Why General Semanticists Should Distrust Theses Based on False Information.

The article "Why General Semanticists Should Distrust Computers" [by Raymond Gozzi, Jr., ETC, Spring 1999] was undoubtedly novel and interesting, albeit rather Luddite in tone. Unfortunately the author seems not to have run his work by someone knowledgeable about computers before going into print. "Unfortunate" because the whole premise on which the article is based is deeply flawed.

The Conclusion section of the article made the following assertions:

1. The computer is a strange device...

2. ...in which the map "is" the territory.

3. It uses a simplified number system which is built upon two-valued processes.

4. We need to face up to the possibility of a bias built into the supposed objective modeling capacity of the computer.

5. This bias may tend towards divisiveness, exclusion, and war.

6. We need to ask the embarrassing question: Is the computer insanely great, or is it just insane?

As I shall demonstrate, not one of these assertions is based on the facts of the case. On the contrary, they all depend upon a thoroughgoing misrepresentation of computers, their workings and their usage.

The Computer is a Strange Device ...

At first glance it might seem that the author means "the computer is a strange device because the map is the territory." But that isn't what the article actually says. No, "the computer is a strange device" is offered, as written, as a statement in its own right. But in what way is the computer a "strange" device? A computer is just a machine. It is no more "strange" than the first car was strange, or the first refrigerator, the first wristwatch or the first digital wrist watch.

What we have here is a highly judgmental and emotive description of "the computer." Viewed in combination with remarks made later in the article I cannot escape the impression that the words have indeed been selected as a means to prepare the reader's mind for the more serious accusations which will follow.

Where the Map is the Territory

The second section of the original article gives us a much fuller picture of the flawed logic that underlies the whole thesis:

Change the software and you have changed the machine.

With one program the machine becomes a word processor, with another program it becomes a photograph editor, with another it becomes a mathematical spreadsheet, and so on.

The machine "becomes" a word processor, a photograph editor, a mathematical spreadsheet and so on? In what sense is the author using the word "becomes," I wonder. And how does a computer "become" anything other than what it is?

Of course there are various software applications which can be loaded onto a computer which will perform word processing functions, graphics editing functions, spreadsheet functions, and so on. But nothing in the computer changes beyond the fact that the hardware is being used by one application or another. The data and the applications are stored and managed the same way. The VDU display is generated, and refreshed in the same manner. Bits and bytes of data are funneled through the same CPU (central processing unit) in the same way, under the control of the same operating system and the same processor instruction set regardless of what application is running.

Indeed, the fallacy of the argument is clearly illustrated by the fact that the current range of desktop computers can comfortably run several applications at the same time. That is to say, I can load up a word processing application, a graphics editing program, a presentations application, and an Internet browser, swap information between the first three, and view the results via the browser. Moreover, using a Windows operating system, I can have all four tasks displayed on my VDU screen simultaneously. In what sense, now, is my computer "becoming" a word processor, a graphics editor, a presentations application, or an Internet browser? And if it is background printing a text document whilst in the foreground I work on a spreadsheet, when does the computer "become" a wordprocessor and when is it a spreadsheet?

Since the initial logic is flawed, it is hardly surprising that when we come to the suggested metaphor that a map (a software program) is the (computer) territory, we find that it is wholly inappropriate. A program is no more a map to the computer it runs on than a car is a map to the road it runs on.

The Multi-valued Proposition

It is indeed true that the basic functioning of a computer is digital, each "bit" being set to either one or zero, or more accurately on or off. But if that was all the computer could do, it wouldn't be much of a computer.

In the original article the author offered us the metaphor of the two- and three-colored traffic lights, arguing that "the actual dynamics of the intersection required a more complex logic than the simple digital "stop-and-go" or the red and green lights." But does the addition of an intermediate yellow light really provide a significant modification of the on/off scenario, as the author suggests? Most definitely not. Each light is still either "on" or "off," just as the "bits" stored in the computer's memory are either "on" or "off." Of course the extra light means that, one level up, we now have six possible settings instead of three: red only, green only, yellow only, red and yellow, green and yellow, red and green. But the same thing is true within the computer.

