on 25 September 2009As I write, artificial intelligence (AI) teams are gathering in Brighton for the 2009 Loebner Prize Contest. This is where chatbots – programs that simulate conversation – compete to pass the Turing Test, convincing a panel of human beings that they’re talking to a real person. Despite the incentive of $100,000, no system has actually won so far, though presumably they all have tearful acceptance speech modules just in case.
The Loebner is a bit like those tests they use to find out if people are telepathic: the rules are painstakingly stretched to make it as easy as possible to pass, but still nobody does. For example, part of a session last year went like this:
HUMAN: Have you any plans for later in the day?
COMPUTER: This is confusing. Why bring the waking hours into this?
And this was the top entry. Second place went to David Beckham.
The trouble with the Turing Test is that it’s based on a daft premise. Even if a computer could think, why would it chat like a human being? Which human being, anyway? Brahmagupta? Chaucer? Milo from the Tweenies? If aliens travelled to Earth in a faster-than-light craft from a far-distant galaxy, would we conclude from their inappropriate responses to our pleasantries about last night’s telly that they were incapable of rational thought? Conversely, a case-based reasoning program clever enough to fool someone for five minutes of aimless banter would hardly mark the advent of the Singularity, although it might get a slot on This Morning.
The Terminator – the original Arnie model, not the cute teenage ones they have now – wasn’t hot on conversation. Like most chatbots, he relied on repeating input back to the speaker or picking from a list of responses such as YES/NO, PLEASE COME BACK LATER and F*** OFF. This probably wouldn’t get him through the first round in Brighton, although the whole naked leather-fetish bodybuilder angle might not go amiss locally.
What he was good at was ruthlessness. This worries Noel Sharkey, professor of artificial intelligence and robotics at the University of Sheffield. As he recently told New Scientist, Dr Sharkey thinks we’re too easily suckered into believing computers are becoming human. It’s a live issue, because armed robots aren’t science fiction any more: the US regularly sends drone aircraft to assassinate foreigners, and the South Koreans are deploying automated sentries along the border with the North. These are basically remote-control toys, but it seems inevitable that a degree of autonomy will creep in. Once you have a system that can track an enemy until an operator’s button-press clears it to open fire, how hard is it to make the argument that national security or the lives of personnel shouldn’t rest on that button-press?
Yet, as Sharkey points out, ‘there is no way for any AI system to discriminate between a combatant and an innocent. Claims that such a system is coming soon are unsupportable and irresponsible.’ Even if we built in Asimov’s Three Laws of Robotics to keep our cybernetic chums on the moral straight and narrow, we couldn’t rely on their judgement. In fact, the original point of the Three Laws was to generate stories about their inadequacies. Decisions on human issues can only be made meaningfully by humans.
Rather than trying to get computers to think for themselves, it’s probably more useful to make them better at doing what they’re told. One of the major contestants in that field is Apple, whose multi-touch interfaces – soon to reach their apotheosis, we’re led to hope, in the much-rumoured Mac tablet – seem likely to outsmart the mouse and keyboard. But the touchscreen is just another kludge. Why should we have to interact physically at all?
Neural interfacing has been somewhat trivialised by chip-implanting experiments that turn living flesh into a garage door opener remote (yes, you, Kevin Warwick). The possibilities are much more interesting than that. Cognitive science has shown that our awareness of making decisions lags behind the actual processes that guide our actions, implying that brain-fed interfaces could respond to choices before we make them. By the time we were ready to notice we’d decided to click something, we’d already be reacting to the result. Instead of struggling to type as fast as we can think, we might type faster than we can think.
Is Apple already working towards this next generation of human-computer interfaces? We don’t know, because the company ‘doesn’t comment on unannounced products’. But I really hope so. Otherwise it could be left to Microsoft, and that’s a vision that would keep even Sarah Connor awake at night.
Adam Banks once asked a chatbot for a date, and she gave him 1 January 1904.
adambanksdotcom @Herring1967 "watching on screen"?! That's not a subeditor, it's a butcher with English as a third language 
