In Jeopardy: computer challenges human race
A computer program is challenging humans in a popular quiz show this week... It's a giant leap forward for AI but questions persist over just how far machines can go...
This week Jeopardy, a popular American quiz show, is hosting an unusual contestant. Two former champions will face a new challenger: a supercomputer called Watson.
Constructed by the tech company IBM, Watson is hailed by its creators as one of the most advanced Artificial Intelligences (AIs) the world has ever seen. So far, its performance on Jeopardy's general knowledge tests has been impressive.
The program runs on a massive bank of 90 high-end servers, around 4000 times more powerful than an ordinary home PC. Housed in a back room, these whirring machines send instructions to a mechanical avatar, which stands next to the human contestants and presses its buzzer with a mechanical hand before delivering its calculated answer.
It's not the first time computers have tried to play us at our own games. In 2007, scientists at the University of Alberta built an AI called Chinook that cannot be beaten at draughts.
Ten years earlier, a computer called Deep Blue pulled off an even more impressive feat when it bested grand master Gary Kasparov at chess. As Kasparov surrendered, he wept, unable to come to terms with the machine's incredible skill.
But, although Watson is an amazing computer, it is not unbeatable.
One Jeopardy player, Greg Lindsay, managed to defeat Watson at the game by exploiting the machine's strengths and weaknesses.
Computers have incredible memories. They can store vast numbers of facts and can access all the information in an instant. Deep Blue beat Kasparov at chess by calculating billions of possible consequences for each move.
But there are things humans do easily that computers find incredibly hard. Computers have not yet mastered human language. Watson can easily tell you who is the President of Guatemala, but he can't understand clues with wordplay or puns. The machine couldn't guess who this clue referred to: 'X marks the spot, man, when this guy opens his peeper.' The human contestant answered easily ('Cyclops' – from the X-men comics).
So AIs have some way to go before they are able to think and talk like in sci-fi films. But they do get closer every year. Analyst Ray Kurzweil thinks that by 2020, we'll see computers that rival the human brain.
Is this a cause for fear? Some see a bright future in which intelligent machines take care of all our problems, doing difficult jobs and allowing humans to live a life of ease. But others warn against AIs. Even the smartest machines, they argue, will never be able to replicate the human soul.
- Can a machine ever really be alive? Can it have a soul?
- If powerful machines could run the world, always making the right decisions, allowing humans to live easy, comfortable lives, would that be a good thing?
- Design a quiz that a human would be able to do but which a robot would find difficult. Test your classmates.
- In groups, rehearse the arguments either for or against Artificial Intelligence. Then pick a representative to argue your case.
Some People Say...
“The human brain is just a biological machine.”
What do you think?
Q & A
- Why can't computers understand language?
- It's complicated. The short answer is that although they can count and remember much better than humans, they cannot truly think.
- Will they ever be able to?
- Maybe. The certain thing is that computers get more powerful each year. Computer power has more or less doubled annually since 1958.
- Will it keep doubling?
- Progress may slow down. But futurists note that as computers get better, they allow us to improve even further on existing designs.A In a way, brains are just wires and electricity too. Brain cells pass electrical signals to each other through complex webs of connections. And yet this web of cells and electricity can produce thought, emotion and perhaps even the soul.