SILICON HUBRIS

Published in slightly different form in Tomorrowsf (http://www.tomorrowsf.com), No. 13 (January-February 1999).

Thomas A. Easton


Artificial intelligence (AI) is one of those areas some people claim we were not meant to venture into (another is artificial life). To attempt to create artificial intelligence is to challenge the gods, they say. The only greater challenge would be to try to create "artificial souls."
Such challenges to the gods are what the ancients meant by hubris. The implication of such a labeling is that--just as in the myths of Prometheus and Icarus, among others--the challengers will be punished.
On the other hand, computer programmers claim hubris as one of their cardinal virtues. They don't have much truck with ancient myths. The gods were grossly unfair to Prometheus when they staked him out to have his eternally regenerating liver torn by an eagle, and Icarus just plain screwed up (Daddy Daedalus told him not to fly too near the sun!).
Therefore, say the programmers, anything that challenges the gods is a right and proper goal to pursue. Artificial intelligence? Why not? Go for it!
So what is artificial intelligence? What is intelligence? We used to think that a simple question: We have it. Animals don't. Machines don't. But we now know animals can be pretty smart, and the machines now warrant a suspicious eye, for there has been an immense amount of progress since the dawn of the computer age.
Yet the critics remain, resisting vehemently the notion that a machine, a mere thing, could ever possibly really think. Nor could a machine possibly be a person! After all:

