[SIGCIS-Members] Firsts, the IAS computer, and Turing.

So collecting my thoughts on the WP article and the discussion in this thread... The striking meta level point is maybe that people continue to care about this stuff. The scholarly history of computing community largely moved on from the 1940s some time ago, but with Isaacson, Dyson, Smiley etc. the histories that reach a broad audience are still obsessed with the earliest computers, invention, and the question of "the first computer." That, in turn, reflects to a large extent the legacy of the ENIAC patent. Another observation is that the various institutions and communities involved do not have much interest in the subtleties of the unofficial "truce" whereby the first generation of history of computing scholars agreed, circa 1980, which string of adjectives went with which “first computer.” Thus we saw repeated attempts by the UK National Museum of Computing to call EDSAC the "first practical electronic computer." Now Princeton CS chair Appel, perhaps confused by (George) Dyson's book, is quoted as saying that "Turing invented computer science and the idea of the computer, and John von Neumann built the first stored-program computer." The usual disclaimer of course applies: people did not necessarily say exactly what journalists quote them as saying. I'll write a little below about the specifics of the two claims. The bigger point, however, is that our attempts to insert various kinds of nuance into the demarcation of firsts simply haven't proved compelling to the broader audiences looking to be entertained or to celebrate the accomplishments of particular people and institutions. That may have something to do with the inherently pedantic nature of inserting limiting adjectives between "first" and "computer" but it also has a lot to do with lack of respect for historical expertise. (See http://www.wsj.com/articles/book-review-to-explain-the-world-by-steven-weinb erg-1423863226). In the world of computing, getting a technical claim wrong carries much greater consequences than getting a historical claim wrong. Indeed, not knowing what one is talking about with respect to the history of computing doesn’t seem to be a problem at all. Specifics re von Neumann and the IAS Computer. Dyson's book, Turing's Cathedral, does put a huge emphasis on the materiality of the IAS computer as a watershed moment in human history. This idea was widely propagated by reviewers who didn’t know much about the topic. For example, the NY Review of Books review essay on the book, “How Computers Exploded,” (http://www.nybooks.com/articles/archives/2012/jun/07/how-computers-exploded /) began The digital universe came into existence, physically speaking, late in 1950, in Princeton, New Jersey, at the end of Olden Lane. That was when and where the first genuine computer—a high-speed, stored-program, all-purpose digital-reckoning device—stirred into action. It had been wired together, largely out of military surplus components, in a one-story cement-block building that the Institute for Advanced Study had constructed for the purpose. Here is how I explained the issue in my essay on Dyson’s book (http://www.siam.org/news/news.php?id=2054) His insistence on the IAS computer and its "fully electronic random access storage matrix" as "as close to a point source" for the origin of the "digital universe" as "any approximation can get" reflects an urge to explain a particular episode as the singular origin of something vast. Pinpointing beginnings is a primal driver of storytelling---consider the Book of Genesis. But historians have spent decades trying to move beyond partisan advocacy for one or another great man as the true inventor of the computer. Looking for a point source leads to history as viewed through a fisheye lens. When discussing the influence of von Neumann on computing, historians traditionally focus on the 1945 "First Draft Report on the EDVAC" circulated under his name. His personal responsibility for many of the ideas set forth in the document has frequently been disputed, but its huge influence on the computer projects initiated over the next few years has not. Historians who have looked more closely at the era also credit an early description of the planned design for the IAS computer, circulated in 1946, and its early revisions as an important influence on many of these projects. The physical, functional computer was much less influential, in part because engineering delays resulted in its completion only after at least one of the machines modeled on its detailed design was already operational. One of Dyson's idiosyncrasies is to write as if these three achievements could not be separated, commenting relatively little on the 1945 "First Draft." He places the full burden of universe-changing historical importance on the physical IAS computer, which ran its first program in 1951, rather than on ideas that many others had already embraced, and indeed extended, years earlier. Dyson boosts the historical