THE FAILURE OF THE DIGITAL COMPUTER

     Remember the digital computer circa 1963? There were banks of flashing lights and spinning reels of magnetic tape drives. Able to do anything with sufficient programming, we looked forward to an age where almost anything could be done by a computer, efficient and infallible. There was a television show called "The Twenty-First Century" where Walter Cronkite extolled the virtues of our technology future and the computer was a central theme.

     There is a growing sensation that something has gone wrong. I find it sad that what was obvious in my circle of friends in 1982 is just becoming visible to the so-called experts in the field. Lawyer jokes are being replaced by programmer jokes. The recurring theme of computer conversation in the twenty-first century is how fallible computers are.

Recognition of the Problem

     The first stage in solving a problem is recognizing that a problem exists. The computer does about as much for our society in 2003 as it did in 1973. We may use computers more than we did then, but Microsoft Word does not produce better documents faster or cheaper than an IBM Selectric typewriter, spreadsheets do not produce better accounting or data analysis than paper charts or punched cards, and PowerPoint presentations do not communicate better than hand-drawn viewgraph or slide presentations. If we count the benefits of e-mail and the World Wide Web and we discount the cost of unreliable software, then I feel it is safe to say that the digital computer does as much for us as it did thirty years ago, neither a lot more nor a lot less.

     We certainly spend more on computers. I don't have actual employment records, but the computer community in the United States seems to have grown from about six thousand to six million. (This is the programming and service community, almost all of us are users in one form or another.) Since the computers themselves cost a lot less, let us settle for a factor of one hundred. That is, we spend 100 (20 dB) times as much on computer and computing now.

     The computers are certainly better and cheaper. The hardware people are making giant strides and every time we think we're getting close to the limits, somehow the technology engineers find a way to do more for less. I would argue that our computer hardware is one thousand times for cost effective in computing speed, memory, and disk storage space. That is, we get 1000 (30 dB) times more computing power for the same cost now.

     Finally, we are solving the same problems now as then. We are computing, analyzing, and sorting data just as we did thirty years ago. I would argue that the computer pioneers, and those who paid them, would not have invested so much without a firm conviction that this was work that would endure. The tasks that we are doing with computers, our ability to reuse the work done in the past, should make those problems one hundred times easier. That is, those who program have a job 100 (20 dB) easier now.

     Doing the full accounting of this analysis, we have lost about 10,000,000 (70 dB). To put things into perspective, that 70 dB factor is the difference in light intensity between the bright daylight sun and a crescent moon. When Arthur C. Clarke compared the reality of computing in 2001 to the computer HAL in his book 2001: A Space Odyssey, he pointed out how wonderfully small and powerful computers were but how disappointing computer programming had become.

     A computer professional who loses 1 dB, 26 percent, of his salary is upset. But the profession has lost 70 dB, 99.99999 percent, of its ability to deliver the goods, and people are just now complaining about it. I think it is safe to say that something is wrong.

Reasons for the Problem

     So we recognize the reality, computers are not fulfilling their promise, not delivering the goods, not working as planned.

     In his Hitchhiker's Guide to the Galaxy books, Douglas Adams wrote about Frogstar, a once-thriving planet that had developed an obsession about shoes so they spent more and more of their economy on shoes. As people insisted on more and more footwear, the quality of the shoes declined and they wore out faster so yet more people turned to shoe production. Eventually, the entire economy of the planet was devoted to producing shoes and it still was not enough. The planet reached the shoe event horizon, its economy collapsed entirely, and it became a barren wasteland.

     We are reaching the software event horizon. Somehow we have employed more and more people to write software of less and less quality to satisfy an insatiable market for computer programs. As these programs perform more poorly than their predecessors, we find ourselves buying upgrades more frequently, and the demand for programmers increases to compensate for their lower effective productivity.

     Programming used to be a fine art, something reserved for a few specialized craftsmen. A programmer was expected to be fluent in the specific language of his computer and in the general language of mathematics. Conversion from base-ten to base-eight (octal), base-sixteen (hexadecimal), or base-two (binary) was a natural notion. The rarefied notions of what is today called "complexity theory" were normal concepts in algorithm design at the computer center.

