The Dream Machine

by M. Mitchell Waldrop

“Those of you who seek salvation, stand up!” So he instinctively leapt to his feet—and smashed his head against the bottom of the keyboard. Instead of finding salvation, he saw stars. This experience, Lick would say, gave him an instant insight into the scientific method: Always be extremely careful in your work—and in your proclamations of faith.

Nonetheless, his vision of high technology’s enhancing and empowering the individual, as opposed to serving some large institution, was quite radical for 1939—so radical, in fact, that it wouldn’t really take hold of the public’s imagination for another forty years, at which point it would reemerge as the central message of the personal-computer revolution. Even more radical, however, was Bush’s second new idea, which was the mechanism for implementing his vision. Instead of storing those countless microfilmed pages alphabetically, or according to subject, or by any of the other indexing methods in common use—all of which he found hopelessly rigid and arbitrary—Bush proposed a system based on the structure of thought itself. “The human mind ... operates by association,” he noted. “With one item in its grasp, it snaps instantly to the next that is suggested by the association of thoughts, in accordance with some intricate web of trails carried by the cells of the brain. ... The speed of action, the intricacy of trails, the detail of mental pictures [are] awe-inspiring beyond all else in nature.”5 By analogy, he continued, the desk library would allow its user to forge a link between any two items that seemed to have an association (the example he used was an article on the English long bow, which would be linked to a separate article on the Turkish short bow; the actual mechanism of the link would be a symbolic code imprinted on the microfilm next to the two items). “Therea...

In any case, Bush continued, once a Memex user had created an associative trail, he or she could copy it and exchange it with others. This meant that the construction of trails would quickly become a community endeavor, which would over time produce a vast, ever-expanding, and ever more richly cross-linked web of all human knowledge. Bush never explained where this notion of associative trails had come from (if he even knew; sometimes things just pop into our heads). But there is no doubt that it ranks as the Yankee Inventor’s most profoundly original idea. Today we know it as hypertext. And that vast, hyperlinked web of knowledge is called the World Wide Web.

Hopper would later gain fame both as a teacher and as a pioneer in the development of high-level programming languages. Yet perhaps her best-known contribution came in the summer of 1945, when she and her colleagues were tracking down a glitch in the Mark II and discovered a large moth that had gotten crushed by one of the relay switches and shorted it out. She taped the dead moth into the logbook with the notation “First case of an actual bug being found.”

By rendering software completely abstract and decoupling it from the physical hardware, the stored-program concept has had the paradoxical effect of making software into something that is almost physically tangible. Software has become a medium that can be molded, sculpted, and engineered on its own terms. Indeed, as the Yale University computer scientist David Gelernter has pointed out, the modern relationship between software and hardware is essentially the same as that between music and the instrument or voice that brings it to life. A single computer can transform itself into the cockpit of a fighter jet, a budget projection, a chapter of a novel, or whatever else you want, just as a single piano can be used to play Bach or funky blues. Conversely, a spreadsheet file can (with a little effort) be run on a Microsoft Windows machine, a Macintosh, or a Unix workstation, just as a Bach fugue can be performed on a pipe organ or by an ensemble of tubas. The bits and bytes that encode the spreadsheet obviously can’t function without the computer, any more than a page full of notes can become music without a performer. And yet the spreadsheet also transcends the computer, in exactly the same way that the Bach fugue transcends any given performance of it. Everything important about that spreadsheet—its on-screen appearance, its structure, its logic, its functionality, its ability to respond to the user—exists, like the harmonies and cadences of Bach, in an abstract, platonic wo...

To do a full mathematical analysis, however, Shannon needed a way to quantify what was being transmitted through those five stages of communication. Happily, he didn’t have to look far. The term “information” had been common parlance around Bell Labs since 1928, when an engineer there named Ralph Hartley had first used it to describe the amount of message flowing through a telephone wire, as opposed to static. Building on earlier work by Bell Labs mathematician Harry Nyquist—who had called it intelligence—Hartley had defined “information” as the useful part of the signal, the part that people wanted to listen to, the part that the engineers were always trying to maximize in the face of noise and distortion. More precisely, he said, “information” ought to measure how much you actually learned from a given message. For instance, if you already knew what the message was going to say, then you learned nothing new by receiving it, and the information content should work out to be zero. But if you didn’t have a clue what the message was going to say, then you learned a lot by receiving it, so its information content ought to be high. Hartley had then gone on to supply a mathematical expression of this idea, which Shannon liked very much. Giving Hartley full credit, he accordingly took it over and made the definition rigorous—and to modern eyes very familiar. At least in the simplest cases, said Shannon, the information content of a message was just the number of binary 1s and 0s...

