Technological Notes: Computers with no common sense

Click to follow
The Independent Culture
WITH COMPUTER power doubling every 18 months (Moore's Law), some ex-perts in artificial intelligence are now bragging that by 2050, robots will surpass the intelligence of humans. They boast that their creations may one day treat us like pets, too feeble and stupid to take care of ourselves. Robots will inherit the earth, and perhaps place us in zoos. The scientists at Carnegie-Mellon Institute, with the largest robotics laboratory in the world, even claim that we are creating our own Darwinian successors, that it's our evolutionary duty to prepare the way for super-intelligent robots, as if it's a law of nature.

But not so fast! The rosy claims of these mathematicians are not without some merit, but they have been made before, with laughable results. As a physicist, I must note that there are a few tiny roadblocks in their predictions, such as the laws of physics.

It's true that Moore's Law has successfully predicted the staggering growth of computers for the past 50 years. This means that every Christmas, our computer gadgets are almost twice as powerful as the previous year's. It's also true that our brains, even when daydreaming, compute at more than 500 trillion bytes per second. By Moore's Law, our chips should hit about 500 trillion bytes per second by 2050.

But there are also tremendous roadblocks. For instance, Moore's Law will collapse way before 2050. In fact, by 2020, transistors will become so small, they will approach the size of DNA coils. Unfortunately, silicon is not structurally stable at such tiny distances. Silicon Valley may become a Rust Belt by 2020 - the Age of Silicon will end, causing chaos in the computer business.

After 2020, physicists will have to replace silicon with very speculative designs, such as computers which compute on laser beams, DNA molecules, protein molecules, or single electrons. There's also the ultimate computer, the "quantum computer", which is so advanced that no prototypes even exist. These are all highly risky designs, so all bets are off after 2020.

Furthermore, scientists have not solved the problem of "common sense". Computers don't know the things that even children know. Computers don't know that water is wet, that animals don't like pain, that mothers are older than their children, that twins age at the same rate, and that when you die you don't come back the next day. How do we know these common- sense things? By experience. But that's precisely what computers don't have. Worse, mathematicians believe that several hundred million statements of common sense are required before a robot can reason like a child. This staggering amount of common sense has effectively stopped progress in artificial intelligence for the past 20 years.

Finally, computers don't see very well. One of the most advanced robots is the Mars Rover, currently on the planet Mars. Although the Mars Rover captured the imagination of millions of people around the world, it has the intelligence of a retarded cockroach. Even a cockroach knows how to flee when spotted on your kitchen table. If a Martian were to swat the Mars Rover with its tentacle, the Rover would take several hours to realise that the tentacle was not a rock!

This is not to say that we won't have computer marvels by 2050. By then, chips will cost a fraction of a penny, the cost of bubblegum wrappers. We will have computers in our watches, clothes, jewellery, furniture, walls, appliances, and perhaps inside our bodies. And we will talk to them in English, albeit a highly stylised and simple English. But don't expect to have robot butlers or maids by 2050.

In other words, don't worry; the mathematicians are wrong. Humanity is not doomed to live in a cage after 2050. (Unless, of course, we physicists can invent the quantum computer . . .)

Michio Kaku is the author of `Visions' (Oxford University Press, pounds 8.99)