Arthur C Clarke's brainchild

Cliff Joseph celebrates the `birth' of Hal 9000, the star of `2001: A Space Odyssey' that set the standard for real artificial intelligence

Yesterday, 12 January 1997, was the birth date of Arthur C Clarke's most famous brainchild. In his novel 2001: A Space Odyssey, that is the date on which Hal 9000, the intelligent computer, "became operational". Hal was the most memorable character in both the novel and the film upon which it was based, thanks in part to the silky voice of actor Douglas Rain. But Clarke and director Stanley Kubrick's vision of a truly intelligent computer had a genuinely profound impact and helped to shape our view of the relationship between man and machine.

Hal might seem to be a somewhat dubious role model, considering that he was eventually disconnected after malfunctioning and murdering most of the crew of the spaceship Discovery. None the less, Hal became the template upon which many later fictional computers were based and, more importantly, he set the standard against which we measure the progress of technology in real life.

To celebrate Hal's "birthday", the scientist David G Stork has compiled a book called Hal's Legacy, in which he and other leading computer experts discuss developments in artificial intelligence and assess how close we have come to creating a real Hal.

As Stork points out, some of Hal's abilities can be matched by computer systems that exist today. In the film, Hal easily beats astronaut Frank Poole at chess, and this is one area where great progress has been made in recent years. Last year, IBM's Deep Blue computer beat the chess grandmaster Gary Kasparov in the first of a six-game series, though it lost or drew the subsequent five games.

But few people would argue that Deep Blue actually exhibits real "intelligence". Its chess-playing is simply a matter of enormously fast number-crunching. Chess is a game that permits a vast but limited number of moves, and is played according to a clearly defined set of rules. Deep Blue simply calculates the outcome of millions of possible moves every second and selects the best moves according to the rules of the game. This is very different from the slower, but more intuitive and strategic approach of a human player.

Yet, as long ago as 1950, Paul Shannon, one of the pioneers of modern information theory, stated that "the investigation of the chess-playing problem is intended to develop techniques that can be used for more practical purposes".

Those practical purposes could include Hal's ability to navigate the Discovery and control all of its systems on the long flight to Jupiter. This sort of task is almost within the ability of current mainframe computers and could well be possible by 2001. Like chess, though, this sort of task is really just a matter of performing lots of simple calculations very rapidly. More complex aspects of human intelligence remain out of reach.

In the film, Hal can understand everything that is said to him by the ship's crew. He can talk and think for himself, and is even able to lip- read with the aid of a video camera. Here again, we already have quite inexpensive computer systems that can recognise speech and respond to spoken commands. Stork even believes that computers will be able to lip- read reasonably well within a few years (though Arthur C Clarke himself admits that this is the one aspect of the film that he didn't believe was possible).

But recognising the sound of words is not the same thing as understanding their meaning. Even if a computer could cope with a wide range of regional accents, it is still easy to confuse antonyms that sound the same such as "hear" and "here". To tell the difference between these words, a computer needs to understand the context in which they are used. The ambiguity of human speech makes it much more difficult to understand than the simple, unambiguous rules of chess.

Hal clearly understands those ambiguities, and that, more than anything else, is what sets him apart from today's computers. When Hal utters those famous words, "I'm sorry, Dave, I'm afraid I can't do that", he is behaving in an extremely human way. He is telling a lie, saying "I can't" when he really means "I won't". Hal is also displaying a very human survival instinct, protecting himself from the human crew who are planning to disconnect him. There's no doubt that this kind of self-awareness is far beyond the high-speed number-crunching of Deep Blue.

Opinion is divided over whether computers in the future will ever be able to achieve Hal's level of intelligence and self-awareness. In Hal's Legacy, writers such as the speech-recognition expert Raymond Kurzweil argue that the human brain is nothing more than a very powerful computer. He believes that human intelligence is merely the natural by-product of the complex interaction of neurons within the brain. Producing a truly intelligent computer is therefore just a matter of duplicating the structure of the brain.

Kurzweil thinks that it will be possible "to build a computer with the raw computing speed of the human brain" by about 2020. By the middle of the century it will also be possible to scan the structure of our brains and duplicate that structure using "neural networks" - complex sets of processors that mimic the workings of neurons in the brain. He even goes so far as to suggest that scanning the brain of a specific individual will produce an exact duplicate of their brain - "including the contents of its memory" - effectively allowing them to "download" their mind and all its memories into a computer.

That may sound even more far-fetched than 2001 itself, but two research scientists at BT, Peter Cochran and Chris Winter, recently attracted a lot of attention with a speech about a "soul-catcher" chip that could, theoretically, store all of a person's memories. Admittedly, the soul- catcher chip isn't even on the drawing board yet but, like Kurzweil, Cochran and Winter estimate that such a chip could be developed by the middle of the next century.

However, many philosophers believe that there is more to human consciousness than the mere physical stuff of the brain. True intelligence and self- awareness, they argue, require an additional, non-physical element that can't be reproduced in a silicon chip. Without that extra element, many people believe that no machine can ever be truly intelligent.

Ideas such as these raise all sorts of questions, both philosophical and scientific. If we do manage to create an intelligent machine like Hal, will it have a soul? And are we playing God - tampering, like Frankenstein, with forces that we ought to leave well alone? Those questions certainly won't be answered in 2001. But if computers keep developing at their current rate, we might get the answers sooner than we think

`Hal's Legacy: 2001's Computer as Dream and Reality', edited by David G Stork (MIT Press, pounds 16.59).

http://www-mitpress.mit.edu/Hal/