Linux.com

Is that free will as in free code or free beer: On Minsky's artificial intelligence

Myself, I never understood the push to endow computers with human-like consciousness. Why do AI researchers believe the tools that they work with (transistors) can be fabricated into things that could think for themselves? I mean, toaster designers don't go around proclaiming that, given enough heating coils, they could build a sentient machine that could converse with us in the universal language of crispy bread. But these AI people, they're such Dr. Frankensteins, lusting to create life itself! -- even if the monsters they do build are pretty pathetic, judging from such sorry examples of human simulation as Eliza.

There are some pretty obvious reasons why AI doesn't work, not the least being scale. As the trade journal Electrical Engineering Times soberly points out ["Chip stack aims for brain-like connectivity]: Brains consist of a trillion ... or so neurons that act as both processor and memory ... Today's microprocessors, on the other hand, have just a few million logic gates to process information." Today's computers are Tinkertoys compared to human gray matter.

But even with the exponential gains future quantum, optical, or superconductor-based computers are supposed to offer us -- emphasis on the "supposed" -- there is still no proof that consciousness can be replicated in a machine, or that humans are merely fleshy input/output devices. Just as it took a non-Euclidian geometry to help Einstein conjure relativity, something other than strings of zeros and ones may be necessary to fire up a human noggin. In other words, the difference between real minds and silicon ones may not just one of degree, but of kind.

Heady stuff, so to speak. Accordingly, most AI researchers have lowered their expectations over time. And divorcing AI research from the goal of achieving consciousness and instead focusing on mimicking simpler human thought patterns has produced scads of useful, or at least workable, results, from beating world chess champions at their own game to gigantic analytical databases that can figure out insurance rates.

But Minsky, a professor of electrical engineering and computer science at the Massachusetts Institute of Technology, is not one to cower before 40 years of failure. He's gone the opposite route. Faced with the failure of AI to achieve consciousness, he attacks the very idea of consciousness itself.

During a May 23 talk at the Game Developers Conference 2001 in San Jose, Calif., Minsky discussed why AI hasn't worked yet. [For a transcript, see Dr. Dobb's TechNetCast.] "The reason consciousness has baffled so many people, especially physicists, is very simple," Minsky said. "There isn't any such thing. Consciousness is a word that we use as a suitcase word. It's a word we use as a name for a dozen very hard problems about how the brain or the mind works, which are quite different from one another."

"Oh, good lord," I thought when I first read that. Because computers can't emulate consciousness, it doesn't exist? Because computers can't become human, he wants us to deny what is human about ourselves in the first place? That, in effect is what Minsky's saying -- at least, that's how it sounds to me. It certainly saves him from having to defend his failures.

This negation of consciousness strikes me as awfully dangerous. I don't believe in the supernatural, but as Descartes argued some 350 years ago [all that "I think, therefore I am" stuff], the only thing we can be certain of in this world is that we exist, by virtue of the fact we have consciousness. We negate that at our own peril. Given that so much of the information technology the Minskys of the world build is simply geared toward getting us to buy more stuff, ridding ourselves of will smacks of brainwashing -- as if Minsky is now operating by technology's imperatives, not humanity's.

I'm not sure if Minsky has ever come out and said publicly before that he doesn't believe in consciousness. But he's been heading in that direction since the mid-'80s, when he wrote the book The Society of Mind, in which he argued that the mind works not as a unified whole but as a collection of many different agents, each handling one simple task. Say you want to get a beer out of the refrigerator. Your mind instructs your body to execute a whole batch of discrete actions: Hefting your butt out of the recliner, grabbing the beer, finding a bottle opener, opening the bottle, etc.

What Minsky doesn't explain is what motivates all these agents of the mind in the first place. He may have explained how we do what we do, but not why we are impelled to do anything, such as get a beer. Is it just self-survival that concocts this illusion of the "I"? Heck if I know. But consciousness, however slippery, is still the only game in town. If Minsky doesn't work towards that end, his intelligence will seem mighty artificial indeed.