viewpoints

Vdoi:10.1145/2542504 Thomas Haigh Historical Reflections Actually, Turing Did Not Invent the Separating the origins of computer science and technology.

he 100th anniversary of the birth of was cel- ebrated in 2012. The com- puting community threw its biggest ever birthday party. TMajor events were organized around the world, including conferences or festi- vals in Princeton, Cambridge, Manches- ter, and Israel. There was a concert in Seattle and an opera in Finland. Dutch and French researchers built small Tur- ing Machines out of Lego Mindstorms kits. Newspaper and magazine articles by the thousands brought Turing’s life story to the public. ACM assembled 33 winners of its A.M. Turing Award to dis- cuss Turing’s ideas and their relation- ship to the future of computing. Various buildings, several roads, and at least

one bridge have been named after him. e ’s of C hristi sy Dozens of books with Turing’s name in the title were published or re- Alan Turing (left); the cover page of Turing’s paper “On computable numbers, with issued. Turing was so ubiquitous that an application to the Entscheidungsproblem” (right). even George Dyson’s book about John von Neumann was titled Turing’s Ca- Still, as this flood of Turing-related points of Turing’s actual career. In

thedral, becoming the first book on the material begins to recede it is time to 1936, just two years after completing his e e court pe r I mag Pa a lle ry; history of information technology to clean up some of the rather bad smell- undergraduate degree, he introduced reach a broad audience since the one ing historical claims left in our meta- the concept now called the Turing Ma- about Nazis with punched card ma- phorical basement. Column space is chine in a paper called “On comput- chines. Publishers are well aware there short, so I will focus here on the idea able numbers, with an application to is a strong audience for books about that Turing invented the computer. the Entscheidungsproblem.” This has Nazis. The public’s hunger for books Very short version: it is wrong. since become the main abstract mod- G l Portrait of N ationa sy about mathematicians and computer In case you spent 2012 in a maxi- el of computation used by computer scientists is less acute, making Tur- mum-security prison or meditating scientists. During the Second World ing’s newfound commercial clout both in a Tibetan monastery, let me briefly War Turing made several vital contri-

unlikely and heartening. summarize the computer-related high butions as part of the British team try- e court T uring Photo

36 communications of the acm | january 2014 | vol. 57 | no. 1 viewpoints viewpoints