importance of the IAS computer by omitting or downplaying information on developments elsewhere before or during 1951, with the exception of a tiny 1948 prototype computer at the University of Manchester from which von Neumann's team took the memory technology. Dyson's evidence is truthful, but startlingly incomplete. For example, while he concedes in his introduction that "the IAS machine was not the first computer," he never mentions EDSAC, operational at the University of Cambridge in 1949, which historians have almost universally recognized as the first useful computer built on the model described by von Neumann in 1945. It was also in 1949 that the Manchester team got its memory technology working in a full-scale computer. In 1951 UNIVAC provided the first commercially manufactured computer to the U.S. Census Bureau, while in the U.K. J. Lyons and Company, best known for its chain of teashops, completed its own computer and applied it to business automation. Other computers were already operational in the Soviet Union and Australia. All those milestones pass unmentioned by Dyson. Dyson asks in conclusion, "How did the von Neumann vector manage to outdistance all the other groups trying to build a practical implementation of Turing's Universal Machine in 1946?" Even an attentive reader might assume that this "outdistancing" involved winning a race rather than losing it by several years. Dyson's implication that the various teams of computer builders inspired by von Neumann's proposed design all saw themselves as trying to implement the computational model described by Turing in his now-celebrated 1936 paper "On Computable Numbers, with an Application to the Entscheidungsproblem" is also likely to raise howls of protest from historians who have looked at early computing. Specifics re Turing The idea that Turing’s 1936 paper somehow provided a crucial “stored program concept” which von Neumann appropriated for the 1945 “First Draft of a Report on the EDVAC” is being actively promoted and surfaces with increasing frequency. I personally do not find it at all convincing. Here’s what I said about it in “Actually, Turing Did Not Invent the Computer” in CACM last year. (http://www.tomandmaria.com/tom/Writing/CACMActuallyTuringDidNotInventTheCom puter.pdf) Where one might leap into fantasy is by asserting the cluster of ideas contained in von Neumann's 1945 "First Draft" are merely a restatement, or at most an elaboration, of Turing's earlier work on computability. Judge for yourself, by placing side by side Turing's 1936 "On Computable Numbers..." and "First Draft of a Report on the EDVAC." They are easy to find with Google, though you might want to pour yourself a fortifying beverage first as neither is particularly easy reading. The former is a paper on mathematical logic. It describes a thought experiment, like Schrödinger's famous 1935 description of a trapped cat shifting between life and death in response to the behavior of a single atom. Schrödinger was not trying to advance the state of the art of feline euthanasia. Neither was Turing proposing the construction of a new kind of calculating machine. As the title of his paper suggested, Turing designed his ingenious imaginary machines to address a question about the fundamental limits of mathematical proof. They were structured for simplicity, and had little in common with the approaches taken by people designing actual machines. Von Neumann's report said nothing explicitly about mathematical logic. It described the architecture of an actual planned computer and the technologies by which it could be realized, and was written to guide the team that had already won a contract to develop the EDVAC. Von Neumann does abstract away from details of the hardware, both to focus instead on what we would now call "architecture" and because the computer projects under way at the Moore School were still classified in 1945. His letters from that period are full of discussion of engineering details, such as sketches of particular vacuum tube models and their performance characteristics. The phrase "stored program concept" has sometimes been used to encapsulate the content of the "First Draft" report, but this underplays its actual impact by implying it held just one big idea. In fact it provided a wealth of intertwined ideas and details. In my current work with Mark Priestley and Crispin Rope I have found it useful to separate these into three main areas. The first, the "EDVAC Hardware Paradigm" described an all-electronic binary computer with a much larger memory than anything ever built previously. The second, the "von Neumann Architecture Paradigm," set out the basic structure of the modern computer: special-purpose registers on which all operations were performed and from which data was exchanged with main memory, separation of arithmetic functions from control functions from memory units, only one action performed at a time, and so on. The third, the "Modern Code Paradigm," concerns the nature and capabilities