Suppose you’re trying to send a birthday greeting down a telegraph line, say, or through a wireless link. The communication channel has an information capacity of so many binary digits per second. And the message likewise carries an average information content of so many binary digits per letter. But taken together, Shannon realized, these two quantities determine a fundamental speed limit, measured in binary digits per second. Above that speed limit, perfect fidelity is impossible: however cleverly you encode your message and compress it, you simply cannot make it go any faster unless you first throw some information away. Below that speed limit, however, the transmission is potentially perfect. Shannon was able to show that there must exist codes that will get you right up to the limit without losing any information at all. Moreover, such perfection is possible even in the presence of noise. Shannon demonstrated that no matter how much static and distortion there may be in a given communications channel, and no matter how faint the signal, messages can still get through with perfect fidelity. Of course, if the signal is very, very faint, you may have to assign a huge number of 1s and 0s to each letter or pixel so that some of them have a chance of getting through. And you may have to devise all kinds of elaborate error-correcting codes so that corrupted parts of the message can be reconstructed at the other end. In practice, diminishing returns will set in: the codes wil...

Ultimately his fundamental theorem explains how, for example, we can casually toss around compact discs in a way that no one would have ever dared do with long-playing vinyl records: error-correcting codes inspired by Shannon’s work allow the CD player to eliminate noise due to scratches and fingerprints. Shannon’s theorem likewise explains how error-correcting computer modems can transmit data at the rate of tens of thousands of bits per second over ordinary (and relatively noisy) telephone lines. It explains how NASA scientists were able to get the Voyager spacecraft’s imagery of the planet Neptune back to Earth across two billion miles of interplanetary space. And in general, it goes a long way toward explaining why today we live in an increasingly digital world—and why the very word digital has become synonymous with the highest possible standard in data quality and reliability.

The technical distinction between information and meaning was too much a violation of common usage, he felt, and would just end up confusing people. Could von Neumann suggest anything better? Von Neumann’s answer was immediate, as Shannon later recounted the story: “You should call it entropy, and for two reasons.” First, von Neumann told the younger man, his formula for the information content of a message was mathematically identical to the physicists’ formula for entropy, a mathematical variable related to the flow of heat.* (Shannon was astounded to learn this; he had derived his formulation totally on his own.) But second, and much more important, said von Neumann, “most people don’t know what entropy really is, and if you use the word entropy in an argument, you will win every time!”

Wiener was not amused. More than once in his long career, the portly mathematician had been known to snore his way through a lecture until the speaker was just reaching his main conclusion: “I did that,” he would snap, suddenly wide awake. “I did that years ago.” And that appears to have been his position on information theory as well. He had a point, sort of. It all depended on how you interpreted “information theory.” If you meant the specific set of mathematical tools that Shannon was inventing for communication engineers, then no, Wiener had not had much, if anything, to do with it. However, if you meant the broader concept of message in all its aspects, then yes, Wiener had been immersed in the field since the 1920s. His inspiration had come from Vannevar Bush, who’d explained to him that electrical engineering actually embraced two separate disciplines: power engineering, where the goal was to optimize the flow of energy through a line, and communication engineering, where the goal was to modulate a comparatively trivial energy flow into a meaningful signal. Wiener had been enchanted with this insight and deeply influenced by it. It had led directly to some of his most important technical work in the 1920s and 1930s, work that was certainly similar in spirit to Shannon’s theory and that ultimately resulted in a definition of information flow that was mathematically identical to his.