1. Machines have no souls, so cannot think;
   
2. Thinking machines cannot be possible, because the consequences would be too dreadful;
   
3. Mechanical reasoning is too limited;
   
4. Machines can have no inner experience to give meaning to their "thoughts" or actions;
   
5. Machines will never be kind, moral, joyous, perceptive, original, etc.;
   
6. Computers do only what we can program them to do;
   
7. Computers don't have nerve cells;
   
8. It is not possible to specify for a machine what to do in every possible circumstance a human can encounter;
   
9. Computers can't have ESP. (List from Moravec, Robot: Mere Machine to Transcendent Mind)
   

Alan Turing tried to dismiss these same objections in 1950, when he first devised the test for machine intelligence that still bears his name (see below). The effort was as futile then as it is today, for artificial intelligence echoes powerfully of myth, of the rabbi's golem and Dr. Frankenstein's monster. To far too many people, it smacks of the demonic.
To computer scientists, on the other hand, this attitude is just "meat chauvinism." They have two definitions of AI: First, AI means simulation (generally but not necessarily by computer) of things we call "intelligent" when people do them, with the aim of replicating or even surpassing people, or of shedding light on how people do intelligent things. The second definition differs in the aim of building "intelligent" systems that do intelligent things but not necessarily the way people do them. One example is the expert system. Another is any computer chess player such as IBM's Deep Blue, which whupped the human grand champion, Garry Kasparov. Both replace insight, intuition, and pattern recognition with laboriously methodical working out of alternatives.
We could also add autonomy, freedom from human control. Initially, intelligence meant pretty much whatever people did in their heads, but early successes rapidly removed many such things (e.g., arithmetic, algebra, geometry, theorem proving, games such as checkers) from the list of "intelligent" activities--at least when machines do them. Today AI indeed means largely autonomy, as might be displayed by a self-driving car. Unfortunately, many people find autonomous machines alarming--they want to be in control of their machines. Many also want to know a human is in control so they know who to sue when something goes wrong--in fact, liability fears have been a major factor in keeping expert systems from being adopted in areas such as medicine (they can be very good at sorting through large but narrowly defined data sets, but we see them mostly where consequences of error are relatively small, as in tax software).
To science fiction readers and writers--and to the general public, which gained its ideas of AI from Hollywood SF--these definitions of AI don't go quite far enough. We picture machines that can act on their own initiative, free of human direction, but we also want machines that can talk back to us. Machines with not just smarts, but personality too! Mechanical people. Robots, if you will.
This is not actually a bad extension of the definition. However we define intelligence (I think of it as general problem solving ability), we must face the question of how to tell when a machine has it. The classic answer is the Turing Test, which demands in effect that an "intelligent" machine be able to pass as human in a conversation with humans (this necessarily omits anything intelligent that can't hold a conversation!). A number of programs have been devised to approach this (the Loebner Prize recognizes each year's closest approaches), and it is worth noting that even very simple ones can be remarkably successful. Meet Mgonz, an Eliza-clone with a rude streak. Since Mgonz can be highly offensive, you might prefer to talk to Eliza herself, and please do note that Eliza has been around since 1966. Some researchers would prefer you didn't waste your time; they say efforts aimed at beating the Turing Test are a waste of resources.
Another effort begins by viewing human intelligence as a kind of story-telling ability--in a sense, it says, we think in stories, and a machine that could tell stories would have a very interesting kind of "intelligence." One attempt to produce a computerized story-teller named BRUTUS was written up in Technology Review a year ago; BRUTUS is a nice illustration of how far the machines have yet to go. See also the Erasmotron.
Assuming that we ever figure out how to make an "artificial" intelligence, should we? This is the big ethical question of AI. It is where the "hubris" lies, the challenge to the gods, and there is Greg Garvey's ACM (Automatic Confession Machine) to raise the question in a very pointed way. Does AI (particularly in this form, perhaps) intrude on the link between human and divine?
Look again at Eliza (and Mgonz). Even simple AI can get people believing in it! And if they do--what does it matter what--human, machine, or god--is behind the screen? Greg Garvey says he's an artist; his ACM sets up a cognitive dissonance that cannot help but be thought-provoking. So does Eliza/Mgonz. They deal in illusion, and Garvey makes the intriguing point that AI doesn't have to be "real" intelligence--illusion will do! That's all it takes to pass the Turing test (it even works for people! See the Sidebar). By other definitions of intelligence (e.g., problem solving), only "real" intelligence will do, though it might not work the way human intelligence does (Deep Blue). Nor may it seem very human.
Many people seem to hate the thought of successful AI because they fear that successful AI would knock us off our last remaining pedestal. It's been quite a while since we could tell ourselves with a straight face that we were the center of the universe, the center of the solar system, the specially created lord of all we survey, or even something finer and better and quintessentially different than mere animals. Now we must face the prospect of not being able to say we're better than mere machines!
There are genuine ethical questions to consider in connection with AI. After all, successful AI would threaten a great many jobs, it could lead to greater disparities between haves (both nations and individuals who can afford the machines) and have-nots (who can't), and it would make it easier for humans to slough responsibility for their decisions (forgetting that machines can be wrong, too).
