Artificial intelligence has taken another leap forward with ‘Watson’, a robot that can beat humans on game shows. But how will he get on in the world of work? Emily Hohler reports.
Who, or what, is Watson?
Watson, named after IBM’s founder, Thomas J Watson, is the result of a four-year project. More than 20 researchers set out to build the smartest machine on earth: one that could answer questions posed in conversational, or ‘natural’ language. Watson is the size of ten refrigerators with 90 servers and 2,880 core processors each able to perform up to 33 billion operations a second. His database holds about 200 million pages of information. IBM refuses to say how much he has cost. Industry insiders put it at anywhere between $100m and $2bn. Watson recently beat two contestants on the US game show Jeopardy!
What is so special about him?
Building a machine that is able to understand a question posed by a person and respond with a precise, factual answer has long been the holy grail of artificial intelligence (AI) research. In 1997, IBM’s supercomputer Deep Blue famously beat the grandmaster Garry Kasparov at chess. But chess is a perfectly logical game and can easily be reduced to maths, which computers handle brilliantly. Language is much more slippery. Jeopardy! contestants need not only encyclopaedic knowledge, but the ability to unravel convoluted clues, often involving word play. In a reverse of the usual quiz formula, contestants are given an answer and have to provide the question.
How does Watson do it?
Watson’s extraordinary speed allows him to use more than a hundred algorithms simultaneously to analyse a question thousands of different ways, generating hundreds of possible solutions. Another set of algorithms ranks these according to plausibility. Once dozens working in different directions all arrive at the same answer, the probability of it being correct increases. But Watson can still appear stupid. For instance, he would struggle with the clue, ‘Look in this direction and you’ll see the wainscoting’, because the answer, ‘What is down?’, is rooted in human social experience. That said, Watson is more than just an expensive publicity stunt. The problem with Deep Blue was that its expertise didn’t translate to real-world business problems and therefore did little for IBM’s bottom line. Cutting the time it takes to answer very difficult questions has obvious appeal. Legal firms, for example, need to sift through case law fast to find a useful precedent or citation. Help-desk workers have to negotiate enormous databases to find answers for irate customers. IBM is already looking at specific applications in medicine and healthcare, help desk, tech support and business intelligence.
So what next?
IBM announced partnership programmes with eight American universities last week, meant in part to explore applications for this new technology. The firm plans to collaborate with Columbia University and the University of Maryland to create a physician’s assistant and with Nuance Communications to add voice recognition. This service could be available in as little as 18 months. There are high hopes. A patient’s symptoms, medical history, physical exam findings and laboratory results present clues that must be synthesised into a diagnosis. Watson would avoid the common trap of fixing on a familiar diagnosis and wrongly attributing symptoms to a single cause.
When might we see a Watson in day-to-day use?
Watson is huge and needs to run on at least one $1m IBM server. But John Kelly, head of IBM’s research laboratories, predicts that within ten years a Watson could run on a server that any small firm could afford, and a few years after that on a laptop. The idea isn’t so far-fetched: the average mobile phone is about a millionth the size and a millionth the price of a computer from the early 1970s. Plus it’s far more powerful.
Should we be afraid?
Humans have always adjusted as machines have taken over chores. Yet such advances inevitably provoke fears. AI systems are not infallible. ‘Millisecond trading’ computers on Wall Street firms may have been to blame for last May’s wild stockmarket plunge. Would doctors be keen to act on a split-second answer from a Watson? If they ignore him, how would it look in a negligence claim? Ultimately, as scientist Stephen Wolfram (inventor of the Wolfram-Alpha search engine) says, Watson may be good with facts, but he lacks two vital human qualities: wisdom and judgement.
Are machines now cleverer than us?
Not yet – Watson excels within very limited parameters. He cannot come up with new ideas, make jokes, or have a conversation. And he’s a long way from passing the ‘Turing test’, whereby a machine could not be distinguished as non-human from its replies. Nevertheless, there are those who believe that that day is not far off: 2029, according to Ray Kurzweil, the American entrepreneur, inventor and scientist. Kurzweil argues that because computing power has proven to grow exponentially rather than linearly (ie, computers get faster more rapidly), by 2045 they will be more intelligent than humans. This point is known as ‘the Singularity’. What would happen next is unknowable because if we could predict the behaviour of these cyborgs, we’d be as clever as them. Although a lot of people think the theory is nonsense, last year’s Singularity summit attracted computer scientists, neuroscientists, nanotechnologists and molecular biologists. Watson is a long way from what’s known as “strong artificial intelligence”. But he may represent a step on that road.