Wiener’s assistant Oliver Selfridge, who was also down in Mexico City that semester, along with Walter Pitts, remembers Wiener’s going through his reasoning for them one evening over coffee as they sat in the lovely rooftop garden of his apartment house. Given the central role of communication in his new science, Wiener explained, his first thought had been to derive a name from the Greek word for “messenger.” Unfortunately, that word was angelos, which in English had long since taken on the specific meaning of “a messenger from God.” Somehow, a new science of angelics wasn’t quite what he was looking for. So instead, said Wiener, he had decided to focus on the theme of control. In Latin, he knew, the word for “steersman” was gubernator, from which we get the English word governor. That was better, since the word governor could sometimes refer to a device used to control the speed of an engine. More often, though, it referred to a human “governor,” or steersman for policy: the wrong connotation again. However, the Latin gubernator turned out to be a corruption of the Greek word for “steersman,” kybernetes. And that, Wiener felt, could be transmuted into English very nicely, as cybernetics. Selfridge and Pitts agreed that cybernetics was indeed an excellent name for the new science. And so, blissfully ignorant that he had just given later generations the means to coin an endless string of buzzwords—cyberspace, cybercash, cyberpunk, cybersex, ad infinitum—Wiener continued wr...

And by 1946, von Neumann, Goldstine, and Burks had laid out a more refined concept of machine architecture in “Preliminary Discussion of the Logical Design of an Electronic Computing Instrument,” a report that would prove to be even more influential than “First Draft.” Among many other things, the 1946 report pointed out how much more efficient a computer would be if it could get access to each memory address at “random”—that is, instantaneously, without having to wait until the correct address came around on a circulating tape or a mercury delay line. Naturally, such a storage scheme became known as Random-Access Memory, or RAM. (Of course, the report also concluded that the memory unit should store the data as charged spots on the face of a cathode-ray tube—a cutting-edge technology in 1946, but now so thoroughly obsolete that almost no one remembers it.)

Among its innovations were conceptual aids such as the “flow chart,” a graphical technique for keeping track of how data flow from one program step to the next. The report also included numerous examples of how complex programs could be built up from simpler blocks of code, which today would be known as subroutines.

Neumann report 1946

He didn’t call it the Information Age; that term would be invented later, by others. But he made it clear that this magical stuff called information lay at its heart. Information was a substance as old as the first living cell and as new as the latest technology. It was the stuff that flowed through communication channels; indeed, it was the stuff that messages were made of. But it was also the stuff that concepts and images and stored programs were made of. It was the stuff that entered the eyes and the ears, that flowed through the brain, that provided the feedback for purposeful action. Information was what computers and brains were about. It was the one central concept that unified communication, computation, and control and made them all seem like different facets of one underlying reality.

Lincoln Lab’s initial guess for the programming requirements on SAGE—that it would require perhaps a few thousand lines of computer code to run the entire air-defense system—was turning out to be the most laughable underestimate of the whole project. True, the Lincoln Lab team was hardly alone in that regard. Many computer engineers still regarded programming as an afterthought: what could be so hard about writing down a logical sequence of commands? Nonetheless, the Lincoln Lab programmers were being asked to create what would now be called a real-time operating system for the most complex computer/communications system in the world, and they had no modern tools to help them—no Fortran, no Cobol, no Algol; no computer languages, period. All they had was the most basic, hardware-level computerese, alphanumeric codes that corresponded to operations like “Add the contents of register A to the contents of register B and place the results in register C.” Anyone who tried to program with such codes quickly discovered that it was terribly easy to make mistakes in even the simplest algorithms. And the SAGE system was anything but simple. Indeed, it soon became all too clear that the air-defense software was going to need something like two thousand programmers, which presented a problem, to put it mildly. First, MIT had no desire to put so many people on the Lincoln Lab payroll for a single project. One day the air-defense programming would be finished, and then what would they d...

Basically, they knew, a proof in logic was just like the proofs that every new generation of high school students had to struggle through in algebra or geometry. The starting point was a set of symbolic statements known as the premises, and the goal was another symbolic statement known as the theorem. To get from one to the other, you just had to combine and modify the premises by using the rules of inference (of which there were about half a dozen). In time, if the rules were applied in the right order, the manipulations would produce the theorem. The trick was to figure out that right order. Although the proof in its final form was obviously “logical,” the process of discovering it was not at all logical. Ask mathematicians how they did it, and they would likely describe a process of befuddled groping as they went through days, weeks, even years of trial and error-followed (sometimes) by flashes of insight and intuition. In short, they would describe bounded rationality in action. Indeed, Newell and Simon realized, in a well-defined domain such as logic-theorem proving, you could even show why rationality had to be bounded, quite aside from the limits on short-term memory or anything else. Imagine for a moment that you wanted Logic Theorist to prove its theorems by brute force. The method sounds easy enough: just start from the given premises and have the program systematically apply every inference rule until the correct solution is found. And in fact, that approach is ...