What the author of the original article appears to have overlooked is the fact that there are several levels of abstraction in the way a computer produces results.

At the most basic level we do indeed find the simple on/off state of the individual bits and bytes. But data is not dealt with primarily as bits but rather as bytes (a group of 8 bits). At this point we are no longer dealing with just l's and 0's (zeroes), but with values in the range 0 - 255.

One step up we see that the flow of bits and bytes is controlled within the microprocessor by tens of millions of "gates," including AND, NAND (not AND), OR, NOR (not OR), XOR (exclusive OR) and so on. At the next level is the CPU, the focus of all activity within the computer. Each CPU has an instruction set which determines what it can actually do with the bits and bytes it processes. These instruction sets vary from manufacturer to manufacturer and from generation to generation of microprocessor.

Another step up we come to the operating system - the "middleman" between the individual applications and hardware of the computer and its peripherals - devices such as printers, scanners, disk drives, etc.

After the operating system comes the programming language used to create the application: C++, Pascal, PROLOG, Java-Script, etc. Some of these languages are procedural, some are declarative; some are compiled, some are interpreted.

And still we aren't finished, because there are plenty of people who have a finger in the pie, so to speak.

There are the analysts and designers who create the specification for each application. There are the individual programmers - some write efficient code, some programmers write very inefficient code. There is the person or people who decide what data will be fed to a particular application. There is the person who designs the reports that (very selectively) record the results from a given application.

In short, the claim that computers are "biased toward modeling situations and solutions in adversarial terms" because "the logic of the excluded middle is a logic of division and exclusion" is an assertion based on a fundamental misconception about the nature of computer processing. And the errors don't stop there.

Bottom Line Thinking

Having laid his foundations, the author now tells us that, "As computers are used for more decision-making processes, we can expect a persistent, unacknowledged divisiveness in their models."

Can he substantiate this claim with firm evidence? No. On the contrary he admits "I can only point to the behavior of the institutions which have used computers the longest: the United States military, big multinational corporations, and what has been called 'Big Science'."

He goes on to say that, "The behavior of multinational corporations has been guided partly by computer models since the 1950s. Has this behavior tended toward division and exclusion, rather than unification and inclusion? I think both processes can be seen in the behavior of these gigantic organizations, but which predominates?"

This paragraph alone demonstrates not one but four important flaws in the author's argument:

1. There is no such thing as "a corporation" in the sense of an autonomous entity which can make decisions. In this respect a corporation is a collection of people. It isn't "corporations" which are guided by computer models, but the human beings who are the decision makers within corporations.

2. Even if we accept the argument that the decision makers have been influenced by computer generated-models, we have to ask what process is involved. Clearly the computers don't plug straight into the decision makers' brains. In reality the decision makers view the computer output and form their own conclusions. Thus the implied argument that computers control corporational policies, even at second hand, is unsupportable.

3. If computers have the inherent bias suggested in the article, how do they manage to dictate both "division and exclusion" and "unification and inclusion"? Unless we suppose that any bias towards "unification and inclusion" is so freakish and infrequent as to be statistically irrelevant, then we must admit that bias that works in both of two diametrically opposed directions is really no bias at all.

4. Most important of all, computers do not create "computer models" out of the blue. "A computer," as one writer noted, "is nothing more than a fast idiot." By which he meant that computers do nothing but what they are programmed to do. GIGO - garbage in - garbage out - is a fundamental tenet of computer usage. Computers have no inherent bias, but computer programs do. Be they expert systems, MIS/EIS (management information systems/executive information systems), data warehouses or anything else, computer programs do indeed have bias built into them - by the analysts, designers, and programmers who create them, and just as importantly, by the choice of the data they are given to process.

Of course we might argue that the very latest technology, such as neural networks, are not programmed in the normal sense and therefore do not suffer these limitations. But we'd be wrong, as this story shows:

A little while ago the military commissioned an "intelligent" weapons system that could locate, recognize, and destroy enemy tanks. The system was to be based on a neural network which would be able to identify the targets. Now the process of initializing a neural network involves showing it hundreds, maybe thousands of examples of whatever you want it to recognize, along with examples that are not whatever you want it to recognize. The idea is that the neural network makes a guess about the identity of each example, and the "programmers" tell it whether the guess is correct. Bit by bit (pardon the pun) the neural network "learns" the distinguishing features of whatever it has to recognize.

