It Began with Babbage (26 page)

However, the ENIAC schema was
not
a subschema of the stored-program computer schema. The differences were too fundamental. The ENIAC and the stored-program computer schemas, in the minds of the people involved in their development, stood apart. And the vital point is that the ENIAC schema
never became a paradigm
, whereas the stored-program computer schema did.

Paradigms in the natural sciences differ in a fundamental way from paradigms of the kind that concerns us in this story. Consider, for example, the Darwinian paradigm in biology. Its vital element is the theory of natural selection, and this is a statement about the organic world as it is. Philosophers would call it a
descriptive
theory. The Darwinian paradigm is, accordingly, a descriptive paradigm, concerning the organic world as it is.

In contrast, what we are calling the
stored-program computer paradigm
is about an artificial world; the computer is an artifact, an invented thing. Any possible science that might be represented by this paradigm is a science of the artificial (see Prologue).
³⁸
And any paradigm characterizing an artificial science, dealing with an artifact (or a family of artifacts) is necessarily
a prescriptive
paradigm. It speaks of a world as it ought to be (see Prologue, Section II).

Natural scientists may agree or disagree about how the world really is; artificial scientists may agree or disagree about how the world ought to be. This distinction between
is
and
ought
is, equivalently, the distinction between what is
true
and what is
right
. It is the distinction between empirical statements that are true or false, and value judgments that distinguish between right or wrong with respect to some purpose or goal. The paradigms that govern the artificial sciences, being prescriptive, express value judgments; the paradigms that govern the natural sciences, being descriptive, express statements about facts. Thus, one can argue whether the special theory of relativity is true—that is, whether it is a correct description of some aspect of the physical world. One cannot argue whether the ENIAC is true—only whether it is achieves its stated purpose, or how well it achieves its stated purpose, how good or bad it is with respect to its intended purpose.

Let us keep this in mind for the purpose of this unfolding story. Let us always keep in mind that any science we associate with computers (which are artifacts), is a science of the artificial. A paradigm for an artificial science is, then, unconcerned fundamentally with truth or falsehood. It is concerned fundamentally with goodness or badness, rightness or wrongness—not in an ethical or moral sense, but relative to goals and purposes.

The EDVAC report, although intended for a few eyes only, was circulated widely. Its status for paradigmhood was a matter of social acceptance by those who were either working on, or were seriously interested in, computers.

One of those who read the EDVAC report was Alan Turing, who, at the end of World War II, joined (in June 1945) the newly formed mathematics division of Britain's National Physical Laboratory (NPL) in Teddington, outside London.
³⁹
Turing's boss at NPL, the man who had recruited him, mathematician John R. Womersley (1907–1958), head of the mathematics division, had shown Turing the EDVAC report.
⁴⁰
On joining NPL, Turing began working immediately on the preliminary design of a stored-program electronic computer.

The critical element in the EDVAC report—the critical factor in the viability of a stored-program computer—was the means to achieve an economical yet fast, large internal memory. Turing's own “universal machine,” the product of his 1936
Entscheidungsproblem
paper postulated a linear tape that would move left and right across a read/write head. In a lecture delivered to the London Mathematical Society in 1947 on his new project, he alluded to this abstract machine, but he pointed out that, although in the kind of computing machine he had envisioned it was necessary for the memory to be infinite, an infinite tape would be impractical in a “real” stored-program machine because of the time that would be spent in searching the tape for information needed at any given point of time.
⁴¹

In fact, there
was
a physical artifact corresponding to Turing's abstract idea. Prior to joining NPL, Turing had been sent to Germany as a member of an Anglo-American team to determine the status of German wartime communication technology.
⁴²
One of the captured German army devices was the magnetophone, an early form of the tape recorder.
⁴³
However, Turing rejected this mechanism for the reason mentioned earlier; it required too much back-and-forth movement.
⁴⁴
Instead, he favored the acoustic delay line as a memory device.