McCarthy has always maintained that the idea simply popped into his head during that same IBM summer of 1955, almost as soon as he’d recognized the problem. If you wanted to create your programs interactively, he’d reasoned, and if nobody was going to give you your very own 704 to do it with, then the obvious answer was to get together with a bunch of other users to share a machine. And not just share it in the lockstep, line-forms-at-the-rear manner of batch processing, either. Really share it. Give everybody a remote terminal so they could all tap in to the big computer through telephone lines whenever they liked, from wherever they liked. Once they were in, moreover, assign each of them a securely walled-off piece of the computer’s memory where they could store data and programming code without anybody else’s horning in. And finally, when the users needed some actual processing power, dole it out to them via an artful trick. You couldn’t literally divide a computer’s central processing unit, McCarthy knew; the standard von Neumann architecture allowed for only one such unit, which could carry out only one operation at a time. However, even the slowest electronic computer was very, very fast on any human time scale. So, McCarthy wondered, why not let the CPU skip from one user’s memory area to the next user’s in sequence, executing a few steps of each task as it went? If that cycle was repeated rapidly enough, the users would never notice the gaps (think of a kindergarte...

Thanks to the proliferation of commercial machines such as the IBM 650 and 704, plus the success of Fortran, the years 1956 and 1957 had already seen new dialects emerging by the dozens (two of the languages produced during this era, Fortran itself and the business-oriented Cobol, are still in widespread use today). However, none of these efforts came close to meeting McCarthy’s standards. In all of the new languages, for example, each symbolic variable had to be assigned to a fixed block of storage in the computer’s memory, so that one block might be allocated to the variable “name,” another block to “salary,” and so on. In fact, those assignments had to be made by the programmer up front, before the program could even begin to execute (in many modern languages, such as C++, this is still the case). But in artificial-intelligence research, such a demand would be sure death. The “variables” in an AI program somehow had to represent the quicksilver fluidity of mental states in human working memory—the images, concepts, possibilities, goals, and alternatives that the problem solver focuses on at any given instant. And there was no way for the programmer to know in advance how big or how complex these variables should be, because they would constantly be changing as the problem solving advanced. Fortunately for McCarthy, however, a solution was already at hand, courtesy of Carnegie Tech’s Allen Newell. Some three years earlier, in the course of creating his Logic Theorist pro...

A Lisp programmer could happily go on indefinitely this way, hooking up simple functions into more complicated ones, then hooking up those functions to create still more complicated functions, until things got as sophisticated as anyone could want. Indeed, this functional structure ultimately turned out to be Lisp’s most powerful and compelling feature. First, it fitted in perfectly with McCarthy’s goal of interactive, question-and-answer-style computing: all you had to do was think of each Lisp function as a question, and its value as the computer’s answer. That was one big reason McCarthy and his students decided not to have Lisp compiled directly into machine code, like Fortran and most other languages. Instead, they created a kind of monitor program called the Lisp interpreter, which would scan each function as the user typed it in, convert it to machine code on the fly, compute the results, and then respond instantly. Second, Lisp’s functional structure gave the programs themselves a striking clarity. Instead of wallowing in undifferentiated masses of numeric machine-language codes (or, for that matter, undifferentiated masses of Fortran statements), McCarthy and his students could build their programs out of meaningful, well-defined building blocks—namely, Lisp functions. And that, in turn, allowed various members of the group to single-handedly tackle what were then considered very large and complex problems—carrying out the symbolic manipulations of ordinary algebr...