Anyway, the weapons system in question was built, and tested, and finally taken out for an "in the field" demonstration before the "top brass." And it failed. One after another, various tanks were driven into the system's visual field, and one after another, the system ignored them all. Why? Because it was raining during the live test, and the only pictures the system had seen up till then had been created on clear, sunny days.

To repeat the point - computers may generate bottom lines, but only under the instructions of programs designed and created by human beings, using data selected and entered by human beings, to meet requirements specified by human beings. The "bottom line thinking" that the article refers to is human thinking, not computer thinking. Even if computers were inherently limited by "the Aristotelian logic of the excluded middle" - and I hope I have demonstrated the fallacy of this argument - this would be irrelevant since computers don't generate bottom line thinking or bottom line logic in the first place. Computers do what they're told, with what they're given. It is human beings who control every step of the process, especially the way in which the computer-generated results are used.

Conclusion

In the penultimate section of the original article the author offers the hypothesis that:

Because digital computers are constituted by circuits embodying binary, off-on, two-valued logics, the models and other products of digital computers will have a bias toward emphasizing processes of division, exclusion and conflict; and will have a bias against emphasizing processes of joining together, inclusion, and negotiation."

This hypothesis is allegedly supported by the six basic statements which I quoted at the start of this article. I suggest that we have seen that all six statements are in error, and that the hypothesis fails in every important respect, thus:

1. The computer is a strange device ... This is an emotive, judgmental assertion, not justified by the evidence.

2. ... in which the map "is" the territory. This conclusion is based on a misunderstanding of the relationship between computer hardware, operating systems and software applications. The conclusion is not justified by the evidence.

3. It uses a simplified number system which is built upon two-valued processes. This is a seriously misleading oversimplification of the real nature of the way in which computer hardware functions, leading to erroneous conclusions, not justified by the evidence.

4. We need to face up to the possibility of a bias built into the supposed objective modeling capacity of the computer. Since "the computer" has no inherent modeling capacity, it clearly cannot be biased. What bias exists in the results generated by computers, and such bias certainly does exist, is entirely attributable to the limitations of the human beings who create the programs, select the data and interpret the results. The conclusion is not justified by the evidence.

5. This bias may tend towards divisiveness, exclusion, and war. Again, the bias is not the responsibility of the computer hardware but of the humans who use it. The conclusion is not justified by the evidence.

6. We need to ask the embarrassing question: is the computer insanely great, or is it just insane? The list of conclusions ends, as it began, with what appears to be nothing more than empty rhetoric. To whom is this "question" actually addressed? On what basis can the question be characterized as "embarrassing"? In what way is it embarrassing? Who will be embarrassed by either the question or the answer? And in what conceivable sense can any 20th century computer be said to be "insanely great," or even "just insane"? For the sixth and final time, the conclusion is not justified by the evidence.

Andrew Bradbury

Hove, East Sussex, United Kingdom

The author notes: "From 1988 to the present I have been working in the systems integration division of a major telecommunications company as Training Consultant and Technical Author, with a special interest in Knowledge Management. In addition to the documentation I produce as part of my work - from training manuals to 'white papers' and end-user documentation - I have had five books on computer programming published in the UK."

Should We Still Distrust Computers? A Reply to Bradbury

I would like to thank Andrew Bradbury for his interest in my article "Why General Semanticists Should Distrust Computers," and for writing a rejoinder. I intended the article to be provocative, and I am glad that one person, at least, was provoked enough to articulate a reply! I believe these are important issues, and I welcome an opportunity for dialogue.

I will discuss Mr. Bradbury's critique of my two claims first (that computers may have a bias toward division and exclusion, and that "the map is the territory" for a computer). I will then discuss three of Mr. Bradbury's claims about computers, which I think are overoptimistic.