of its instructions. For example, instructions were expressed as through a small vocabulary of operation codes followed by argument or address fields. These were held in the same numbered memory cells as data. While executed by default in a particular sequence, the machine could jump out of sequence and the destination of this jump could be modified as the program ran based on the state of the computation. Taken together, von Neumann's cluster of ideas guided the construction of computers that were much cheaper, smaller, more reliable, and more flexible than their predecessors. ENIAC, the first general-purpose electronic digital computer, used almost 18,000 vacuum tubes. The more tubes a machine held the more expensive it was to build and, as they eventually burn out, the less reliable. Its immediate successors held 1,000 or 2,000 tubes yet could handle problems of greater logical complexity and were easier to program. This efficiency made possible the construction of computers in cash-strapped Britain following the war, and made computers affordable and useful enough that they were rapidly turned into commercial products and applied to business tasks as well as scientific computations. According to Copeland, "the fundamental conception of the stored-program universal computer" was Turing's. Von Neumann merely "wrote the first paper explaining how to convert Turing's ideas into electronic form." But what actually would have been different about von Neumann's "First Draft" report if Turing had never written his now famous paper? My answer to that question is: nothing (with the possible exception of the neuron notation he appropriated to describe logic gates, whose creators cited Turing). Copeland has gone so far as to argue the basic idea of a single machine that could do different jobs when fed different instructions can be traced to Turing. But Charles Babbage had that idea long before, and as mentioned earlier, several computers controlled by sequential instruction tapes had already been constructed with no influence from Turing and were well known to von Neumann before he wrote his report. EDVAC went far beyond this to store a program in addressable internal memory rather than on a sequential instruction tape. To suggest this advance came from Turing is odd, as the machine Turing described had no internal writable memory and took its instructions from a tape. Von Neumann brought a concern with logic and preference for minimal, general-purpose mechanisms to the design of EDVAC but he did not need Turing to teach him that. He was a mathematic genius with a deep pragmatic streak and an astonishing track record of productive collaborations across a huge range of fields. Turing's 1936 paper lacks many novel and fundamental features found in the "First Draft" such as addressable memory locations. Neither did Turing describe instruction codes followed by arguments, the building blocks of computer programs. The suggestion that the EDVAC design was merely a conversion of Turing's paper implies these features are trivial, and the single important idea in each document is that code and data should be treated interchangeably so programs can modify themselves. Yet while Turing's paper showed one machine could, in modern terms, emulate the functioning of another it never described a machine altering its own instructions. Furthermore, at the very end of the "First Draft" von Neumann expressly forbade EDVAC from overwriting the operation fields in its instructions, even though he relied on modifications to their address fields to accomplish basic operations such as conditional branching. This address modification was a very influential idea in the "First Draft," but was, of course, absent from Turing's paper as his machines did not use addresses. In other words, the capability for unrestricted self-modifying code von Neumann is said to have copied from Turing is something Turing did not describe and von Neumann's design explicitly prohibited. So my challenge for anyone who wants to argue that Turing’s 1936 paper was a prerequisite for the invention of modern computer architecture is to find specific and essential features of the modern computer which 1) Are present in the 1945 “First Draft of a Report on the EDVAC.” 2) Are present in Turing’s 1936 paper. 3) Are not present in the designs of the Harvard Mark I, ENIAC, or the Bell Labs relay computers, all were entirely familiar to von Neumann by early 1945 and were designed in complete ignorance of Turing’s paper. 4) Are not present in the work of Babbage (though I cannot remember seeing any discussion of whether von Neumann was familiar with the specifics of Babbage’s ideas for the analytical engine, and it seems quite likely that he wasn’t). 