A Contrarian Future for Minds and Machines

By Selmer Bringsjord

Most prophecies tend to be sensational, and today's predictions about tomorrow's computers and robots are no exception. In the June 19 issue of Time, devoted to "The Future of Technology," the author and inventor Ray Kurzweil tells us that by 2030, microscopic robots, called "nanobots," will be able to make a blueprint of your brain after you swallow—yes, swallow—a few of them. The blueprint will make it possible for a computer to replicate your brain in hardware that is 10 million times faster than the sluggish, old-fashioned gray stuff inside your cranium; the result will be an artificial intelligence immeasurably more clever than you.

In The Chronicle of Higher Education's July 12 edition, Vernor Vinge, an associate professor of mathematics and computer science at San Diego State University, gives us a more compressed time line. In view of Moore's Law, which holds that the power of computer hardware doubles every 18 months, he thinks that computers will be more intelligent than humans within 20 years. After that point, he says, humans will be left in the dust by machines that get exponentially smarter by the day—if not by the nanosecond. That will be the age of posthumanity.

Hans Moravec, a well-known roboticist at Carnegie Mellon University, thinks it will take a little longer for robots to exceed humans in all respects, from running companies to writing novels. In his latest book, Robot: Mere Machine to Transcendent Mind (Oxford University Press, 1999), he predicts that robots will be that advanced by 2040. Later, they will evolve to such lofty cognitive heights that next to them, we will seem as primitive as single-cell organisms are compared with us. Many others in the field of artificial intelligence predict the same sensational future, unfolding on about the same rapid schedule.

Depending on your point of view, the future that I envision is either brighter or bleaker than those predictions; it's certainly not as sensational. I believe that computers and robots are irremediably inferior to the human mind. I'd happily wager the value of my retirement savings in, say, 2030, that what I will be able to accomplish then on an average workday will be utterly beyond the smartest computing machine available in that year. My confidence stems from my own research, which shows that processing speed is irrelevant to what's most impressive in human cognition. To sit down with a pen before a blank piece of paper and produce a play like Hamlet involves doing something that no computer, however fast, can pull off.

Before I explain why a silicon Shakespeare is impossible, it's important to get a few mathematical facts straight. We know that certain problems can be solved by computation—and hence, in principle, by some computer or robot. We also know that certain problems can't. For example, the problem of playing invincible chess, or at least chess that no human can beat, can be solved computationally. That had been established mathematically long before I.B.M.'s "Deep Blue" trounced Garry Kasparov. On the other hand, countless problems exist that are computationally unsolvable, like determining whether an arbitrarily selected computer program will ever stop running.

That means that no matter how fast a computer computes, it cannot determine whether a computer program will keep going forever. Alan Turing proved that over 50 years ago; I and countless others teach his and related proofs every year.

Now, a little logic makes it plain that would-be prophets like Kurzweil, Vinge, and Moravec must assume that the human mind is incapable of solving any computationally unsolvable problem—given that the prophets base their prognostications on the ineluctable increase in the speed of computational hardware. After all, if the human mind can solve problems that computers cannot, then computers will not be able to outperform humans in 2040, in 3040, or in a quadrillion millennia.

Why would the prophets assume that the human mind lacks the capacity to solve computationally unsolvable problems? I haven't a clue. As I say, my own research, which is largely aimed at getting a computer to write fiction, has convinced me that humans routinely solve such problems.

As an example, suppose that the novelist Mark Helprin wants to write a compelling scene featuring a youth confronted with the sudden and violent murder of his parents. Helprin adopts his character's point of view, expresses in English what it's like to be that character under those circumstances, and then judges the English that he's written, which may need to be modified—again, based on his imagining what it's like to be that character.

I have worked for a decade—with the help of David Ferrucci, of I.B.M.'s T.J. Watson Research Center, and others—to design a computational system that can do what Helprin does. One problem that has plagued me during that decade is that computers don't have, and therefore can't adopt, points of view. They don't have feelings; they have inner lives on a par with those of rocks. No amount of processing speed is ever going to surmount that obstacle. I see that now, as clear as day.

So why do I pursue research designed to teach computers to write? The answer is that it's fun and can be productive, even profitable. Although no computer, however fast, will ever be able to write a scene as Helprin does, Ferrucci and I have discovered some interesting tricks that allow computers to write simple fiction. One of those tricks, used by our Brutus story-generation system, involves mathematizing and manipulating the concept of betrayal at work in Shakespeare's plays. We can't capture Shakespeare's imagination, but we can capture the structures that his imagination manipulated. That kind of trick will soon allow computers to compose more straightforward documents, like contracts, job descriptions, and proposals. Some artificial-intelligence companies are inches away from that sort of machine intelligence.

Proponents of artificial intelligence would do well to see themselves as helpful engineers, or clever tricksters, rather than the inventors of machines that will usher in a posthuman age. Many useful devices that rely on blindingly fast computation can be created with brilliant engineering: a car that drives itself, an automated air-traffic-control system, a tractor that cuts my lawn on its own, an artificial teaching assistant that grades my students' assigned problems in logic, and so on. Even when it comes to problems as hard as those faced by Helprin and his counterparts, tricks and shortcuts can bring us close to solutions.

Of course, some of my colleagues regard the job of engineer as uninspiring, and the job of trickster as demeaning. They will continue to strive for a race of machines that operate light-years beyond Homo sapiens. They will always fail. But they will never fail to tell us that we are soon to lose our position atop the intellectual food chain. © Selmer Bringsjord

Selmer Bringsjord is a professor of logic and cognitive science, and director of the Minds & Machines Lab, at Rensselaer Polytechnic Institute. His most recent book, written with David A. Ferrucci, is Artificial Intelligence and Literary Creativity: Inside the Mind of Brutus, a Storytelling Machine (Lawrence Erlbaum, 2000).