How, then, do you teach a computer common sense?
The word: Common sense
15 April 2006
From New Scientist Print Edition.
SOME things are just obvious. We all know that people don't walk on their heads, for example, or that if you go out in the rain you're likely to get wet. It's common sense.
But some things that seem obvious to one person may seem obscure to another if they are from another culture, religion or background. Common sense is not always common to everyone. This is especially true when beliefs play a strong part in how people perceive things. It may seem obvious to me that a drought was caused by a random change in weather patterns, but to someone who believes in supernatural beings it could seem just as obvious that it's because they've displeased the rain god. Einstein summed it up thus: "Common sense is the collection of prejudices acquired by age 18."
Spare a thought, then, for those trying to design computers that have common sense. It took each of us humans many years to build up a portfolio of wisdom about the way the world works. When a baby repeatedly knocks over towers of bricks it may look like a futile activity, but what the child is really doing is exploring the world and filling its brain with fundamental rules. By the age of 3, it will have acquired more common sense than the most sophisticated computer. This is the big headache for artificial intelligence (AI) researchers: they can design a computer that might beat Garry Kasparov at chess, but you couldn't have an intelligent conversation with it because it has no grasp of ordinary life.
How, then, do you teach a computer common sense? Researchers at a company called Cycorp in Austin, Texas, are trying to find out. Since 1984, they have been incorporating a huge collection of everyday knowledge in an AI project named Cyc. Cyc now contains around 300,000 concepts, such as "sky" and "blue", and around 3 million different assertions, such as "the sky is blue", in a format that can be used by computers to make deductions. It can now generate hypotheses from the facts it contains, and confirm them by searching the web. Cycorp has also just launched a trivia game for the public that will help fill in gaps in Cyc's knowledge, in which the computer generates statements that the user has to describe as true, false or incomprehensible. You can see a prototype at cyc.com.
There's a long way to go. Despite more than 20 years' work, the Cyc project contains only about 2 per cent of the information its designers think it needs to operate with something like human intelligence. It also begs a crucial question: can a computer really acquire common sense without experiencing the world directly, given that the latest neuroscience shows we base our judgements on gut feeling and emotion rather than a rational assessment of the facts?
15 April 2006
From New Scientist Print Edition.
SOME things are just obvious. We all know that people don't walk on their heads, for example, or that if you go out in the rain you're likely to get wet. It's common sense.
But some things that seem obvious to one person may seem obscure to another if they are from another culture, religion or background. Common sense is not always common to everyone. This is especially true when beliefs play a strong part in how people perceive things. It may seem obvious to me that a drought was caused by a random change in weather patterns, but to someone who believes in supernatural beings it could seem just as obvious that it's because they've displeased the rain god. Einstein summed it up thus: "Common sense is the collection of prejudices acquired by age 18."
Spare a thought, then, for those trying to design computers that have common sense. It took each of us humans many years to build up a portfolio of wisdom about the way the world works. When a baby repeatedly knocks over towers of bricks it may look like a futile activity, but what the child is really doing is exploring the world and filling its brain with fundamental rules. By the age of 3, it will have acquired more common sense than the most sophisticated computer. This is the big headache for artificial intelligence (AI) researchers: they can design a computer that might beat Garry Kasparov at chess, but you couldn't have an intelligent conversation with it because it has no grasp of ordinary life.
How, then, do you teach a computer common sense? Researchers at a company called Cycorp in Austin, Texas, are trying to find out. Since 1984, they have been incorporating a huge collection of everyday knowledge in an AI project named Cyc. Cyc now contains around 300,000 concepts, such as "sky" and "blue", and around 3 million different assertions, such as "the sky is blue", in a format that can be used by computers to make deductions. It can now generate hypotheses from the facts it contains, and confirm them by searching the web. Cycorp has also just launched a trivia game for the public that will help fill in gaps in Cyc's knowledge, in which the computer generates statements that the user has to describe as true, false or incomprehensible. You can see a prototype at cyc.com.
There's a long way to go. Despite more than 20 years' work, the Cyc project contains only about 2 per cent of the information its designers think it needs to operate with something like human intelligence. It also begs a crucial question: can a computer really acquire common sense without experiencing the world directly, given that the latest neuroscience shows we base our judgements on gut feeling and emotion rather than a rational assessment of the facts?
posted by jonas at
Wednesday, April 12, 2006
0 Comments:
Post a Comment
<< Home