5) Are not generic features of mathematical logic with which von Neumann would in any event have been familiar with. Simply waving one’s hands vigorously while invoking a “stored program concept” will not do. How about the claim that Turing founded computer science? That’s not true in any literal way either. As I addressed that idea in the CACM article: Turing provided a crucial part of the foundation of theoretical computer science. There was no such thing as computer science during the early 1950s. That is to say there were no departments of computer science, no journals, no textbooks, and no community of self-identified computer scientists. An increasing number of university faculty and staff were building their careers around computers, whether in teams creating one-off computers or in campus computer centers serving users from different scientific disciplines. However, these people had backgrounds and appointments in disciplines such as electrical engineering, mathematics, and physics. When they published articles, supervised dissertations, or sought grants they had to be fit within the priorities and cultures of established disciplines. The study of computing always had to be justified as a means, not as an end in itself. Ambitious computer specialists were not all willing to make that compromise and sought to build a new discipline. It was eventually called computer science in the U.S., though other names were proposed and sometimes adopted. To win respectability in elite research universities the new discipline needed its own body of theory. The minutiae of electronic hardware remained the province of engineering. Applied mathematics and numerical analysis were tied too closely to the computer center tradition of service work in support of physicists and engineers. Thus, the new field needed a body of rigorous theory unique to computation and abstracted from engineering and applied mathematics. Turing was not, in any literal sense, one of the builders of the new discipline. He was not involved with ACM or other early professional groups, did not found or edit any journal, and did not direct the dissertations of a large cohort of future computer scientists. He never built up a laboratory, set up a degree program, or won a major grant to develop research in the area. His name does not appear as the organizer of any of the early symposia for computing researchers, and by the time of his death his interests had already drifted away from the central concerns of the nascent discipline. When building a house the foundation goes in first. The foundations of a new discipline are constructed rather later in the process. Turing's 1936 paper was excavated by others from the tradition of mathematical logic in which it was originally embedded and moved underneath the developing new field. In several papers historian Michael S. Mahoney sketched the process by which this body of theory was assembled, using pieces scavenged from formerly separate mathematical and scientific traditions. The creators of computer science drew on earlier work from mathematical logic, formal language theory, coding theory, electrical engineering, and various other fields. Techniques and results from different scientific fields, many of which had formerly been of purely intellectual interest, were now reinterpreted within the emerging framework of computer science. Historians who have looked at Turing's influence on the development of computer science have shown the relevance of his work to actual computers was not widely understood in the 1940s. Turing's 1936 paper was one of the most important fragments assembled during the 1950s to build this new intellectual mosaic. While Turing himself did see the conceptual connection he did not make a concerted push to popularize this theoretical model to those interested in computers. However, the usefulness of his work as a model of computation was, by the end of the 1950s, widely appreciated within large parts of the emerging computer science community. Edgar Daylight has suggested that Turing's rise in prominence owed much to the embrace of his work by a small group of theorists, including Saul Gorn, John W. Carr, and Alan J. Perlis, who shared a particular interest in the theory of programming languages. His intellectual prominence has been increasing ever since, a status both reflected in and reinforced by ACM's 1965 decision to name its premier award after him. By the way, “The Imitation Game” (which I have so far avoided watching) apparently brings whole new levels of nonsense to the discussion. This time the NYRB has taken a rather more critical stance, branding the film “monstrous hogwash”: http://www.nybooks.com/blogs/nyrblog/2014/dec/19/poor-imitation-alan-turing/ . Best wishes, Tom -----Original Message----- From: members-bounces@sigcis.org [mailto:members-bounces@sigcis.org] On Behalf Of Dag Spicer Sent: Wednesday, February 25, 2015 9:47 AM To: PeterEckstein@comcast.net; John Impagliazzo; Computer, SIG Subject: Re: [SIGCIS-Members] Washington Post article about Turing BTW, Merton’s original Mathew Effect paper is available here: <http://www.garfield.library.upenn.edu/merton/matthew1.pdf> http://www.garfield.library.upenn.edu/merton/matthew1.pdf Dag On Feb 24, 2015, at 5:44 