Such considerations are important, but it may already too late for them. George B. Dyson (son of famed physicist Freeman Dyson) thinks AI may already be here. In Darwin among the Machines: The Evolution of Global Intelligence (Perseus Books, 1997), he says that in a very real sense it has been developing ever since Thomas Hobbes recognized in his Leviathan (1651) that in any mass of interacting, mutually adjusting parts there is a sort of intelligence. Hobbes was not well received by those who thought intelligence was particularly human rather than something that could belong to a society. But, says Dyson, the idea took root and grew as calculating engines and communications networks (from semaphore and telegraphy to the Internet) were devised, until today we can glimpse a growing symbiosis of human and machine that already holds an intelligence both more and other than the merely human. We can't talk to it, so it can't pass the Turing Test, but it is no less there.
Or maybe it's only almost too late. For over a decade, Carnegie Mellon University's Hans Moravec has speculated on the replacement of biological intelligence by machine intelligence. To him, robots will be the creations of our minds just as our children are the creations of our bodies, and since it is our minds that make us most distinctively human, our "mind children" will be our children in the more important sense. At the same time, mind transfer into robot brains and bodies would offer us true immortality.
In "When will computer hardware match the human brain?" Journal of Transhumanism, Vol. 1, 1998, Moravec describes Garry Kasparov's reaction to Deep Blue: "Kasparov ... claims to see into opponents' minds during play, intuiting and exploiting their plans, insights and oversights. In all other chess computers, he reports a mechanical predictability stemming from their undiscriminating but limited lookahead, and absence of long-term strategy. In Deep Blue, to his consternation, he saw instead an 'alien intelligence.'" Moravec adds that the hardware needed "to match general intellectual performance of the human brain ... will be available in cheap machines in the 2020s." At that point, "The visceral sense of a thinking presence in machinery will become increasingly widespread. ... There will be machines that can interact as intelligently as any human on any subject. The presence of minds in machines will then become self-evident."
In his latest book, Robot: Mere Machine to Transcendent Mind (Oxford University Press, 1998), Moravec reviews at length the development of robots up to the point where they can take control of their own continuing improvement or evolution. Here he pegs the achievement of machines capable of matching (or holding) a human mind at about 2040, and he devotes considerable space to what a civilization dominated by autonomous machines must mean for human economies. He does not shrink from the thought that most humans will be content to eat the lotus in idle bliss, supported by a horde of advanced robots programmed to be happy slaves. That won't last, though, he assures us. Some folks will tire of idleness and give up their humanity to become "Exes," robots animated by transferred human minds, and hie off into the far reaches of the galaxy.
Unfortunately, those Exes will have to move fast. Natural selection guarantees that our unprotesting robotic servitors will give rise to more independent sorts, even to predators eager to absorb the materials, energy supplies, and programs of anything--even ex-humans--that can't resist or elude them. In other words, there will eventually be a vast and diverse robotic "ecosystem" out in space, with predators.
In time, says Moravec, mere robots and Exes will be surpassed by superminds, which in turn will give way to what can only be described as an intelligent universe, a universe converted to computational apparatus and sheer mind in which all reality is virtual. Indeed, he tells us, within this vast "cyberspace," the "real" world will exist only as a simulation of itself. This includes plants, animals, and intelligent beings, and the simulation will be so perfect that no one will be able to tell the difference. It sounds rather like Robert Charles Wilson's Darwinia (TOR, 1998), which begins with the replacement of Europe by a strange wilderness full of very unfamiliar plants and animals, quite as if a piece of another world had been grafted onto Earth's flank. In due time, Wilson reveals that at the End of Time, when the universe is winding down, the godlike (to us) intelligences of the era constructed an Archive that would store and replay the history of the universe, including planets such as Earth. Unfortunately, it was almost immediately infested by "subsentient virtual entities, relics of a war that had devastated a distant galaxy long before [who] were using the Archive as a platform to preserve their algorithms against thermal death. They lacked moral regard for any entity not themselves, but they were fully aware of the purpose of the Archive and of its designers. They had not simply captured the structure, they had taken it hostage." Changes were being made. Past lives were being converted from records to new lives, living within the Archive with absolutely no awareness that they were not "real," and the Archive's masters were at war to expel the virtual foe.
A very interesting question raised by both Wilson and Moravec, as well as by a host of writers who have focused on the charms of virtual reality, is whether it truly matters whether one lives in a "simulated" or "real" world. Do we need a new, expanded version of the Turing Test that would apply equally to conversation, mental downloading, and even cosmic simulation? Put simply, it would say that if "real" and "synthetic" feel equally real, they are.
Surely we have a long while to wait before we must confront that question. Or perhaps not. There are a few scholars out there who think Moravec is a little conservative. For instance, Nick Bostrom of the Department of Philosophy, Logic and Scientific Method, London School of Economics, writes that "we will have superhuman artificial intelligence within the first third of the next century" (emphasis added).
If there's anything to the idea of hubris--that those who challenge the gods get hammered for their presumption--then the next few decades should generate some very interesting history, unless Wilson is right and we have to wait till the End of Time for our comeuppance.


Dr. Thomas A. Easton is Professor of Life Sciences at Thomas College in Waterville, Maine. He has been the Analog book columnist for almost 20 years. His latest novel, Unto the Last Generation, is available only on-line, from Mind's Eye Fiction. Last year's Silicon Karma (White Wolf, 1997) was well received; copies are now available from the author (teaston@acadia.net). His latest nonfiction books are Taking Sides: Clashing Views on Controversial Issues in Science, Technology, and Society (Dushkin Publishing Group, 1995, 2nd ed., 1997, 3rd ed., 1998) and Periodic Stars: An Overview of Recent Science Fiction (Borgo Books, 1997).
Previous columns are available here.