     Today we have a different programming community. It is considered reasonable to be a computer programmer without proficiency in mathematical or algorithmic reasoning. I find myself explaining to self-proclaimed programmers why a system with two yes/no inputs cannot discern and produce five different outcomes. I find myself explaining distinctions, getting resistance that "I already know that," finally getting the idea across, and still getting software that produces the wrong answer.

     I find myself dealing with smart people who are comfortable not knowing how their systems work. When two components interact in an unforeseen and undesirable way, I want to know which elements of the system are causing the clash. I make changes in the system to diagnose the problem so I can prevent a recurrence. I find members of the new community make changes in the system until the problem goes away and never mind if it will just recur a few days later with new input data.

     We have a tool user community who expects all tools to be "dumbed down" to the lowest level. We used to say making a tool foolproof would create a tool only a fool would want to use. Spending a month writing program so that a thousand users can save a day learning how to use it sounds very efficient, but it is often a false economy. The community of people using a program are more effective if they understand something about the program. If a thousand users save a day learning how to use a program and then double the time it takes them to use the program because of that ignorance, then we spend more programmer time to spend more user time.

Resources for Programming

     I don't know how many people were actually involved in programming computers in 1963 or in 2003, but a leap from 6000 to 6,000,000 seems about right, at least here in the United States. But what increase in productivity is represented by the thousand-fold increase in body count?

     I would argue that in programming, like most things, there are a few people who have the knack. On any sports team, aren't there one or two people who seem to have a wealth of talent? How many pilots at the airport fly like Chuck Yeager or Bob Hoover? Of the tens of thousands of kids taking violin lessons, how many play like Jascha Heifetz?

     I would also argue that in programming, unlike most things, most of what gets done gets done by people with the knack. One athlete does not make a team, one pilot is not an airline or a fighter squadron, one musician does not make an orchestra, but one good programmer can do enormously more than a less capable team. In spite of all the babble about not being able to get a baby in one month by impregnating nine women, most people want to believe that team efforts produce the best results. Here in the United States we gear our entire corporate lingo around teams and teamwork.

     While cooperation is a good thing, software is a natural monopoly of talent. One good programmer can out-produce thousands of not-so-good programmers. The extra overhead of group interaction, defining interfaces and managing timelines, makes a single good programmer more productive than thousands of others.

     In short, the 6000 people who are attracted to computers and computing as it was presented in 1963 represent much, perhaps most, of the talent in programming. Expanding the crowd a thousand-fold to 6,000,000 might increase the total talent tenfold. Expanding the crowd to the entire world of 6,000,000,000 might increase the total talent another tenfold. And that's it—there is no more programming talent to be found on this planet. Like oil or zinc or diamonds, once we find all the good programmers in the world there are no more.

     Let us consider the job of writing a computer program to do a job for somebody else. In 1983 I wrote a program called AUTOGROW that calculated growth of cellular telephone systems to meet growing demand. Using expertise I accumulated over months of working with the best cellular engineers, it took me a week to write the algorithms for the first version of AUTOGROW. I wrote a programmer-style interface that asked seven questions of the AUTOGROW user, parameter inputs. Because I had a small community of knowledgeable and friendly users, I could stop there. The next six months of AUTOGROW effort went into better algorithms, displaying cellular layouts so people could understand them, and doing studies for system and financial planning.

     The next phase of user-interface would be a menu-driven system that guides a user through the program parameter choices. A menu interface would have taken weeks to write, going over the choices, figuring out the options, and working with users to make the system "friendly."

     The final phase of user-interface would be a graphical user interface (GUI), the sort of point-and-shoot system that is popular today. That would have taken many months to design, program, and test. And when the GUI was all done and perfected, at best it would give the AUTOGROW graphical user the same capability as the seven dorky questions that AUTOGROW users answered for the fifteen years of its life.

     I call it the UI pyramid. Designing a programmer-friendly interface takes one unit of effort, designing a text-based user-friendly interface takes ten units, and designing a post-modern graphical interface takes 100 units. The GUI that took 100 times as much effort to produce makes it easier for people who do not understand. But rather than squandering very-scarce programming resources to cater to people who do not understand, wouldn't it make sense to educate those users so they have a sense of what the program is doing? Then we could produce ten times as many useful programs.

