Email me:

A Computer's Eye View (from 1972)

This excerpt is from Man and the Computer by John G. Kemeny, published in 1972 by Charles Scribner's Sons. This except is from Chapter 4: Division of Labor, p. 44-47.

A Computer's Eye View

In the preceding chapter time sharing was examined from the point of view of human beings. It will help provide a different perspective on time sharing to examine it from the point of view of the computer. Since people have great difficulty in adjusting their minds to the incredible speed of the computer, I shall provide an analogy.

Let us suppose that space travelers discover a higher order of living beings. These creatures are much more intelligent than men, but their metabolism is much slower. We recognize their great insights into problems and hope that they will help us to resolve a number of issues currently beyond human capability. On the other hand, they are so frustratingly slow that they must rely on our much faster response time in order to get the job done.

Specifically, I will assume that the amount of routine mental work (for example, computation) that a man can do in an hour will take these creatures a hundred years. They marvel at our great speed in carrying out calculations and yet they look down on us as lower order of intelligence, whose strength lies in routine tasks, but which lacks their fundamental insights.

Suppose that a man goes to work for such a creature, hoping to combine the creature's higher intelligence with man's much greater speed. The relationship is quite frustrating. First of all, the creature takes two years to formulate the problem for his human companion. Man promptly goes to work and carries out the required calculations and arguments in a couple of weeks. When he presents his results to the creature, he must again wait months or years until the creature has absorbed these results and is ready to make a further suggestion. Needless to say, the human worker feels that his talents are wasted. He therefore offers his services to an entire tribe of creatures and finds he can keep up the requests of one hundred of them.

I will make the following assumptions about the various times involved in this analogy: The creature takes from three months to five years to get ready to enter a new request. The exact time may depend on how useful he finds the human results and how deeply he wants to think about the next step. Let us say the average "turnaround time" is two years. The task is such that it may require from one day to a month of human effort, and will assume that on average it takes a week. Under these assumptions the human being can give good service to one hundred of the higher-order creatures.

The strategy will be as follows. He will work on a given job for a few days to see if it can quickly solved. If it can, he reports his results and turns to the next task. If he has failed to complete the job in—let us say—five days, he will set his calculations aside for awhile and work on some other creature's request. It is not efficient for him to work on each job until it is completed, because he may need to put in thirty days on a difficult task while the next task could have been completed in a single day. Since his main motivation is to give excellent service to his employers, he will try to complete short jobs as quickly as possible. It is not unreasonable for a creature that needs a full month of calculation to wait longer for the results, but one who needs only a day's computation expects his answers in a hurry.

I have tried out such a system on a computer (in a mode known as "simulation," which is discussed in Chapter 11), and it appears that the human being would be able to give excellent service to all 100 employers. I have simulated the behavior of the human-creature system over a period of twenty years. On the average, each creature has to wait two months for his results. Since he then takes from several months to several years to decide what to do with the results, this will seem like very fast service. In twenty years creatures can on average have nine to ten jobs completed by their human servants. Since they take two years on the average to decide what to do next, it is clear that the service cannot be improved very much. Each creature will be spending more than 90 percent of the time thinking about what he wants done and less than 10 percent waiting for the next result. This is a highly satisfactory working relationship.

The human being is kept busy all or most of the time. Occasionally a fairly long backlog of several months' work will build up, but typically he will have only a half dozen jobs to worry about at a given time. This is due to the fact that, from the human point of view, these intelligent creatures are so very slow in making up their minds about what they want next. In five different attempts to simulate the twenty year period, I found that the human being in the best case caught up with all his work at the end twenty years while in the worst case he had eleven jobs pending with a backlog of three and half months. The slowest of the creatures had only four jobs completed in twenty years, while some who seemed to response rapidly to human results succeeded in completing seventeen jobs, which is superbly fast action on the time scale of these slow creatures.

If the reader will now substitute computers for human beings, human beings for the more intelligent creatures, and reduce the time scale by a million, he will understand the computers' point of view about time sharing.

How far has computer technology advanced since this was written in 1972? In the proceeding chapter, Kemeny describes the then-new GE 635 machine that ran the Dartmouth Time Sharing System: the "dual processor system is capable of some 10 million multiplications per minute."

That works out to be 166,166 multiplications per second. (I'm assuming these are "fixed-point" multiplies: no decimals allowed.)

A Pentium 4 can do at least one multiplication per clock cycle; for a 1.7GHz P4, that's 1.7 billion multiplications per second. Rounding a bit, that 2002 Pentium 4 is about 10,000 times faster. (And if that GE 635 cost 10 million dollars—a wild guess, but perhaps not far off—then that $1000 P4 is also about 10,000 times cheaper.)

But here's the kicker: that GE 635 was shared among 100 people. The P4 is 10,000 faster, but it's all dedicated to you: so your P4 is effectively one million times faster than the machine you might have shared in 1972.

How would this increased speed affected Kemeny's example? Depends how you play it. If you kept the human speed fixed, that one hour of human work that used to take the creatures 100 years would now take the creatures one million years. If you assumed humans had sped up 10,000 times, that hour of work would now take about 1/3 of a second. Either way, the difference between human speed and computer speed is all but incomprehensible.

Last update: 4/8/2003; 9:01:52 PM.