In response to:

How the Computers Exploded from the June 7, 2012 issue

To the Editors:

In Jim Holt’s review of George Dyson’s Turing’s Cathedral he falls into the trap that Dyson set for him [“How the Computers Exploded,” NYR, June 7]. Dyson has amplified the importance of John von Neumann’s MANIAC project to a point where Holt got the impression that it was the first useful computer and it started a revolution. It wasn’t. It didn’t.

John von Neumann’s MANIAC was not the first computer. Nor was it, as Holt dubs it, the first “genuine” computer, or the first high-speed, stored-program, all-purpose computer. John von Neumann did not invent the stored-program architecture that often bears his name. By the time the MANIAC came online, several stored-program machines were operating and actually for sale in England and the US. Any number of computer history texts will bear this out.

Dyson does include these facts in his book. Yet they are mostly brushed over, in a mad love affair with all things von Neumann. So it’s not surprising that Holt’s review gets basic facts wrong.

If Holt had only doubted this revisionist history enough to check another source, he could have found that the ENIAC was very busy cranking through a variety of different computational problems from 1945 to 1955 (including one for the H-bomb). By 1948 it had a stored program. In 1949 the Manchester Baby and Mark I, the EDSAC and the BINAC were running. Eckert and Mauchly had contracts in government and industry to deliver UNIVACs. The world was lousy with computers!

The idea that von Neumann was some kind of torch carrier who convinced the world that computers were important just does not wash with the facts. It does, apparently, sell books. The insiders were convinced in 1946, when the ENIAC was revealed and the description of Eckert and Mauchly’s EDVAC was disseminated under von Neumann’s name. The population was convinced in 1952, when UNIVAC predicted the election on national TV. The fact that von Neumann continues to get credit for Eckert and Mauchly’s work is maddening.

Bill Mauchly
Berwyn, Pennsylvania

To the Editors:

Jim Holt begins his very interesting review of George Dyson’s book Turing’s Cathedral with the statement that the “digital universe” came into existence in 1950 with the construction of the MANIAC—an all-purpose computer built in Princeton. I believe that this statement is incorrect. The first such computer was the ENIAC, which was built at the University of Pennsylvania in 1946. It was conceived and designed by the engineers John Mauchly and J. Presper Eckert. It had all the features of what was later called the von Neumann architecture. Von Neumann got into this around 1944 as a consultant and reformulated the engineering into a mathematical structure. Among the features of the ENIAC that had been conceived by Mauchly and Eckert was what is known as “stored programming.” Hints of this can be found in previous work by Alan Turing. Programs are numerically stored in the computer’s memory where they can be manipulated like numbers so the computer can modify its program as it goes along.

Von Neumann was such a compelling figure that it is tempting to build the story around him, which by the way Dyson does not do. Holt makes much of the fact that the MANIAC was used to compute features of the hydrogen bomb. But this was first done by the ENIAC. Von Neumann had an abolute paranoia about the Russians and favored a first nuclear strike. Einstein referred to him as a Denktier, a think animal.

The history of the electronic computer is very complex and parts of it are still disputed, but Holt’s account is too simple.

Jeremy Bernstein
New York City

To the Editors:

Arguing about precedence of near- simultaneous inventions is generally fruitless, but it is strange that Alan Turing’s Pilot ACE (Automatic Computing Engine), which has at least some claim to be the first high-speed, stored-program, von Neumann-architecture digital computer, is often written out of the history: Jim Holt in “How the Computers Exploded,” which emphasizes Turing’s role, identifies MANIAC as the first practical realization of Turing’s “universal computer,” but makes no mention of the Pilot ACE.

Pilot ACE, which was based on Turing’s original design for a full ACE version, was built at the UK National Physical Laboratory (NPL) starting in 1946, and ran its first program in May 1950. Turing left NPL before its completion, frustrated by slow progress, but the work was completed by a team including Donald Davies (later director of what came to be called the Informatics division of the NPL), Jim Wilkinson, and Michael Woodger.

