|Computing by Committee|
Issue 04 (January 2005)
Only yesterday, it seems, small computers were mere hobbyist curiosities. Today they are machines of extraordinary power. Advances in this field have been nothing less than astonishing. And things have just begun to happen. More than one nation is now working feverishly to develop the so-called fifth generation computer. The goal? Affordable, truly user-friendly, artificially intelligent computers on our desktops and in our homes within the next decade.
Resulting fanfare would have us believe that we will soon talk to our PCs rather than program them, that our machines will learn from experience, becoming more intelligent with time; that virtuaUy any information or service will be ours simply for the asking. Our computers, for example, will be able to serve as personal financial analysts, tailoring investment strategies to suit individual tastes, and perhaps, when granted proper authority, effecting market transactions in our behalf without our direct intervention or knowledge.
Although ideal in all outward appearances, the situation is not without its disadvantages. Unfortunately, the existence of such a constructive technological capability also provides the protential for equally sophisticated mischevious electronic activity. I foresee the need in fact for future legislation to deal with the entirely new field of automated malfeasance. As true artificial intelligence emerges, less principled computers may be inclined to embezzle funds on their own accord, issue bogus stock certificates to other machines, or even sell retirement properties in uncharted regions of the Great Bald Cypress Swamp to gullible mortals. Future auditors would be required to ferret incriminating evidence from deep within hidden recesses of RAM where not merely two, but thousands of different sets of "books" might secretly be kept to suit a multitude of devious purposes.
On a more cheerful note, the new technology may enable us to commission our PCs to labor at our occupations by proxy. Under such conditions, enterprising fast-track professionals could manage multiple careers with teams of computers, simultaneously dabbling in manufacturing, marketing, medicine, mathematics and the like. Others might be inclined to turn on the machine and proceed to the nearest substantial body of water with rod and reel in hand. The potential scope of application is constrained only by the imagination.
In order to achieve this capability, scientists say, a machine one million times faster than today's fastest supercomputer must be developed. Unfortunately, a speedup of this magnitude can't be realized with conventional one-step-at-a-time serial computer designs. Recognizing the need for architectural innovation, researchers intend to gain the required performance by embedding many parallel central processing units, or CPUs, into a single computer. Working in close coordination, each processor will share a portion of the task at hand to complete the job with the required speed.
Results of recent experiments in this direction, however, have been disappointing. Prototype parallel machines are exhibiting diminishing performance returns as more and more processors are added to the computing broth. Like the proverbial cooks, large crowds of CPUs spparently do not function efficiently together. There is confusion, too many chips speaking at once, and conflicts arising over the use of resources such as memory cells and communications paths.
To add to the chaos, the recursive nature of the inference processes used in artificial intelligence algorithms tends to generate many subproblems that in turn spawn additional sub-subproblems, and so on. In no time, affairs become bogged down with the sheer magnitude and complexity of it all.
Sound familiar? Apparently in their efforts to solve a problem too big for one processor to handle, scientists have created the electronic analog of that very mechanism by which procrastination in its purest form is achieved and through which confusion and disorder are propagated -- the human committee. If history lessons are but remotely useful, the project is doomed to fail.
Similar in nature to its human counterpart, the electronic committee functions well if the number of CPUs is held to a manageable level and there is a clear division of responsibility. In a machine with two processors, for example, one chip might handle data input and output while the other performs computations. This has proven to be an extremely efficient arrangement in many currect applications, particularly if the processors get along well together.
When large numbers of CPUs are assigned similar tasks, however, turfs begin to overlap, the territorial imperative emerges, and the committee phenomenon prevails. Rather than concentrating on the job at hand, processors become preoccupied with manuevering to establish or reinforce their personal positions. Unity and cohesion disintegrate; individual units act out of context. As a result, consensus is never reached and there is little or no progress toward resolution of the fundamental problem. The outcome is a diluted compilation of the views of all members -- appeasing everyone, satisfying no one. I foresee a typical fifth generation computer solution, in fact, as containing only inoffensive generalized results laced with disclaimers and caveats and finally, a recommendation that additional committees of computers be convened to continue the investigation.
It would appear as if the outlook for the technology of parallel processing is not without severe limitations -- a fifth generation pipe dream. Not to despair, however. Almost invariably, we manage to accomplish the improbable, despite overwhelming evidence to the contrary. And that is as it should be. Otherwise, we would not have crawled, thin-legged and scaled, from the primordial sea, much less come to the wonders of art and literature, and of science.
So, just in case, I'm keeping a clear space on my desk at the office for my electronic replacement.
And my fishing gear ready at home.
Computer maven KENNETH BALLIET is an engineering manager with the FedGov. He has written pieces for the Commodore computer user's magazine, as well as Mother Earth News and the Mensa Bulletin.