This is an attempt to predict the future, and in particular, the future of computers as they relate to humanity.
Like all attempts to predict the future, it is bound to be fraught with error. "Fools rush in where angels fear to tread." Indeed. And onward anyway.
A good place to start would be Deep Blue's defeat of the world chess champion Gary Kasparov. Kasparov is considered by many to be the greatest chess player who ever lived. Deep Blue was a supercomputer built by IBM to play chess. For the first time, Kasparov was defeated in a match.
Though such feats are impressive, attempts to model computers after human intelligence have so far failed miserably, primarily because humans possess something called "common sense," which is proving to be extremely difficult to program, since it involves millions of small but complexely interconnected pieces of information, such as "Before you put liquid in a glass, the open side has to be facing up," and so on.
The lack of common sense will not prevent computers from evolving into extremely competent but narrow specialists like Deep Blue. Deep Blue is now supreme in its field, but if it's taken even slightly out of the realm of chess it becomes completely lost and idiotic.
In contrast, if you take humans out of their normal environment they will normally find a way to adapt, using their seemingly ordinary but actually quite profound and subtle "common sense."
Let's pause for a moment, and look briefly now at some of the most interesting new areas of computing.
One of the newest branches is genetic algorithms, a technology which basically allows software agents to compete, mutate and evolve just like Darwinian evolution in nature.
The "fittest" software agents survive to the next generation, and then compete again with other agents, so that agents evolve which are better and better adapted to the designated goal, whether that goal is recognizing human emotions or human speech or oil-drilling geological patterns.
The difference is that such evolution can take place billions or trillions of times faster than it does in nature, with some very interesting and successful results to date.
Another new branch is neural nets, which are modeled after neurons in the brain. Each software "neuron" has an "axon" or channel through which it it "fires" or communicates its signal to other neurons. And each software neuron also has "dendrites," through which it receives signals from other software neurons.
Together, like neurons in the brain, they form a complex net which, it turns out, has the ability to learn. Neural nets learn through feedback loops and by automatically adjusting the strengths of the connections between the various software neurons.
Neural nets have successfully learned, by themselves, to recognize different human faces. They've learned to recognize patterns in the stock market, complex patterns in the weather and many other things.
Cellular automata, another new branch of computing, has software "cells" or squares which evolve on a horizontal or 3-dimensional grid by following simple rules having to do with their neighbors, such as, "If 3 of my 6 surrounding neighbors on this grid are black, this cell will turn from white to black in the next generation. And of course each generation, or "beat," can take place in a billionth of a second, making possible very rapid development.
All sorts of strange and wonderful forms have developed in cellular automata, including forms which seem to contain the possibility of reproducing themselves, considered the central feature of life.
Expert systems, or rule-based systems, have also come a long way. They essentially distill a body of facts and rules used by experts in a given field, including the ability to reason by those rules. They have been used quite successful recently in diagnosing illness, laying out complex circuits on computer chips, controlling the motion of robots, and so on.
Genetic algorithms, neural nets, cellular automata and so on are all examples of complex adaptive systems," that is, systems which have no leader but which self-organize from the relatively simple behavior of many individual units.
Nature abounds in examples of complex adaptive systems, or CAS's. For example, an ant hill is a CAS; so is a turbulence pattern in a river; so is a cancer growing, a hurricane forming, economic manias and panics, a bee hive and so on.
Such systems display emergent properties, that is, complex behavior that could not have been predicted by even intense study of the simple units making up the CAS.
An excellent example of a CAS and an emergent property is the internet, which has arisen as a net of independent units in a communicating grid, with no leader. It's a self-organizing system, just as a hurricane or an ant-hill or an epidemic is.
Complex adaptive systems in computer software are currently all examples of idiot savants, that is, intelligence which can be extremely penetrating and brilliant in a particular narrow area, but is otherwise an idiot in the wide external world. (Hmm; sort of like me.)
But in these narrow areas, computer intelligence will evolve to become extremely deep, capable of analyzing data and reaching comclusions which far surpass the ability of human beings. Deep Blue is only the first example of many, because computer intelligence is evolving at a rate which is many billions of times faster than nature's rate.
We are already very dependent on computers and the net to which they're connected. And we'll become more so as time goes on, because computer intelligence will keep evolving in various areas until they make better decisions in those areas than we humans do, and so increasingly we'll let them make decisions in those areas.
An example is the latest jet aircraft, whose wings are adjusted thousands of times per second by computer. The pilot has to let the computer "fly" the plane in certain ways because it can do a better job than the human can, indeed, in a way that a human could not even attempt.
In business, executives more and more will need to trust decisions made by deep computer intelligence in particular areas, just to stay competitive. And as consumers, we'll become more dependent on the comforts of postbiological evolution, such as computerized personal companions.
In the near future, there will be brain catalogs just as there are computer catalogs today. In a brain catalog, you buy the components of a brain in a system, sort of like specifying an 800 megahertz chip and a 60 gigabyte hard drive in a computer today. There will be various brain systems that we can buy.
Then, for our purchased brain, we'll get to buy peripherals and enhancements for it, including input and output units such as vision and speech and so on. We'll hook all this up in a system much as we put together a computer system today at home or work.
We'll purchase these retail and mail-order brains for much the same reason we buy computers today—to stay current, competitive and viable in our society, whether as a person or a business. We'll use such systems to survive in a world where they're increasingly used by everyone else. In this process we'll become vastly more dependent on them than we already are.
Okay. Now things start getting interesting...
(This is the end of Part One. Go to Part Two.)
—jim sloman, for 6/14/01.
|