PM, <mailto:PeterEckstein@comcast.net%3cmailto:PeterEckstein@comcast.net> PeterEckstein@comcast.net<mailto:PeterEckstein@comcast.net> wrote: I consider this interview and the article to be nothing more than PR puffery on the part of Princeton. Isaacson is quoted as saying that Turing did not invent the automatic electronic digital computer, but somehow the article goes on to give as much glory as possible to Princeton but never mention the names of those Isaacson said were the real inventors--John Mauchly and Presper Eckert. They built ENIAC before there was any clear technology in which to store a program, and, after the ENIAC design was frozen, they and several others at Penn set out to develop the architecture of a next-generation stored-program computer that they named EDVAC. At some point von Neumann, who had never seen a computer before, joined the group and contributed to its thinking, and he volunteered to write up their conclusions, which were issued in a typed (and widely distributed) paper that bore his name alone. Von Neumann was, as I believe the article calls him, a protean genius, but attributing the stored-program concept to him is, indeed, an example of the Matthew Effect gone wild. ________________________________ From: "John Impagliazzo" < <mailto:John.Impagliazzo@Hofstra.edu%3cmailto:John.Impagliazzo@Hofstra.edu> John.Impagliazzo@Hofstra.edu<mailto:John.Impagliazzo@Hofstra.edu>> To: "Paul Ceruzzi" < <mailto:CeruzziP@si.edu%3cmailto:CeruzziP@si.edu> CeruzziP@si.edu<mailto:CeruzziP@si.edu>>, "sigcis" < <mailto:members@sigcis.org%3cmailto:members@sigcis.org> members@sigcis.org<mailto:members@sigcis.org>> Sent: Tuesday, February 24, 2015 2:02:42 AM Subject: Re: [SIGCIS-Members] Washington Post article about Turing Thanks, Paul. Below is the link to the article for those who do not have access to it. <http://www.washingtonpost.com/national/health-science/what-imitation-game-d idnt-tell-you-about-alan-turings-greatest-triumph/2015/02/20/ffd210b6-b606-1 1e4-9423-f3d0a1ec335c_story.html> http://www.washingtonpost.com/national/health-science/what-imitation-game-di dnt-tell-you-about-alan-turings-greatest-triumph/2015/02/20/ffd210b6-b606-11 e4-9423-f3d0a1ec335c_story.html John John Impagliazzo, Ph.D. Professor Emeritus, Hofstra University IEEE Life Fellow ACM Distinguished Educator Editor-in-Chief, ACM Inroads From: <mailto:members-bounces@sigcis.org%3cmailto:members-bounces@sigcis.org> members-bounces@sigcis.org<mailto:members-bounces@sigcis.org> [ <mailto:members-bounces@sigcis.org> mailto:members-bounces@sigcis.org] On Behalf Of Ceruzzi, Paul Sent: Saturday, 21 February, 2015 08:47 To: sigcis Subject: [SIGCIS-Members] Washington Post article about Turing On the front page of today's Washington Post is an article by Joel Achenbach about Turing's 1936 paper and its influence on computer science. All well and good, except later on he quotes the Chair of the Computer Science Department at Princeton as saying "...Turing invented computer science and John von Neumann built the first stored-program computer." An example of The Matthew Effect ("them that's got shall have; them that's not shall lose"). Overall, Achenbach has written an very good summary of Turing's contributions. He also gets one thing right (unless I am mistaken): we really don't know to what extent von Neumann and Turing discussed these concepts when both were at Princeton. _______________________________________________ This email is relayed from <mailto:members@sigcis.org%3cmailto:members@sigcis.org> members@sigcis.org<mailto:members@sigcis.org>, the email discussion list of SHOT SIGCIS. Opinions expressed here are those of the member posting and are not reviewed, edited, or endorsed by SIGCIS. The list archives are at <http://sigcis.org/pipermail/members/> http://sigcis.org/pipermail/members/ and you can change your subscription options at <http://sigcis.org/mailman/listinfo/members> http://sigcis.org/mailman/listinfo/members _______________________________________________ This email is relayed from <mailto:members@sigcis.org%3cmailto:members@sigcis.org> members@sigcis.org<mailto:members@sigcis.org>, the email discussion list of SHOT SIGCIS. Opinions expressed here are those of the member posting and are not reviewed, edited, or endorsed by SIGCIS. The list archives are at <http://sigcis.org/pipermail/members/> http://sigcis.org/pipermail/members/ and you can change your subscription options at <http://sigcis.org/mailman/listinfo/members> http://sigcis.org/mailman/listinfo/members _______________________________________________ This email is relayed from <mailto:members@sigcis.org> members@sigcis.org, the email discussion list of SHOT SIGCIS. Opinions expressed here are those of the member posting and are not reviewed, edited, or endorsed by SIGCIS. The list archives are at <http://sigcis.org/pipermail/members/> http://sigcis.org/pipermail/members/ and you can change your subscription options at <http://sigcis.org/mailman/listinfo/members> http://sigcis.org/mailman/listinfo/members