(This is Part 11 of a series. Go back to Part 10.)
2d. The rise of transcendent intelligence
Aside from the increasingly degraded and perilous biosphere, the greatest long-term trend, and ultimately "news story," will be the rising intelligence of machines. Eventually this development will become an absolute fixation of the media and the human race in general.
This rise in machine intelligence will be greatly facilitated by two developments:
On the hardware side, there are a number of promising technologies in the hopper, but I imagine that nanotube technology will win out, at least for awhile, because it offers so many advantages.
Nanotubes are minute closed-end tubes made from carbon atoms. As the analyst Ray Kurzweil points out, 50,000 of them placed sideways would equal the width of a human hair. Whew!—that's small!
But more important is their computing ability. Nanotubes, it turns out, can be turned into infinitesimal transistors orders of magnitude tinier than today's transistors on integrated chips. Not only that, but nanotubes would facilitate the building of 3-D chips—computing chips not limited to a single layer, but with thousands of layers—and with thousands or millions of processors in each layer.
Such chips would have many trillions of transistors and be capable of computing at speeds millions of times faster than today's chips and trillions of times faster than the human brain. That's t as in trillions of times faster. Such chips will also be massively parallel, with millions of processors grinding away on a single chip. Later on, photon computers using lightbeams will be faster and more complex yet.
Structurally, machines of the future may also be made of nanotubes. Nonotubes are strong yet amazingly lightweight. A car made out of nanotubes would weigh only 50 pounds yet be far stronger than today's cars. So the bodies of androids, at least for awhile, will most likely be composed of this remarkable material of the future.
On the software side, the increasing use of neural nets and evolutionary algorithms will enable types of intelligence not available from even the best "expert systems." Unlike rigid rule-based systems, neural nets model the way the human brain works, so that they can learn through experience. Meanwhile, genetic algorithms provide a speeded-up version of evolution among competing software agents to provide the most robust survivors. In both cases, this is an intelligence that can learn, evolve, "grow."
Present-day computers are only a few years away from complete mastery of natural language, which will enable them to go on the internet and learn by themselves from the billions of documents there. Later on, machines will also learn from video and audio and virtual reality on the web. And sooner or later androids will be out in the real world, learning either as the companions of their proud purchasers or, somewhat later, on their own.
This is not even to mention quantum computers, which compared to digital computers are like a battleship to a rowboat. Because quantum bits—qubits—can hold a 0 and 1 at the same time, using quantum effects, quantum computers will eventually be able to do in seconds computations that might take digital computers years or centuries to accomplish, if they could accomplish them at all.
Take all this together and the conclusion becomes inescapable that computers will first equal and then surpass human intelligence. And after that, exponentially surpass human intelligence.
Computers will not remain rigid and brittle, as they mostly are now. As the various fields of computing converge, and as machines become as complex as humans—and later much more complex than humans—intelligence will "emerge." It will be an emergent property, just as intelligence in humans was an emergent property once our brains grew large enough.
The genetic code between humans and chimpanzees, our closest neighbors, varies by only 1%. What does vary is that we have three times as many neurons as chimps do. Researchers are now speculating that this simple change in quantity alone was enough to trigger the emergent property of intelligence. This follows the well-known law that a large enough increase in quantity becomes a change in quality.
The same will be true of our machines. When the complexity of their circuitry rivals that of the human brain, which Kurzweil expects around the year 2020, then the intelligence of machines will begin to rival that of humans. And in subsequent decades, when the complexity of their circuitry begins to far outstrip that of humans, their intelligence will also begin to far outpace that of their human masters. It will be nothing less than the rise of a new intelligent species.
(This is the end of Part 11. Go to Part 12.)
—jim sloman, for 1.5.06
|
