What’s clearly needed is an alternative to the Turing Test. Something less likely to become a laughing stock. Something like laughter, perhaps?
7 Dec 2015 by qwbdchbctt.d qwbdchbctt.d
As the UK languishes in the supercomputing doldrums, Russ Swan asks: Are we having a laugh? If not – then maybe we should be…
As the UK languishes in the supercomputing doldrums, Russ Swan asks: Are we having a laugh? If not – then maybe we should be…
Whatever happened to artificial intelligence? There was a time, not so long ago that the imminence of sentient silicon was a hot topic of conversation almost everywhere. Various predictions for the moment of singularity, when machines reached parity with human consciousness, came and went, yet nothing happened. What went wrong?
The subject of AI is inextricably linked with that of high performance computing (HPC) – or supercomputers in everyday language. Where the one is the processes and software, the other is the hardware and machinery. There can be no synthetic mind without an artificial brain in which to house it.
Now, the world of supercomputers is a bit odd. Around the globe, various nations and national agencies compete to create the fastest and most powerful calculating machine on the planet, in the certain knowledge that after a couple of years it will be surpassed by a rival, and within a decade or two it will be a quaint museum piece.
You’ve heard the aphorism that there is more computing power in your phone than took the Apollo astronauts to the moon. That doesn’t even tell half the story. Today there’s probably more processing power in your Visa card.
The first supercomputer to really gain widespread fame was the Cray-2 of the mid 1980s, thanks to prominent exposure in the Guinness Book of Records. This awesome machine had a peak performance of about 1.9 Gflops – a gigaflop being a billion floating point operations per second.
A decade later in the mid-90s, after a period of particularly intense progress which even exceeded the predictions of Moore’s Law, the bar was raised to the 300-400 Gflops range. How good is that today? You know that tablet computer, the one you play music on and use to share pictures of kittens? That’s a Cray-2 in your hands, right there. Good to know that all that incredible computational power is being used for the greater benefit of mankind, isn’t it?
Incidentally, for a nation that was once a colossus of intellect and leading scientific endeavour, the UK seems to have given up in the supercomputer race. The Top500 list is compiled twice a year to provide a league table of the world’s most advanced processing hardware, and according to the latest list the highest performing machine in the UK ranks just 41st in the world. That is behind machinery from China, USA, Japan, Switzerland, Germany, Saudi Arabia, Australia, Italy, Korea, France, Russia, and Poland.
The computational hardware is only the beginning, of course. Anything even vaguely like AI will require reams of sophisticated code in order to emulate (or replicate?) consciousness and intellect. Despite the incredible advances being made, this moment still seems a long way off.
The Turing Test has long been regarded as the benchmark for this, and a chatbot program called Eugene Goostman was the first to ‘pass’ this test, at an overhyped event at Reading University last year. The conversational algorithm succeeded in fooling one-third of a panel of human judges into believing it was a human, specifically a 13 year-old boy from Ukraine. In fact, it really only showed how poor the test was, or perhaps how gullible the judges were. Eugene’s online version was swamped with interest, first going offline through too much demand, and then being unceremoniously switched off when it was universally ridiculed.
The engine management system fitted to VW diesels showed enough intelligence to tell lies, and to change those lies according to context. I’d say that was more intelligent than young master Goostman.
What’s clearly needed is an alternative to the Turing Test. Something less likely to become a laughing stock. Something like laughter, perhaps?
It’s a popular interview question – “tell me a joke” – and interviewees are well advised to always have one or two up their sleeve. Could a computer tell a joke, or better still invent one? This would be a real indication that it was beginning to understand context and display creativity, which sounds a bit like intelligence to me.
And now, suddenly, it gets interesting. A couple of scientists at Northwestern University in the USA recently pitted their own AI program, called Manatee, against a human stand-up comic called Myq Kaplan. At least one of those must be an acronym, but I’m not quite sure which one.
They set both the challenge to come up with a new gag on the template of ‘I like my X like I like my Y’, and for my money the program won the battle. Kaplan’s output included “I like my jokes like I like my robots… efficient and killer”. Meanwhile the joke bot came up a couple of corkers, including the surprisingly rude “I like fingers like I like notes…sticky”, and “I like my thrills like I like my flights…cheap”.
I, for one, welcome our new silicon overlords.