The Toronto Star
Sunday, January 12, 1997

Film star HAL was ahead of his time

No computers today can do the things he did

`I became operational at the HAL Plant in Urbana, Illionois, on January 12, 1997.'
-- The computer HAL on his origins,
in Arthur C. Clarke's novel,
2001: A Space Odyssey

by Joseph Hall

HAL was born a tad premature.

On this, the day of his fictional birth, nothing like the laid-back, lip-reading computer who went looney in Stanley Kubrick's 1968 film version of the sci-fi classic is even close to being turned on - in Urbana or any other city.

And while artificial intelligence experts are working to replicate aspects of HAL, none will be standard ware in the foreseeable future, computer scientists say.

Indeed, some facets - like HAL's capacity for human emotional outbursts that ended with him killing four of his space ship's crew - will likely never be incorporated into computers.

Some still believe ``gofai'' (good old-fashioned artificial intelligence), with its focus on logic and representation, just needs time to develop and is on the cusp of important breakthroughs.

``I would say there's been disillusionment overall'', says David Stork, editor of Hal's Legacy, a new anthology of articles on artificial intelligence. ``But there's renewed excitement. We know it's going to come.''

The most widely accepted definition of artificial intelligence is still the one proposed by British mathematician Alan Turing in 1950.

If a human interrogator conversing by teletype with a computer cannot tell if it is human or machine, he said, that constitutes intelligence. The test has so far proved impossible to pass.

To understand how close scientists are to recreating HAL, his ``character'' needs to be dissected.

At his simplest, HAL spoke like a normal human being (if a touch on the mellow side, courtesy of Canadian stage actor Douglas Rain).

``He used language . . . in a very human way'', says Graeme Hirst, a University of Toronto artificial intelligence researcher who has seen the film about a dozen times.

``Now language understanding is something that we've worked on for years . . . and some of the things that HAL did are things we can do today, or almost do today.''

Some simple conversations, such as when Dave the astronaut tells HAL to open a space pod door, are within the capabilities of any decent speech recognition system today, says Hirst, himself a computer language researcher.

But current systems require users to enunciate clearly - A. Word. At. A. Time. - and they cannot, like HAL, comprehend the natural slurring of human speech.

``But where HAL went way beyond anything we have today is that he could deal with incredibly complex conversations'', Hirst says.

``. . . If it's a genuinely new conversation with the computer not expecting anything about the questions, we're nowhere near that yet.''

Emotional Overtones

Some of HAL's speech in the film, especially as he's going ga-ga, is also charged with emotion.

``When he traps Dave outside the ship (as he tries to kill him), HAL's response is really very emotional'', Hirst says.

``Dave asks HAL what's wrong and HAL responds in sort of classic psycho-game way with words to the effect of, `I think we both know what the problem is here, Dave.' ''

Hirst says this is one of the areas where HAL goes far beyond anything we have, almost have, or would ever want to have in our computer programs.

``We couldn't imagine that this computer that's been programmed to run this spaceship could have ever been programmed or somehow learned a manipulative and game-playing emotional response like that'', he says.

``Certainly it's not something that would have been put into the machine on purpose, even if you were sort of programming it with emotional overtones to make it easy to talk to.''

HAL's attempt to kill Dave, however, is based on the type of seat-of-your-pants ``opportunistic'' thinking capacities that artificial intelligence programmers are currently trying to develop.

``In fact, it's interesting in retrospect that much of what HAL does, and the way HAL schemes, can be mapped quite nicely into the kind of planning and procedures that we've developed over the past 20 years'', Hirst says.

HAL's plan to kill the astronauts, who themselves are planning to shut him down, involves getting the Frank character out of the ship to deal with a questionable antenna. HAL then cuts him loose in the vacuum of space.

After this, Dave leaves the ship in a pod to try and retrieve his colleague, leaving his survival helmet behind.

``Dave leaves so fast he forgets to put on his helmet . . . and HAL could not have relied on this happening'', Hirst says.

``This is crucial now, because when Dave's outside in the pod, HAL refuses to let him back in and the potential success of HAL's scheme relies on Dave's mistake in not wearing his helmet.''

HAL is using what artificial intelligence researchers call opportunistic planning.

``That is, even as his plan is in action, he's able to take advantage of new developments, new knowledge, things that come along, and integrate them into his plan'', Hirst says.

Computer scientists pursuing the grail of opportunistic planning have a way to go before they can duplicate HAL's abilities. He simply knows far more than today's computers.

``HAL has just an incredible amount of knowledge about human behavior and about the world in general, quite apart from the standard things he must know about running the ship and so on'', Hirst says.

American artificial intelligence researchers are working to create an encyclopedic computer knowledge base. And yet, a computer with HAL's mental dexterity may still elude them.

``We don't have the techniques that would allow the computer to decide extremely rapidly which facts are relevant and which facts aren't'', Hirst says.

Neural Network ``We don't have the techniques that would allow it to know even which facts can be completely ignored and relied on not to change. But people can do this very easily, so it's possible, somehow.''

One way to rapidly process knowledge would be through ``neural network'' computers, devices being developed to mimic the human brain.

But neural naysayers maintain that such computers involve a few hundred or thousand or tens of thousands of simulated neurons. The human brain has billions and billions.

``There's no evidence your dinky neural systems can ever be scaled up to anything the size of the human brain'', says Hirst. ``That's simply hope and optimism right now.''

HAL was also able to read lips in silence and from a profile, something beyond the ability of even experienced human lip readers.

Although scientists today are working on lip-reading programs to assist computers in word recognition, they all require a full-face view and must be accompanied by audio input.

In some crucial ways, however, scientists have gone beyond the futuristic computer.

HAL's name - an amalgam of the words heuristic and algorithmic and not a play on IBM - would suggest he was a flexible thinker.

Heuristic (``guiding or helping one to discover'', says the Gage Canadian Dictionary) basically means, in the field of artificial intelligence, a rule of thumb, Hirst says.

``What people figured out quite early on in artificial intelligence is that many aspects of intelligence are in fact good rules of thumb but don't work 100 per cent of the time, that is they aren't algorithmic'', he says.

Human intelligence comes from the ability to recognize these rules of thumb, apply them, and then cope with what happens.

While computer scientists are currently working on heuristic programs, HAL was woefully short on this ability, despite his name.

``If HAL is truly a heuristic and algorithmic computer, then he must . . . be flexible, have the ability to make a mistake, because we recognize that sometimes the choice will be wrong.''

As it turns out, HAL was completely inflexible and convinced, to his ultimate downfall, of his own infallibility.

``So that's perhaps one area where we've gone well beyond HAL. We understand crucially, both at the hardware and software level, the concept of fault tolerance and flexibility in response.''

Modern computing has done HAL's futuristic world one better in another area.

``In the movie (astronauts) Frank and Dave are doing their routine work with clipboards, pencils and paper'', says Hirst. ``They don't have little portable computers or interfaces to HAL or laptops or anything else.

``They don't have video games.''

with files from Los Angeles Times

Copyright © 1997 by The Toronto Star. All Rights Reserved. Reprinted with permission.