Pilot ACE used 1,450 vacuum tubes, had a mercury delay line main memory of 384 32-bit words (equivalent to 1,526 bytes in modern terms) and a clock rate of 1 MHz. A 4,096-word “hard drive” drum memory was added later. Between 1950 and 1954, when it was retired to the London Science Museum, Pilot ACE saw significant practical use, most notably in the analysis of the airframe structural stresses that caused the catastrophic breakup of some early Comet airliners (the first commercial jet).

Advertisement

Incidentally, Mike Woodger, Turing’s assistant from 1946, whom I worked with at the NPL in the early 1970s, never accepted that Turing’s death by poisoning was suicide. He claimed that Turing’s careless work habits with photographic chemicals made accident much more likely, and that suicide was not consistent with Turing’s personality whatever the stressful circumstances.

Mark Dowson
Leesburg, Virginia

To the Editors:

In 2001 I heard a talk by Kay Mauchly, John Mauchly’s widow, that detailed how the “von Neumann architecture” of the stored-program binary computer got that name. I don’t think she used the word “stolen,” but it was clear that after a certain meeting with Eckert and Mauchly, John von Neumann went off to Princeton to write up their ideas as his own in his paper on the EDVAC computer. In the 1970s I heard similar opinions in chats with both Presper Eckert and Grace Hopper. As for the claim that von Neumann’s 1952 MANIAC was the first digital computer—if not EDVAC of 1945 then there was the Eckert/Mauchly BINAC of 1949, the British LEO machine of 1951, and first commercial production machine, Eckert/Mauchly’s UNIVAC 1 in 1952; and even Konrad Zuse’s Z-series machines of the 1930s and 1940s deserve a vote. Early computing history will always have many fathers, but von Neumann’s worthy reputation in mathematics does not extend to what should rightly be termed the “Eckert/Mauchly architecture.”

As for Turing, given room in Trafalgar Square for another column, instead of an apology he might well be placed atop it.

David K. Adams
former UNIVAC employee
Castine, Maine

Jim Holt replies:

I began my review by claiming that John von Neumann’s MANIAC, which became operational in 1950 at the Institute for Advanced Study in Princeton, was the first “genuine” computer—that is, the first stored-program universal Turing machine. A number of readers wrote to champion the claims of other machines to this title. Both Bill Mauchly and Jeremy Bernstein back the ENIAC, which was codesigned by Mauchly’s father. But the ENIAC, as I pointed out in the review, was not built as a stored-program machine; it had to be laboriously rewired by hand for each task. Mauchly says of the ENIAC that “by 1948 it had a stored program.” That is true, but only because the ENIAC had been retrofitted, at von Neumann’s suggestion, so that it could be programmed in a primitive sort of way. In any event, as the distinguished logician Martin Davis observed in his 2000 book, The Universal Computer, the gap between the thinking behind the ENIAC and the ideal of a universal computer was “immense.”

Mark Dowson, by contrast, nominates as the first computer the Pilot ACE—built in the late 1940s in Britain and partly based on the elegantly minimalist design of Alan Turing himself. That is more plausible, even though the Pilot ACE proved far less influential than von Neumann’s Princeton machine as a template for later computers.

Still other readers maintained that the EDSAC, designed by Sir Maurice Wilkes at Cambridge University, was the first working stored-program computer. As George Dyson notes in Turing’s Cathedral, Wilkes managed to coax his machine into operation in 1949, beating von Neumann’s by a year or so. (Its first programmed job was to print out a list of prime numbers.) Turing, though, was unimpressed by Wilkes’s EDSAC design, commenting that it was “much more in the American tradition of solving one’s difficulties by means of much equipment rather than by thought.”

Whether MANIAC or ENIAC or EDSAC or EDVAC or ACE deserves the laurel as the first genuine computer—not to mention UNIVAC, BIVAC, MARK I, LEO, Zuse Z4, and much of the rest of the alphabet—two things are certain. One is that the “Von Neumann architecture” for the modern computer is something of a misnomer, since the ideas behind it were not exclusively or even primarily due to von Neumann. The other is that “Turing machine” is not a misnomer at all, since it was Turing who came up with the original idea of an all-purpose computer. For that, and for his quietly heroic (and shabbily rewarded) role in helping Britain avert defeat in World War II, Turing probably does, as David K. Adams suggests, merit a column in Trafalgar Square.