Friday, May 27, 2016

Smarter than the best brains: IBM builds a new kind of genius

Final Jeopardy: Man vs. Machine and the Quest to Know EverythingFinal Jeopardy: Man vs. Machine and the Quest to Know Everything by Stephen Baker

It’s more than five years since a computer called Watson beat two quiz champs on Jeopardy, the American TV game show. The achievement of that day, witnessed by millions, seems, if anything, more interesting today as developments in artificial intelligence have moved centre-stage.

The spectre of middle-class jobs lost to AI has become part of conventional wisdom. As people spend more and more time exchanging data with distant computer servers, knowing little about what happens between their keystrokes and the results they study onscreen, the systems which control information, whether classified as AI or not, are ever more sophisticated and central to our lives.

The Watson experiment on Jeopardy was both a triumph of scientific and technological research and a kind of homage to the great tradition of computers in the States. Watson was built by IBM’s research team and named after the company founder. Thomas J. Watson and his son, between them, turned the company from a cash register business in Chicago to the epitome of corporate modernity, selling mainframe computers to customers who first had to learn what a computer was.

On a shorter timescale, Watson was the follow-up to another IBM triumph, when its computer Big Blue beat world chess champ Gary Kasparov in 1997. That was an extraordinary feat, but at least chess is a game with a limited number of possible moves – albeit a very large number.

But how could they make a machine that could deal with the natural language used in Jeopardy questions? Especially since the tradition of Jeopardy was to ask witty, punning questions, a bit like crossword clues? To make it more difficult, as a result of the game show scandals of the 1950s, where popular contestants were given the answers to keep them on shows and improve ratings, Jeopardy had been designed to prevent such a possibility by giving the contestant ‘the answer’, and requiring them to formulate the right question.

So that was the challenge IBM’s research team took on, less than four years before the show in which their computer won. It was partly a question of computer speed: even if Watson knew the answer, it had to be able to produce it before the human champions that were its opponents. These winners dealt in split seconds, hitting the buzzer often, apparently, before their conscious minds had an answer. As one put it “you find your thumb pressing the buzzer while the brain races to catch up.”

An early version of the computer was so slow that the programmers would ask it a question and then go to lunch, hoping it might have produced something (usually the wrong answer) by the time they returned.

Stephen Baker, an experienced business and technology journalist, was given privileged access to IBM’s team as they tackled their audacious challenge. The result is a technology thriller, with no shortage of intriguing characters, incidents and, well, jeopardy. The story brings together the East coast world of IBM, and the West coast world of network television – another venerable US institution, harking back to the innocent days when home entertainment meant sitting as a family, choosing between the three networks and a couple of local stations.

Network television, as much as IBM, was on a difficult journey to adapt to the modern world – a world in which TV was one of many choices of screen entertainment beckoning from a variety of devices. To lose Jeopardy’s academic fustiness, its producer Harry Friedman had broadened its agenda. Now, as well as the traditional, fact-based questions, there were many that required a knowledge of pop culture or just ordinary life. When weaved into tricky ‘answers’ by the show’s writers, they made Watson’s life harder. How could a computer possibly get this right?

Answer (question): “Here are the rules: if the soda container stops rotating and faces you, it’s time to pucker up.”
Question (answer): “What is Spin the Bottle?”

Baker’s account gives enough detail to appreciate at least the principles with which the IBM team approached their challenge. For instance, they broke it down into sub-tasks: understanding the question, assembling a massive library of information, creating a list of candidate answers and assigning a level of confidence to each. The latter because a Jeopardy contestant is also required to gamble money on its chance of getting an answer right, and must even take a view on how its opponent will bet.

The story raises the question of how intelligent machines should be presented to human beings. What sort of ‘character’ should Watson be given? After thinking about tones of voice, visual representation and physical form, the team decided to create a screen view of Watson’s brain: activity in the computer would produce a display that showed Watson ‘thinking’. But there would be no attempt to turn the computer into humanoid form, as that might encourage fears of computers taking people’s jobs. It would have a calm male voice and wouldn’t attempt to mimic emotion – triumph, frustration or disappointment. That might produce an unintentionally comic effect. Instead, Watson would remain “relentlessly upbeat”, whatever was going on in the game.

As well as giving IBM some good publicity – and risking the opposite if it had failed – the Watson project had serious business potential. Not only could a Watson-related machine master huge libraries of information, it could also analyse all the online information being produced every second. As Baker puts it:

“A new generation of computers can understand ordinary English, hunt down answers in vast archives of documents, analyse them, and come up with hypotheses. This has the potential to turn entire industries on their heads.”

Medicine might be one of the first fields to benefit, but it won’t just be the limitations of technology that determines how it goes; it will also be human foibles, especially pride. As one doctor put it: “Doctors like the idea of having information available. Where things get more psychologically fraught is when a damned machine tells them what to do.”

The impact of AI is examined in Martin Ford's Rise of the Robots, which I wrote about here

View all my reviews

No comments:

Post a Comment