Ray Kurzweil bugs me. The futurist has been predicting for the last decade or so that some lucky people alive today will see the day when machines become smarter than humans. Thanks to the exponential growth of computer processing power, he foresees the coming of "The Singularity" â€“ the point at which self-aware machines take over their own evolution and leave humans in the dust. Rather than seeing the rise of artificial intelligence as a dystopia, a la The Matrix or Terminator, Kurzweil believes that the explosion in intelligence will sweep up mankind, ultimately making us immortal when our consciousness can be transformed into non-biological media.
Seeing our eventual immortality in machines has always struck me as wrong â€“ both erroneous and wrong-headed. In his Are You a Robot? lecture, Kenny Felder explicitly argues against the notion that lots and lots of really fast computations will ultimately result in consciousness. Consciousness, he said, was a completely different phenomenon than computation, and a bazillion calculations per second does not translate into the conscious experience of life and thought. That position resonated with me philosophically, but I until recently I hadn't seen anyone else try to back up that concept with harder science.
In the latest issue of Wired Magazine, though, Mark Anderson does flesh out the current science that challenges the notion of singularity. I was gratified to see that he starts exactly where Kenny did: "This notion sweeps under the rug a messy philosophic problem: An algorithm is only a set of instructions, and even the most sophisticated machine executing the most elaborate instructions is still an unconscious automaton." But even setting that aside, he pointed out the scientific puzzles that could derail the notion of immanent AI:
The existence of gamma waves. Brain scientists have long correlated gamma waves to consciousness; when you're awake and conscious, the brain pulsates in a measurable frequency. And yet, no one knows why. Cells that have no direct connection between them manage to stay in synch. Even more confounding, the pulsing is happening at the neurons input, in the dendrites, instead of in their output through the axons. Even with our current biological understanding of the brain, we may be completely confused about how the wiring actually works . . . and, by extension, how the brain could give rise to consciousness.
Computers within computers. Most neuroscientists have been paying attention to the networks of neurons in the brain. But new research suggests that the sub-cellular structures of microtubules might themselves be providing trillions of computations per second inside each neuron. Imagine a supercomputer inside each individual cell of your brain, and you start to get the idea of how phenomenally complex our brains really are.
Quantum phenomena. Some researchers go so far to suggest that the complexity might not stop at the molecular level. According to some physicists, some perplexing aspects of visual perception are most easily explained by quantum mechanics. Quantum mechanics, it seems to me, is the point at which science squints really hard to explain things, where uncertainty reigns and the most mysterious things happen. If consciousness turns out to be a quantum phenomena, then it may lay at the very extreme edge of what is even comprehensible.
None of this disproves Kurzweil's vision. Humanity, together with our very powerful computer friends, may yet sort it all out and construct consciousness. But it may take a few centuries longer than we expected.
I'm not sure about the discussion about gamma waves, but in general I agree. A close examination of a neural mechanisms at the cellular level show that common estimates of how much computation power is really in a brain may very well be not only quantitatively wrong by several orders of magnitude, but qualitatively wrong as well. In any case, with our best technology we have yet to build anything close to intelligent as an insect. There are algorithms which capture the algorithmic essence of an ant-colony for instance, but as far as simulating a full insect goes, we've got nothing. Consider the up close behavior of wasps, the preying mantis, hunting spiders (yes, a spider is an arachnoid, not technically an insect), even the lowly house fly and the evidence quickly stacks up to say that all of the accumulated exponential growth of computational power we have experience in the last few decades (which incidentally, in terms of computation per area is starting to show signs of decelerated growth) falls woefully short.