     There is a time for graphics. Finding something on map or seeing a statistical relationship is usually easier with a graphical display. To paraphrase the old cliche, a computer picture is worth 1024 words. But when the only purpose of a graphical interface is to make things pretty, then maybe we should skip it. GUIs are a little bit faster for novices and a lot slower for experienced users. Why expend extra resources to slow down the people who use a program the most often?

     There is another side to the resource-squandering equation of graphical interaction. A text-based interface can be designed to work from a script of input commands. The command script can be stored in a file. When a DOS or UNIX user calls and asks how to do something, the helper can send a command file. When a Windows user calls and asks the same question, we have all heard the conversation from the cube next door: "Go to the desktop and double-click on the icon. Do you see the ‘file open’ bar? Okay, click on that and look for ‘properties’ and right-click on that. See the waving bunny rabbit? When he turns purple, double click on the ears and wait for the menu to come up." Come on guys, we waste so much time on this kind of frustrating communication, can't we put that into a script file?

     We are reaching the software event horizon, if we are not there already. Every resource we can find, every smart person we know, is busily generating software to satisfy an insatiable demand.

Recovery from the Problem

     There is a saying used to justify many of the truly awful things in the post-modern era. If you do things the way you used to do them, then you will get things the way you used to get them. We used to get ten million times (70 dB) more efficacy from our computers than we do now. It's time to do some things the way we used to do them to get the results we used to get.

     Get smart young people interested in programming. There is a pool of innate talent out there in the world's young people. In spite of how dumb they may seem to us, I believe today's young people have just as much genetic potential to be smart, effective, and useful human beings as we did when we were their age. (And don't forget that we did a few dumb things in our day.)

     I remember the computer projects I did as a high school student: writing a program to play Nim and another to generate mathematical hexaflexagons, both with "graphical" displays on a teletype terminal, When our Latin teacher gave us an assignment to scan many pages of ancient Roman poetry, to mark each syllable as long or short, a friend and I enumerated the specific rules and wrote a program to do it so we could type in the poetry and the computer would type it back with long and short marks on top of the text. My final high school programming project was a program that would play the card game of Hearts. All of these were written in BASIC using a teletype terminal with round, gray, elephant-foot keys and a 110 baud modem.

     Today's young computer users download and learn to use graphics programs, design intricate and sophisticated web pages, and play elaborate computer games. While all of these exercise their minds, these are fundamentally different from the programming challenges we chose. Specifically, these newer intellectual activities do not lead their participants into the programmatic and computer-organized thinking that produces decision-support software, or solid, robust, simple programs in general.

     Design tools for smart people. In the electronic, computerized office of the twenty-first century we have surrounded ourselves with a host of computerized office tools designed for users who do not work well with computers. It is an attempt to make an IBM personal computer work like a MacIntosh.

     The whole mouse concept went the wrong way, in my opinion. It was designed as a high-tech solution at Xerox, a three-button screen pointer matched to a five-button keyset for eight bits straight into the machine. The skilled mouse-cum-keyset user knew the American Standard Code for Information Interchange (ASCII) and could input text directly. The keyboard was used for large blocks of contiguous typing, such as the text I'm typing now. Editing was done with mouse and keyset.

     Then the mouse was dumbed down to a point-and-shoot interface (or grope-and-grunt as I call it) that makes it easy for dumb people (who don't know what is going on) to use the system. The trouble is it doesn't make it any easier for smart people than dumb people. As the dumb people learn the system and become smart people, even their efficiency is diminished by the claimed-easy-to-use, graphical interface.

     At the text level, we have systems like Microsoft Word. It may be intuitive for somebody who has never used a computer to generate italics by leaving the keyboard, grabbing the mouse, groping for the [I] box. It may be more intuitive, the first three times, than typing <i> in HTML or \it in TeX. By the fourth or fifth time, the extra time is annoying and we learn the Word shortcuts including control-I for italics. But even if I learn the shortcut, I have to go through all of them each time I type a formula like Eb/N0 because it is difficult or impossible in the point-and-shoot world to define abbreviations or to make global substitutions.

