Happy birthday, HAL

Last month a very important celebrity marked his birthday. He wasn’t an actor, though he was in a movie; he wasn’t an author, though he appeared in a book. And strangest of all, he died almost 30 years before he was born.

He was HAL, the artificial intelligence that guided the spacecraft Discovery to its rendezvous with an alien artifact in 2001: A Space Odyssey. HAL, though very smart, was unfortunately also a trifle unbalanced, and after an ill-considered decision on his part to kill most of the human crew of the spacecraft, had to be deactivated by the survivor, astronaut Frank Poole.

In the book, during that deactivation, HAL identifies where and when he was born: “I became operational at the Hal Plant in Urbana, Illinois, on January 12, 1997.” (The movie, as movies often do, got the date wrong: it put his birthdate in 1992. As if.)

In celebration of HAL’s birthday, the University of Illinois in Urbana is staging a “Cyberfest” in March, but a much more significant celebration of HAL’s birth has been announced by NASA: it has plans to develop a new artificial intelligence–and launch it into space. Not only does this tickle the fancy of those, like me, for whom 2001: A Space Odyssey was an eye-opening experience in the theatre (even though I saw it at a Saturday matinee with a bunch of other 10-year-olds who didn’t appreciate it nearly as much as I did), it should also perk up the ears of Star Trekfans: NASA is calling this new artificial-intelligence-housing space probe Deep Space One.

What makes this a fitting tribute to HAL, apart from the timing, is the fact that NASA admits that Arthur C. Clarke’s tale sowed the seeds for this new spacecraft.

Deep Space One, which weighs 430 kilograms, is scheduled to fly by asteroid McAuliffe in 1999 and the planet Mars in 2000. It’s shaped like an octagonal cylinder with two large rotating solar panels. But what makes it really special is the artificial intelligence–NASA calls it a “remote agent”–that will be going along for the ride.

This “remote agent” has three capabilities that will help it control the spacecraft. The first, “high-level planning and scheduling” means that the agent can schedule its own spacecraft activities, and reschedule and re-plan them depending on circumstances. It also distributes resources such as electrical power. This means that ground control doesn’t have to send a series of repeated commands: it can just give DS1 an overall goal and leave it up to the remote agent to achieve it. This, in turn, greatly reduces the number of people needed at ground control, and the overall cost of the mission. (When everything has to be done from the ground, as has been the case up to now on major interplanetary missions, literally hundreds of people are required.)

The second, “model-based fault protection,” is a virtual version of Star Trek‘s Scotty, overseeing the functioning of the spacecraft’s machinery and checking to see if everything is working properly. And if something does go wrong, it can diagnose the problem and suggest ways to fix it or work around it.

Finally, “the smart executive” executes the plans laid out by the other two components or received from ground control, overseeing all the details necessary to reach the goals set for it.

Besides reducing manpower and overhead costs, the new artificial intelligence also improves communication, because it doesn’t have to wait for orders from ground control, delayed by the time it takes radio waves to cross the enormous distance between Earth and the probe, before sending data.

“It really helps to have visionaries like Arthur C. Clarke write about what might be possible someday,” says Dr. Douglas Bernard, an engineer at the Jet Propulsion Laboratory. “I hope that someday we make HAL 9000. This is just a first step in that direction, but it’s a giant leap in the levels of autonomy that spaceships have.”

Maybe, but it’s still several more giant leaps from a semi-automated space probe to HAL, who could read lips, play chess, and decide that humans were a bit of a nuisance and he’d be better off without them. (NASA did take one other hint from 2001; in the event the remote agent goes haywire, it can be deactivated–“We can perform a lobotomy,” as one scientist put it.)

Just how far are we from that type of truly “thinking” machine?

That depends on who you ask. Deep Blue, a chess-playing computer from IBM, almost (but not quite) beat Gary Kasporov last year. At MIT, a robot called “Cog” is (its creator hopes) learning about the world the way babies do, programming and reprogramming itself by interacting with the people and things around it. Its creator hopes someday to have a conversation with it. Down in Austin, Tex., another research team is feeding information into a system called CYC, as they have been for more than a decade, typing in the rules of “human consensus reality” (a.k.a. “common sense”), on rule at a time: “Bread is a food.” “You’re wet when you sweat.” CYC has received a million of these rules so far, and its creator hopes that after it gets another million or so it will be able to simply swallow whole encylopedias by itself and then ask for clarification of any points that puzzle it.

Will either of these machines–or their undreamed-of descendants–think like we do? Probably not. But that doesn’t mean that they won’t become sentient–that is, conscious of themselves as thinking individuals. Even philosophers are beginning to take seriously the notion that we may be creating a whole new life-form, one which will see the world quite differently than we do.

How will we know when a machine achieves consciousness? In 1950, in an essay in the journal Mind, British computer scientist Alan Turing proposed the following test for artificial intelligence: suppose an interrogator is communicating by keyboard with several entities he or she can’t see. Some are people and some are computers, and the interrogator has to guess which is which. To the extent the computer can consistently fool the interrogator, you can say that it thinks.

Turing thought that by the end of the century, computers would be able to pass that test. So far, none has been able to convincingly. Certainly, no computer in NASA’s DS1 is nowhere near that smart. Nor are Deep Blue, Cog or CYC. Clarke was apparently as overly optimistic as Turing by giving HAL a 1997 birthdate.

Nevertheless, there’s a good chance that we will see HAL’s equivalent in the next 30 years. One can only hope that it, unlike HAL, won’t decide that humans are rather a nuisance and it’d be better off without us.

Unlike 2001‘s Frank Poole and DS1’s ground-based scientists, we might not be able to pull the plug.

Permanent link to this article: https://edwardwillett.com/2003/02/happy-birthday-hal/

Leave a Reply

Your email address will not be published.

This site uses Akismet to reduce spam. Learn how your comment data is processed.

Easy AdSense Pro by Unreal