx1628, @xcxcx1629, and so on).”

Beneath all this, there might be another, fundamental flaw in the report’s assumptions. Demos argue that there is no firm evidence that online material actually radicalises people, and that much of the material extremists view and share is often from mainstream news outlets.

But even if total censorship was possible, that doesn’t necessarily make it desirable. Bartlett argues that deleting extreme content would diminish our critical faculties, and that exposing people to it allows them to see for themselves that terrorists are “narcissistic, murderous, thuggish, irreligious brutes.” Complete censorship would also ruin social media for innocent people.

“All the big social media platforms operate on a very important principal, which is that they are not responsible for the content that is placed on their platforms,” he says. “It rests with the user because if they were legally responsible for everything that’s on their platform – and this is a legal ruling in the US – they would have to check every single thing before it was posted. Given that Facebook deals with billions of posts a day that would be the end of the entire social media infrastructure.

“That’s the kind of trade off we’d be talking about here. The benefits of those platforms are considerable and you’d be punishing a lot of innocent people.”

No one is denying that social media companies should do as much as they can to tackle terrorism. Bartlett thinks that platforms can do more to remove information under warrant or hand over data when the police require it, and making online policing 24/7 is an important development “because terrorists do not work 9 to 5”. At the end of the day, however, it’s important for the government to accept technological limitations.

“Censorship of the internet is only going to get harder and harder,” he says. “Our best hope is that people are critical and discerning and that is where I would like the effort to be.” 

Amelia Tait is a technology and digital culture writer at the New Statesman.