     This is more than a preference of one tool over another. Once we understand that it is the smartest 0.1 percent doing most of the useful work, we should realize that it is not only desirable but essential to use tools that use their time efficiently.

     This goes far beyond Microsoft Word and beyond Excel, PowerPoint, Visio, and all the other tools designed for novices (and designed to keep them that way). This mindset also applies to software development systems, so-called integrated development environments (IDEs), that makes developing software easier for people who do not know how to write programs. Pardon my pedantry, but people who don't know how to write programs should get a text editor, a command-line compiler, and a good book and learn how to write a computer program to do a job. There are tools like make and grep that make the job easier for both dumb and smart people.

     Make systems last a long time. The investment in computer programming, like any other high technology, is large. It was large in 1973 to pay for mainframe computers to be purchased and maintained and it is large in 2003 to pay for large programming and infrastructure-support staffs. The break-even payback period for any given piece of computer software depends on its application and user community, but I expect it takes years for a program to pay for itself. The same goes for the effort of learning how to use a user interface.

     Consider the investment in writing a paper in Word or a spreadsheet in Excel. Now consider the likelihood that this paper or spreadsheet will still be useful in ten years. By having systems change file formats and user interfaces every six months, software vendors pad their own nests at the enormous expense of their user community. Continuing to patronize these vendors, to use these software packages for information more easily and effectively managed in simpler ways is irresponsible.

     My own record has been phenomenal in this regard. AUTOGROW, my first real computer program in the workplace, lasted fifteen years serving various purposes in cellular radio telephony. My airline planning tools written from 1990 through 1994 almost all enjoy frequent use in 2003. A graphical tool I wrote one afternoon in 1999 for analyzing an interlocking (the railroad term for multiple track intersections) is still in regular use. The fleet assigner I wrote in 1990 was made obsolete by mathematical software breakthroughs a few years later, but its data organization still stands as the foundation of the newer models. I'm not sure I know all I did right in writing these programs, but I am sure that my attitude, my approach to programming and system design, and my choice of tools have a lot to do with it.

     Put program structure into the program, not the tools. The invention of the COmmon Businessman Oriented Language (COBOL) for accounting software and the FORmula TRANslation (FORTRAN) language for mathematical computation during the 1950s set the stage for decades of computer programming. These languages say what needs to be said with a small vocabulary exquisitely used.

     Since that time, the programming community embarked on a quest for new and better languages. They gave us ALGOL, APL, BASIC, C, C++, Forth, InterCal, Java, Modula, Pascal, SNOBOL, SQL, and Visual Basic. Many of these languages add value to the computing and programming universe, BASIC for introducing young people (including myself) to computing, ALGOL for expressing algorithms, InterCal for creating the mindset of post-modern programming, but they are not tools I would use every day. To work with the post-modern software community, I use a carefully engineered subset of C that looks like FORTRAN.

     For the work I have done and cherished, FORTRAN says what should be said just as Shakespeare's English has said what needs to be said for five centuries. Writing algorithms in C++, Java, or Visual Basic has the same appeal as writing sonnets in Esperanto. Think what a wonderful world we would have if half the energy spent on new languages had been spent on writing better programs in the languages that we already had that work so well. I'm no Shakespeare, but my FORTRAN/C source communicates on several levels.

     The intricacy and complexity of my programs is woven into the code that I write rather than hidden in the language itself. This is more than a preference of one tool over another. Once we understand that the post-modern languages obfuscate rather than illuminate, we can return to a style of programming that builds tools to last for years and to adapt to new user needs for decades.

     I would also argue for well written software in simpler languages on the basis of performance. In today's world of faster computers and larger programming budgets, the notion that C++ runs thirty times slower than FORTRAN and takes at least that much more development time seems like worrying about gas mileage with today's low oil prices. Instead of seeing faster computers as an excuse for dumbed-down tools and sloppy programming, I have seen faster computers as an opportunity to solve harder problems with more intricate algorithms.