A Hypothesis, Not a Conclusion

I think Mr. Bradbury has perhaps not understood that my "Digital Division Hypothesis" is just that - a hypothesis. I regard it as a proposition to be tested. It could be right, it could be wrong. It reads:

Because digital computers are constituted by circuits embodying binary, off-on, two-valued logics, the models and other products of digital computers will have a bias toward emphasizing processes of division, exclusion, and conflict; and will have a bias against emphasizing processes of joining together, inclusion, and negotiation.

I do not think all products of digital computers will exhibit such tendencies, but I think some products might - particularly the kinds of sophisticated calculations that go on in making decisions in game theory, or strategic projections in complex situations. The mathematics of fractals tells us that when processes are repeated many times, similar shapes appear at different levels of the structures generated. The computer certainly repeats binary processes many times - do these two-valued shapes reappear at higher levels of their functioning?

Ultimately this is an empirical question, to be determined by testing the products of digital computers against, for example, hand calculations (using a base 10 number system), or perhaps against the outputs of analog computers. If this hypothesis finds no support in such instances, I am quite ready to give it up as a bad idea. But I think it should be tested, and I pointed out various reasons it might be plausible. Mr. Bradbury treats the Digital Division Hypothesis as a "conclusion ... not justified by the evidence." No, it is a hypothesis, not a conclusion, and I'd like to see more evidence.

Is a Program a Map?

Mr. Bradbury also takes issue with my contention that a program is metaphorically a map which guides a computer's operation. He says "A program is no more a map to the computer it runs on than a car is a map to the road it runs on." I must confess this analogy eludes me, requiring two different senses of the word "run." It would also be more consistent to say in the second clause that "... a road is a map to the car that runs on it."

Is the Map the Territory?

The real issue which has upset Mr. Bradbury in this instance is my contention that, for the computer, "the map is the territory." When a different program runs on a computer, it "becomes" a different machine. Mr. Bradbury is correct to point out that, in most cases, the operating system and the more basic machine language functions remain the same, regardless of which applications program is running. He provides an "abstraction ladder" for different levels of functioning of a computer. I appreciate his comments and accept them, up to a point.

Mr. Bradbury asks:

... how does a computer "become" anything other than what it is. Of course there are various software applications which can be loaded onto a computer .... (b)ut nothing in the computer changes beyond the fact that the hardware is being used by one application or another.

I can feel my colleagues who support E-Prime (the use of English without the verb "to be") gritting their teeth at this. The simple verb "is" hides some very complex questions about what constitutes a computer. A computer "is" both hardware and software, both material elements and patterns of electrical impulses. The functioning computer usually blends several kinds of hardware and several kinds of software into a dynamic system. When viewed as a functioning system, it is difficult to separate out the pieces and say that one is separate from the others.

Therefore, I think that the different "steps" on Mr. Bradbury's "abstraction ladder" are not always as separate as the metaphor implies. For example, we read occasionally about "Trojan Horse" programs which are sent as e-mail, but which can take over (and reprogram) other applications and the operating system as well. (Clifford Stoll's The Cuckoo's Egg (1989) was written about such an intruder, which took over the entire operating system of the Lawrence Berkeley Labs mainframe.) More recently, the "Chernobyl Virus" even rewrote the BIOS chip - the basic input-output system - on infected computers. So the lines between "levels" of software are not as sharp or as inviolable as Mr. Bradbury implies.

And what of networked computers, particularly where much of the software and operating instructions for the computer reside on the network? Where do we draw the lines to determine what "is" one computer? Is just "one" computer the proper unit of analysis in this case? Our conceptual boundaries, so neatly inscribed in our language, blur when dealing with computers and networks.

I think Mr. Bradbury is right to object that there are portions of the computer-territory which do not necessarily conform to the program-map. I thank him for pointing this out. However, I think sometimes the program-map does determine more of the computer's functioning than Mr. Bradbury allows. And I do not follow him into his static definition of the computer, assuming that it "is" what it "is."

Fantasies of a Pure Technology

Mr. Bradbury makes a series of cogent arguments against various of my claims in the article. These arguments are often advanced implicitly by defenders of particular technologies, Mr. Bradbury is to be credited for making them explicit, where they can be discussed. I would list them as:

1. "A computer is just a machine. It is no more 'strange' than the first car was strange. ..."

