YOUR HORSE IS stuck in a ditch. You’ve tried luring it out with a sugar lump, but it’s not budging. Whipping out the smartphone, thumbs shaking, you Google “Annie”, the RSPCA’s virtual assistant. Reassured, you see Annie has handled this sort of thing before.
“My horse has fallen down a ditch, what should I do?” is a Frequently Asked Question. A human operator might have offended by asking how your horse got there, whether you’d tried the sugar lump trick or some commentary from the Grand National? But Annie plays it safe: “Contact the RSPCA’s 24-hour cruelty and advice line immediately.” Why didn’t you think of that?
Chatbots such as Annie can only answer questions they’ve been scripted to answer. More sophisticated digital helplines can be part human, part automated, so you know your particular crisis has been understood and the advice given is appropriate. One day, though, fewer human ghosts will be needed in these machines.
David Levy is the organiser of this year’s Loebner Prize competition for the most human-like conversational computer, or chatbot. “The point is simply to create the ability in software to conduct an interesting, entertaining and informative human-like conversation which could be used in all sorts of different environments – commercial environments, social services environments, and just as entertainment,” says Loebner.
For the moment, this is something of a pipe dream – as I found out as one of this year’s Loebner Prize judges. It’s an annual competition based on a problem posed in 1950 by the great mathematician and second World War codebreaker Alan Turing. Many regard Turing as the father of Artificial Intelligence (AI). Can a machine think?
Turing suggested a test for a thinking machine. If, in a blinded, text-based conversation, a machine could convince a human judge that it too was human, then it could be considered a thinking machine. And so the famous Turing Test was born.
For 20 years, Hugh Loebner, a Hawaiian-shirted American philanthropist, has been footing the bill for a Turing Test-based competition. This year, to mark the centenary of Turing’s birth, it was held in the cloistered, wood-panelled confines of Bletchley Park, where Turing and his colleagues cracked the German Enigma code during the second World War.
Over two hours, we judges were to pit our wits against four different pairs of anonymous humans chatbots. Could we spot the difference?
You could almost hear the drum roll as we sat poised at our computers and, to a countdown Nasa would be proud of, launched into action. Judges initiated conversations. After that it was a free-for-all. We quickly realised the chatbots stuck out a mile and by Round three, I was getting silly …
Conversation A
Me: “Greetings earthling.”
Response: “Nanu nanu.”
Conversation B
Me: “Why did the chicken cross the road?”
Response: “I give up. Why?”
You might think that “nanu nanu” is an odd utterance for a human, but it had me laughing out loud, not something you’d expect from a chatbot. However, even a humourless human would surely know why the chicken crossed the road, wouldn’t they? B was clearly the chatbot. B was this year’s Loebner Prize winner, Chip Vivant.
Mohan Embar, Chip Vivant’s programmer, later tipped that if you’re ever in doubt about whether your online correspondent is human try asking this simple question: “where is the nose on your face? A human will think of the three-word answer (in the middle), but computers, even those with on-screen avatar faces, will never give this response.
Chatbots don’t actually understand us. They just pretend to, answering our questions based on sample human exchanges in their database. The bigger the database, the better they are at responding appropriately (some mine Twitter for this). The humanity of their answers and their ability to deflect or clarify difficult questions depends on the skill of their programmers.
Digital helplines are their most obvious application. But if they could be more human, would we prefer talking to them?
Surprisingly, not always. A 2007 survey by Callcentres.netfound that 67 per cent of Australians would rather deal with an Aussie-accented speech-recognition system than an offshore human call-centre agent.
If my experience judging this year’s Loebner Prize is anything to go by, humanity’s unique claim on thinking is safe for now. Like my three fellow judges (another journalist, an accountant and an archaeology professor), I concluded that a typing Orang Utan, like The Jungle Book’s King Louie could “be like me” more effectively than any of this year’s finalists.
One chatbot annoyed with a cacophony of questions. Another seemed rude and aggressive (a ruse used to stop difficult conversations mid-flow and revert to a more familiar subject). A third had the kind of verbal diarrhoea you dread from a dinner party guest.
None of the above is ideal in a helpline chatbot. But Chip Vivant, the winner, seemed to have potential and an important message from its maker – forget about apeing humans; be yourself; people will love you for it. Customers will love you for it. Given the current state of technology, Chip’s winning approach makes sound commercial sense.
Digital helplines must be fast and effective to satisfy customers. For now, says Erwin van Lun, founder and chief executive of chatbots.org(an online community of chatbot developers, academics and users), the human element is unnecessarily costly and risks doing more harm than good. “If your computer breaks, you want to know how to fix it. You don’t want to read something like ‘Your computer is broken? That’s a pity for you. Everything breaks sometime’.”
The most effective digital helpers are industry-specific and specialised, says van Lun. Few could pass the Turing Test, and whether or not they have a human face depends on the company’s branding strategy.
Anna, Ikea’s virtual assistant, has a face. So does Jon, Morrison’s straw-boatered virtual fishmonger. But Siri, the iPhone’s digital helper, doesn’t. And Quark, Creative Virtual’s “v-assistant” looks like an animated motorbike helmet.
And although in Ireland organisations seem slow to embrace automated helplines, they are bound to appear soon on a website near you.
A recent survey of the world’s top 1,500 companies by Gartner Research predicted that by 2015 half of all online self-search activities will be carried out by chatbots or “virtual assistants”. Company benefits lie in slashing 5 per cent off the cost of customer service, building brand loyalty and giving customers a bit of fun through playing with a robot.
Building charm into your chatbot can have customers coming back for more.
I found chatting to Cleverbot quite addictive. This is a common response, says its developer Roll Carpenter, twice winner of the Loebner Prize (2005 and 2006). Last year, Cleverbot won a Turing Test at the Techniche Festival in India, achieving a score of 59.3 per cent human.
Since then, Carpenter says, millions of people talk to Cleverbot every month for entertainment – the longest conversation on record lasting 11 hours.
“Despite it clearly being a bot,” people do regularly become convinced otherwise. Its because Cleverbot learns from people and imitates them … and has a huge variety of responses.”
Through his company Existor, Carpenter has two new projects on the go, a virtual assistant for a Russian bank and his cutest chatbot yet, Pupito, an animated puppy app for iOS. Pupito yaps and talks in text in response to typed input or, for a small fee, will chat with its “owner” in what Carpenter describes as puppy-like language (“Me is extra-specially clever, you knows!”).
But digital pets could have loftier aims than passing the time on the morning commute. They could be companions for people who, for some reason, can’t have the real thing. Elderly people in care homes, for instance.
Comforting the lonely is what Mohan Embar is all about. The US-born Indian programmer of this year’s Loebner Prize winner says that ELIZA is his inspiration. ELIZA was one of the worlds first ever chatbots, developed in the mid-1960s at MIT. It’s DOCTOR script simulated a form of psychotherapy which was so effective it completely fooled some of its “patients”.
“Imagine what we could do to provide comfort to people, using today’s technology,” says Embar. “That is my goal in life … Elderly people and those living in isolation, unable to obtain normal human interaction, could be helped by the “synthetic comfort” that an artificial agent could provide. I don’t know how rich this idea would make me, but the idea of being able to make an extension of myself that provides comfort to a lot of people is very appealing to me.”
And wouldn’t a bit of digital comfort be useful on a customer helpline, too?
No comments:
Post a Comment