As a Lisp programmer continued to link simpler functions into more complex ones, he or she would eventually reach a point where the whole program was a function—which, of course, would also be just another list. So to execute that program, the programmer would simply give a command for the list to evaluate itself in the context of all the definitions that had gone before. And in a truly spectacular exercise in self-reference, it would do precisely that. In effect, such a list provided the purest possible embodiment of John von Neumann’s original conception of a stored program: it was both data and executable code, at one and the same time.**

Lick began the article with a metaphor: “The fig tree is pollinated only by the insect Blastophaga grossorum [the fig wasp]. The larva of the insect lives in the ovary of the fig tree, and there it gets its food. The tree and the insect are thus heavily interdependent: the tree cannot reproduce without the insect; the insect cannot eat without the tree; together, they constitute not only a viable but a productive and thriving partnership. This cooperative ‘living together in intimate association, or even close union, of two dissimilar organisms’ is called symbiosis. ... The purposes of this paper are to present the concept [of] and, hopefully, to foster the development of man-computer symbioses.”22

symbiosis meant humans and computers working together in a partnership, with each side doing what it did best: “[Humans] will set the goals and supply the motivations. ... They will formulate hypotheses. They will ask questions. They will think of mechanisms, procedures, and models. ... They will define criteria and serve as evaluators, judging the contributions of the equipment and guiding the general line of thought. ... The information-processing equipment, for its part, will convert hypotheses into testable models and then test the models against data. ... The equipment will answer questions. It will simulate the mechanisms and models, carry out the procedures, and display the results to the operator. It will transform data, plot graphs. ... [It] will interpolate, extrapolate, and transform. It will convert static equations or logical statements into dynamic models so that the human operator can examine their behavior. In general, it will carry out the routinizable, clerical operations that fill the intervals between decisions.”

The first and foremost challenge, said Lick, was what he called the speed mismatch: “Any present-day computer is too fast and too costly for real-time cooperative thinking with one man.” His phrasing suggests that he had already rejected as impractical the idea of everyone’s having a computer of his or her own, as Wes Clark or Ken Olsen might have advocated. “Clearly,” he wrote, “for the sake of efficiency and economy, the computer must divide its time among many users.” He was deliberately noncommittal about how this would be accomplished, noting only that time-sharing systems were currently under active development. Never one to hold his imagination in check, however, he couldn’t resist following the notion to its logical conclusion—namely, an on-line “thinking center” not unlike today’s World Wide Web. “[It] will incorporate the functions of present-day libraries together with anticipated advances in information storage and retrieval and the symbiotic functions suggested earlier in this paper. The picture readily enlarges itself into a network of such centers, connected to one another by wide-band communications lines and to individual users by leased-wire services.”

“No one knows what it would do to a creative brain to think creatively continuously. Perhaps the brain, like the heart, must devote most of its time to rest between beats. But I doubt that that is true. I hope it is not, because [interactive computers] can give us our first look at unfettered thought. It can allow a decision maker to do almost nothing but decision making, instead of processing data to get into position to make the decision.”

A model, in short, is any convenient simulation of reality. However, as Lick noted in another paper29 from the 1960s, there are models, and then there are models: “Ordinary mathematical models are static models. They are representations in symbols, usually written in pencil or ink on paper. They do not behave in any way. They do not ‘solve themselves.’ For any transformation to be made, for any solution to be achieved, information contained in the model must be read out of the static form and processed in some active processor, such as a mathematician’s brain or a computing machine. A dynamic model, on the other hand, exists in its static form only while it is inactive. The dynamic model can be set into action, by one means or another, and when it is active, it does exhibit behavior and does ‘solve itself.’” The Curtiss Robin was a dynamic model in this sense, he explained: once the fan was turned on, it flew. It was active. The same was true of the analog electronic simulations he had once experimented with at the MIT acoustics lab. And digital computer models? In principle, said Lick, they combine the best of both worlds by being both static and dynamic—and more. Indeed, he argued, it is this characteristic that gives software its unique power as a modeling medium. A model sculpted from software is static when it exists as binary code on a disk or a tape. As such, it can be stored, transmitted, archived, and retrieved, just as ordinary text can be. It in fact is a kind o...

“By far the most numerous, most sophisticated, and most important models are those that reside in men’s minds. In richness, plasticity, facility, and economy, the mental model has no peer.” Included among those mental models are images recalled from memory, expectations about the probable course of events, fantasies of what might be, perceptions of other people’s motives, unspoken assumptions about human nature, hopes, dreams, fears, paradigms—essentially all conscious thought. Of course, Lick and Taylor would continue, “[the mental model] has shortcomings. It will not stand still for careful study. It cannot be made to repeat a run. No one knows just how it works. It serves its owner’s hopes more faithfully than it serves reason. It has access only to the information stored in one man’s head. It can be observed and manipulated only by one person.” But if you could join mental models to computer models, Lick reasoned, and if you could get the two of them into just the right kind of symbiotic relationship, then you could overcome every one of those limitations.