     Have computer users become computer literate. Our customers in the programming world are the users of our software. As these customers clamor for simpler interfaces, we provide dumbed-down, graphical interfaces that make it easier for novices to use our software at the expense of smart or experienced users. This moves the programming effort to the wide base of the UI pyramid where most of the effort adds comfort rather than performance. My own customers soon learned that having ten times as much performance was comforting enough.

     What constitutes computer literacy? In the old days, it was finding the computer center, feeding a deck of punched cards into a card reader, browsing a magazine while the computer did its thing, and reading output on broad line-printer paper, usually with green and white stripes across the page. It usually entailed using a keypunch to type input parameters that went into the card reader at the end of the deck after the source code of whatever was running. In the more-recent old days, it entailed being using a command line interface, being able to create, copy, and delete files, knowing about directories, and using a text editor to enter data and to write command scripts. This was the case on machines and operating systems from International Business Machines (IBM), Digital Equipment Corporation (DEC), Hewlett Packard, and so on. Later on, UNIX and DOS followed the same paradigm.

     So what has changed other than the expectation that today's computer users are too stupid to type commands and too lazy to use a text editor? The atomic elements of files, command lines, and text editors form a basic vocabulary of using a computer. I have augmented that basic vocabulary with menus and graphics, but I never lost the fundamentals in the process.

     Other programmers have replaced the basic vocabulary with menus and graphical interfaces that keep their users in the dark. The result is that these users now have no idea how to get to the essential elements of their own work. This describes the Microsoft Office suite of programs as well as what I have seen in the Apple MacIntosh.

     This is more than a preference of one tool over another. Once we understand that files, commands, and editors are the fundamental tools, then we can build user interfaces with tremendous power. When we try to take away the basics, the results are predictably bad.

     Insist on excellence. This sounds like one of those post-modern business clichés, like "keep focused" or "work as a team," but it is a great deal more than the mindless use of a word like "quality." It is an attitude, a mindset, in designing and writing programs.

     When I was a youngster in modern times, the computer did not make mistakes. When something went wrong, it was attributed to "human error." The movie "Dr. Strangelove or How I Learned to Stop Worrying and Love the Bomb" made macabre fun of that expectation of perfection. The one thing that never occurred to anybody in that film is that the software might not work as planned.

     Today's computer community in post-modern times has quite a different expectation. If it doesn't work, then there must be a computer involved somewhere. The notion that standards are followed well enough so a user can read the manual, use a software tool, and get it right the first time is absurd today. When a user finally figures out all the tricks and gets it right, the new release of the tool breaks those workarounds so the user is back to figuring it out all over again. Comparing the modern TeX to the post-modern Word for document preparation seems silly, but there is still a community of Word users even after decades of abuse.

     This is more than a preference of one tool over another. Once we understand how much effectiveness we lose using inferior tools, it becomes imperative that we migrate away from sloppy suites of fancy tools and learn to use simpler tools that work reliably and well.

Reality

     Much of what is wrong with post-modern computers and computing is no accident. There is a large community of so-called Information Technology (IT) professionals who make their living from the inadequacies of the equipment they serve. In some sense they are no different from those who service our automobiles, people who profit from the fallability of our machines.

     But there is a charlatan aspect to today's IT community and computer consumers are being sold a bill of goods. If computers worked as well as promised, then most of the IT community would be doing something else, actually working for a living. And if consumers knew how far programming has declined, then they would not buy computers and most of the IT community would be doing something else, actually working for a living. So much of the IT community makes its living on a perception arbitrage. They will work hard to perpetuate the lie.

     There are good, decent, honest, effective computer professionals and there are shysters and charlatans out there. The proportion is somewhere between the legal profession and insurance salesmen. The likelihood of a diatribe like this one cleaning up the IT profession is about the same as a book on government corruption cleaning up graft in congress. It is up to each of us in the computing community, providers and consumers, to uphold these values ourselves and to insist on them from others.

This is a FLASH-FREE web site.
Today is 2018 December 19, Wednesday,
4:44:36 Mountain Standard Time (MST).
One visit to this web page.


  Wikipedia Affiliate Button

THE ADAM HOME PAGE

adam@the-adam.com