Such a delay line had been built by Presper Eckert and others at the Moore School in summer 1943, and was studied for application to radar.
⁴⁵
When plans for the EDVAC were being drawn up, Eckert and another member of the ENIAC team worked on a preliminary design of an acoustic delay memory using mercury.
⁴⁶
The credit for the idea lay entirely with Eckert, according to Goldstine.
⁴⁷

In the EDVAC report, von Neumann made no reference to any specific acoustic memory, Eckert's or otherwise. Rather, he discussed at great length, in an abstract sort of way, a “delay memory” and its characteristics.
⁴⁸
He also described an alternative (and for him, preferred) form of memory “along the lines of the
iconoscope
,”
⁴⁹
an electrostatic storage tube that stored electrical charges and that, he believed, held distinct advantages over the delay line as memory.

The basic principle of the acoustic delay line memory is that because sound takes a certain amount of time to travel through a physical medium such as a liquid like mercury, for
that duration of time, the medium may be said to be storing the sound wave. Consider a tube of mercury of some length. An electrical signal could be transformed into an acoustic disturbance, propagate through the mercury in the tube from one end, and be transformed back into an electrical signal at the other end. If this “output” electrical signal was fed back to the “input” to the tube, then one would have a memory device. Sound travels through mercury at the rate of 1450 meters/second. A tube of mercury 1.45 meters long will give a delay of 1 millisecond.

Translation of electrical pulses into an acoustic wave and the reverse could be done using a phenomenon called the piezoelectric effect, discovered in 1880 by French scientists Pierre Curie (1859–1906) and his brother Jacques (1856–1941). In this instance, an electrical disturbance imposed on a quartz crystal would cause the crystal to oscillate. When such a crystal was attached to the input side of a tube of mercury, the oscillation produced an acoustic pulse, which propagated through the mercury, taking a certain amount of time. After a delay, it caused another quartz crystal to vibrate, producing, in turn, an electrical pulse. The latter is returned, through appropriate circuits that reshape and amplify the pulse, to the input side of the tube.
⁵⁰

Suppose now a binary 1 is represented by a pulse 0.5 microsecond wide and a 0 is noted by the absence of a pulse. A pulse position (corresponding to a 1 or a 0 followed by a gap before the next pulse [1] or nonpulse [0]) would be 1 microsecond. Then, a single delay tube 1.45 meters long would hold a sequence of 1000 binary digits. To reshape and amplify the output pulse before recirculating it would need about 10 vacuum tubes. Thus, 1000 binary digits could be stored on a mercury tube at the cost of 10 vacuum tubes. If, on the other hand, vacuum tubes were used for memory (as in the ENIAC), then 1000 binary digits would need 1000 vacuum tubes. In terms of vacuum tube use, the cost of an acoustic memory reduces by a factor of 100 to 1.

Acknowledging that Eckert was the originator of the idea of using a delay line as memory, Turing chose this mechanism over alternatives because it was “already a going concern.”
⁵¹

Turing named the machine he began to design the ACE (Automatic Computing Engine). In 1945, he wrote
Proposal for Development in the Mathematics Division of an Automatic Computing Engine
(
ACE
), which was perhaps as extensive in scope and length as the EDVAC report.
⁵²
Although believing his own paper to be a completely comprehensive account of the design of his proposed machine, he added that his paper should be read alongside the EDVAC report.
⁵³
Thus, von Neumann's paper appeared to have played a part in shaping Turing's ideas about a practical stored-program computer—a curious and pleasing
quid pro quo
if, in fact, von Neumann's idea had been shaped by Turing's 1936 work.