computers are so good at processing vast quantities of data, but much more important, according to Lick, is their potential to give us a fundamentally new way of representing knowledge. In addition to our classic formats—text, tables, diagrams, equations, and the like—we now have the power to represent knowledge as a process, an executable program. Imagine the equations that describe, say, the development of a hurricane. And now imagine a computer simulation that shows us that hurricane wandering across the Caribbean to devastate southern Florida: the equations have been brought to life by the computer in precisely the same way that the score of Beethoven’s Fifth Symphony is brought to life by an orchestra. Just as in music, drama, dance, or any other performing art, Lick declared, “information is a dynamic, living thing, not properly to be confined (though we have long been forced to confine it thus) within the passive pages of a printed document. As soon as information is freed from documental bounds and allowed to take on the form of process, the complexity (as distinguished from the mere amount) of knowledge makes itself evident.”32

McCarthy’s talk had an impact, not least because at the very end of it he finally stated in public what he’d long been mulling over in private: “If computers of the kind I have advocated become the computers of the future,” he said, “then computation may someday be organized as a public utility, just as the telephone system is a public utility. We can envisage computer service companies whose subscribers are connected to them by telephone lines. Each subscriber needs to pay only for the capacity that he actually uses, but he has access to all programming languages characteristic of a very large system. The system could develop commercially in fairly interesting ways. Certain subscribers might offer services to other subscribers. One example is weather prediction[;] other possible services include ... programming services. Some subscribers might rent the use of their compilers. Other subscribers might furnish economic predictions. The computing utility could become the basis for a new and important industry.”42

IBM had promised that the first version of OS/360 would be available by the end of 1965, and that through the selective omission of various bells and whistles, it would run on System/360 processors with as little as sixteen kilobytes of memory. By summer, however, with the programmers falling further and further behind schedule every day, it was clear that the planners had grossly underestimated the magnitude of the task. Indeed, OS/360 would eventually go down in history as one of computerdom’s classic horror stories, just as project chief Fred Brooks’s rueful meditation on the lessons he’d drawn from that experience, The Mythical Man-Month,41

And by all accounts the wildest of the bunch was Utah’s Alan Kay, a guy who was so far out in the future that not even this crowd could take him seriously—yet so funny, so glib, and so irrepressible that they listened anyway. When it was his turn to give a presentation, Kay told them about his idea for a “Dynabook,” a little computer that you could carry around in one hand like a notebook. One face would be a screen that would display formatted text and graphics and that you could draw or write on as if it were a pad of paper. The Dynabook would communicate with other computers via radio and infrared. It would be so simple to program that even a kid could do it. And it would have lots and lots of storage inside—maybe even a hard disk!

Taylor also tried to keep the proceedings as easy and as informal as possible, to the point of having the conference room furnished with beanbag chairs. He even let the speakers set the rules for how each meeting would proceed, much as a card dealer could call the game in Las Vegas; thus their nickname, Dealer Meetings. And when the arguments got heated, which they often did, the minister’s son would do his best to convert a “class-one” disagreement—one in which the combatants were simply yelling at each other—into a “class two” disagreement, in which each side could explain the other side’s position to the other side’s satisfaction. You don’t have to believe the other guy, he would tell them. You just have to give a fair account of what he’s saying. And it worked. As one CSL member later explained it, Taylor’s class one/class two exercise was amazingly effective at clarifying unspoken assumptions and ferreting out facts that one person knew and another didn’t. “So by the time you get done,” he said, “you all know the same set of things, and you end up concluding the same thing.”4

In fact, says Lampson, he and his colleagues were increasingly coming to realize that the decision to focus on graphics displays undermined the most fundamental premise of time-sharing—namely, that computers are fast and humans are slow. As he would express it in a later account, “[This] relationship holds only when the people are required to play on the machine’s terms, seeing information presented slowly and inconveniently, with only the clumsiest control over its form or content. When the machine is required to play the game on the human’s terms, presenting a pageful of attractively (or even legibly) formatted text, graphs, or pictures in the fraction of a second in which the human can pick out a significant pattern, it is the other way around: People are fast, and machines are slow.”15