2. "Computers have no inherent bias, but computer programs do."

3. "Computers do what they're told, with what they're given. It is human beings who control every step of the process, especially the way in which the computer-generated results are used."

I would like to discuss each of these positions, because I think they embody widely held fantasies about the nature of computers and their influence in the world. I think each is subject to serious question.

Is the Computer Strange?

In my article, I called the computer "a strange device." Mr. Bradbury indicates that he has many years experience as a technical author and training consultant in the computer field, so perhaps he is used to these machines more than the rest of us. But I would like to point out some aspects of computers which are so unique that I think it is accurate to call them "strange."

Current computers run on chips, small squares of silicon. The Pentium Pro processor, released in 1995, contains 5.5 million transistors on a space of a couple of square inches. There is nothing else human-made which contains so much in such a small space - except newer chips, which are getting even more components on them.

Further, these chips operate extremely fast. A PC chip released by IBM and Motorola in 1996 operates at 200 MHz - meaning it runs on 200 million pulses per second. The DEC Alpha workstation operates at 500 MHz. These times are so incredibly short that we have nothing to compare them with in our experience. Almost by definition, these devices are strange - not like anything else we know. (By contrast, the first car was much like a wagon or carriage.)

Do Computers Have No Bias?

I agree with Mr. Bradbury when he points out that computer programs have biases. However I disagree with the claim that computers themselves have no bias.

The image of a neutral technology, which has no inherent bias, is often advanced. Yet, when examined, it is clearly a utopian and unrealistic claim. As Neil Postman points out in Technopoly (1993), every technology exists to do something, and not do other things. A technology exists to construct the world in a certain way and not in other ways. This means it is biased toward some ways of functioning and not others. Furthermore, this means it will facilitate some kinds of human behavior, and discourage others. (This does not mean that a technology cannot be used against its bias - it just means that the technology will tend to make certain uses easier, and certain other uses more difficult or less common.)

As General Semanticists should be aware, all symbol systems have biases. They involve abstractions which require making choices. I argue that the binary number system embodied by digital computers contains (at least one) bias - toward division and exclusion. But even if this bias is adequately compensated for by the functioning of the computer, as Mr. Bradbury contends, I would say that the technology still must have a bias - every technology does. Postman (1993) argues that the computer is biased toward seeing the world in terms of numbers, and this bias leads to the claim that all our problems can be solved by fast calculations. (Needless to say, he disagrees with this claim.)

Do Human Beings Control Every Step of the Process?

It would indeed be reassuring if humans did completely control the computer systems they have created. However, we almost daily see evidence that these systems are to some extent beyond human control. I am writing this before the turn of the millennium, so I do not know how serious (or not) the Y2K problems will actually be. But the very degree of concern, and the amount of money being spent on this problem, indicates that many people are starting to realize how much control has slipped away from humans.

Also, the frequent reports of "viruses," "worms," and "Trojan Horses" roaming the world's computer networks would suggest a lack of control.

Do humans always control the uses computer outputs are put to, as Mr. Bradbury claims? As I recall my readings from Organizational Communication, the claim "the computer says so" has been a powerful rhetorical device for years in the management of many firms.

Summary

In sum, I would not be so ready as Mr. Bradbury to assume that we can control computers - particularly networks of computers. I am also reluctant to draw the sharp lines that Mr. Bradbury does between the functioning patterns of computer systems and their "identity" as computers. I do not share the utopian belief that the computer has no bias at all. It is toward the end of discovering the computer's bias that I have proposed the Digital Division Hypothesis, to be tested and either confirmed or rejected.

I would like to thank Mr. Bradbury again for his thoughtful critique, and providing the opportunity to clarify and debate some of these issues. I think they are important issues, about which reasonable people may disagree.

Raymond Gozzi, Jr.

Ithaca, NY

REFERENCES

Postman, N. (1993). Technopoly. New York: Vintage Books.

Stoll, C. (1989). The Cuckoo's Egg. New York: Pocket Books.