It was an ambitious plan. The ACE would use a delay line memory of some 6400 words of 32 binary digits each, held in 200 long mercury delay lines.
⁵⁴
Unfortunately, the ACE project had more than its share of problems, mainly in communications among its designers in NPL's mathematics division and the implementers who were in the electronics section. Eventually, the problems were resolved and, in 1949, detailed design began on a much-stripped down version of the ACE, called
Pilot ACE
. Its memory capacity was much smaller than what Turing had originally planned, comprising 300 words of 32 binary digits each.
⁵⁵

Public demonstrations of the Pilot ACE were given in May 1950,
⁵⁶
but by then at least two other stored-program computers were operational (as we see later) both, as it happened, in England. The ACE project is significant to this story not because of priority-of-invention reasons; rather, it represents an almost instant recognition by Turing (and Womersley) at NPL of the significance of the stored-program concept. It was an important piece of evidence pointing to the acceptance of the stored-program computer as a computational paradigm.

Well before the public demonstrations of the Pilot ACE, however, Turing had left NPL. In September 1947, he took sabbatical leave from NPL and returned to Cambridge to resume his war-interrupted fellowship at King's College.
⁵⁷
A year later, he resigned from NPL and moved to Manchester to take up a position created for him and funded by the Royal Society at the University of Manchester. His new position was deputy director of the Royal Society Computing Laboratory.
⁵⁸
The University of Manchester harbored computational ambitions.

If a paradigm, when born, elicits prompt recognition and acceptance by members of the relevant scientific community, if it initiates research programs (“normal science” in Kuhn's language), if it is quickly assimilated and internalized as a mental schema by individuals, then, like Turing, but more
effectively
, the work by Englishman (later Sir) Maurice Wilkes (1913–2010) and his colleagues in Cambridge University between 1946 and 1949 epitomized these phenomena. Wilkes recognized the significance of the ideas put forth in the EDVAC report as soon as he first encountered them in May 1946. In his memoirs, written in 1985, he records how, on reading the report, he immediately recognized that this was “the real thing,” that this was the way the development of the computer would follow.
⁵⁹

It would be a travesty to call the work that Wilkes and his collaborators did in those first 3 years of peacetime as “normal science.” There was nothing remotely normal about it. Wilkes was on
terra incognita
, a terrain unknown not just to him, but also to everyone else. What Wilkes did was design and implement successfully and completely a computer called the Electronic Delay Storage Automatic Calculator (EDSAC) that not only
adhered to the general principles of the stored-program computer, but also demonstrated
empirically
the viability of the concept. It was, in the realm of the artificial sciences, what in the natural sciences is called
experimentum crucis
, a crucial experiment that could make or break a theory or hypothesis. It was not the only experiment under way at that time to test the viability of the stored-program concept, but it was the first such
complete
experiment. In fact, it is fair to say that beginning with the EDSAC project, Wilkes's scientific life entailed one long series of experiments that revealed how an artificial science (in this case, computer science) can be an empirical science.

The EDSAC project was significant in another sense. It opened up an entirely new region of the stored-program computer paradigm, a new class of artifacts the likes of which had never been seen before: computer programs or software.

Wilkes obtained his BA degree in mathematics from Cambridge University (”mathematics tripos” in Cambridge language) in 1934, the same year as Turing. While Turing was a student (and later, fellow) of King's College, Wilkes studied at (and would later become a fellow of) St. John's College a few hundred yards away, but apparently they did not meet until much later.
⁶⁰

Wilkes was very much a “Cambridge man” who, on entering the university in 1931 as an undergraduate, spent his entire academic life in Cambridge until his retirement in 1980. When he died in Cambridge, he was emeritus professor of computer technology and a much-honored man, including the award of a knighthood.
⁶¹

Wilkes's forte during his early years was in applied mathematics, especially as applied to radio physics. He did his doctoral research at the Cavendish Laboratory (where the fabled Lord Rutherford was still the Cavendish professor and director), and obtained his PhD in 1938.
⁶²

By then, however, he had already encountered the lure of automatic computing. While working at the Cavendish, he came to know of the differential analyzer from a lecture in 1936 given by Douglas Hartree (1897–1958), a mathematician and numeric analyst and professor of applied mathematics at the University of Manchester. Hartree had built a differential analyzer for his department. The lecture was followed by a demonstration of a model version of the analyzer (built from components of what, in England, is called a meccano set and in America, an erector set), and was housed in the physical chemistry department of the university.