ing to decipher intercepted German of his death his interests had already communications, which were encoded drifted away from the central concerns using specialized machines and had I could fill many of the nascent discipline. been thought unbreakable. Immedi- columns doing When building a house the founda- ately after the war Turing designed an tion goes in first. The foundations of a electronic computer, the ACE, for the nothing more than new discipline are constructed rather National Physical Laboratory. A series skewering ridiculous later in the process. Turing’s 1936 pa- of machines based on the design were things written about per was excavated by others from the V eventually built, including one of the tradition of mathematical logic in first commercial computer models, Turing, many of them which it was originally embedded and though Turing departed for the Univer- moved underneath the developing sity of before serious con- by people who ought new field. In several papers historian struction began. He worked there with to know better. Michael S. Mahoney sketched the pro- one of the earliest modern , cess by which this body of theory was but soon turned to more abstract and assembled, using pieces scavenged philosophical questions. Pondering the from formerly separate mathematical possibility of what we would now call and scientific traditions. The creators artificial intelligence, Turing proposed of computer science drew on earlier we should judge a computer intelligent disciplines. However, these people had work from mathematical logic, for- if someone could not reliably tell it from backgrounds and appointments in dis- mal language theory, coding theory, a real human after conducting a typed ciplines such as , electrical engineering, and various conversation with both. This procedure mathematics, and physics. When they other fields. Techniques and results is now called the “Turing Test.” Turing’s published articles, supervised disserta- from different scientific fields, many career came to an abrupt end in 1954 tions, or sought grants they had to be of which had formerly been of purely with his death, usually attributed to fit within the priorities and cultures of intellectual interest, were now rein- suicide following various humiliations established disciplines. The study of terpreted within the emerging frame- inflicted by the authorities after a legal computing always had to be justified as work of computer science.a Historians conviction for homosexuality. a means, not as an end in itself. who have looked at Turing’s influence That is a remarkable career by any Ambitious computer specialists on the development of computer sci- measure, with enough tragedy and ge- were not all willing to make that com- ence have shown the relevance of his nius to hook a broader audience and promise and sought to build a new dis- work to actual computers was not make Turing an unlikely gay icon. I cipline. It was eventually called com- widely understood in the 1940s.1,4,5 do not have the expertise to evaluate puter science in the U.S., though other Turing’s 1936 paper was one of the the common claim that Turing’s work names were proposed and sometimes most important fragments assembled shortened the war by several years but adopted. To win respectability in elite during the 1950s to build this new in- even a more cautious evaluation of the research universities the new disci- tellectual mosaic. While Turing him- impact of his wartime accomplish- pline needed its own body of theory. self did see the conceptual connection ments would make him a mistreated The minutiae of electronic hardware he did not make a concerted push to national hero. To celebrate Turing is remained the province of engineering. popularize this theoretical model to therefore to celebrate freedom and Applied mathematics and numerical those interested in computers. How- decency, as well as genius. Let’s just analysis were tied too closely to the ever, the usefulness of his work as a make sure we do our cheering in a his- computer center tradition of service model of computation was, by the end torically responsible manner. work in support of physicists and en- of the 1950s, widely appreciated within gineers. Thus, the new field needed large parts of the emerging computer Retroactively Founding a body of rigorous theory unique to science community. Edgar Daylight Computer Science computation and abstracted from en- has suggested that Turing’s rise in Turing provided a crucial part of the gineering and applied mathematics. prominence owed much to the em- foundation of theoretical computer Turing was not, in any literal sense, brace of his work by a small group of science. There was no such thing as one of the builders of the new disci- theorists, including Saul Gorn, John W. computer science during the early pline. He was not involved with ACM Carr, and Alan J. Perlis, who shared a 1950s. That is to say there were no de- or other early professional groups, did particular interest in the theory of pro- partments of computer science, no not found or edit any journal, and did gramming languages.3 His intellectual journals, no textbooks, and no com- not direct the dissertations of a large prominence has been increasing ever munity of self-identified computer cohort of future computer scientists. since, a status both reflected in and scientists. An increasing number of He never built up a laboratory, set up a reinforced by ACM’s 1965 decision to university faculty and staff were build- degree program, or won a major grant name its premier award after him. ing their careers around computers, to develop research in the area. His whether in teams creating one-off com- name does not appear as the organizer a See part three of Mahoney’s Histories of Com- puters or in campus computer centers of any of the early symposia for com- puting, cited in the Further Reading section at serving users from different scientific puting researchers, and by the time the end of this column.

january 2014 | vol. 57 | no. 1 | communications of the acm 37 viewpointsCACM_JOCCH_one-third_page_vertical:Layout 1 7/30/09 5:50 PM Page 1

So Who Did Invent the Computer? On a conceptual level, the break- This question, asked at a party, will through was inventing what we could cause any responsible historian of now call a computer architecture able to computing to blanch and mumble an take advantage of the flexibility of these excuse before scurrying to the safety new memories. Historians agree that of the drinks table. The whole way we the first wave of modern computers un- write and think about the computers of der construction around the world dur- ACM the 1940s is an attempt to avoid having ing the late 1940s were all inspired by to provide a single answer to that ques- a single conceptual design, an unpub- Journal on tion. Instead we award each early ma- lished typescript cryptically titled “First chine, and its main inventor(s), a meta- Draft of a Report on the EDVAC.” This Computing and phorical trophy engraved with a phrase unfinished document summarized such as “first general-purpose automat- discussions among the team working Cultural ic electronic digital computer.” These on a successor to ENIAC. Its title page trophies adorn the figurative man- named only John von Neumann as its telpieces of John Atanasoff, Konrad author, though the extent to which he Heritage Zuse, J. Presper Eckert, John Mauchly, personally created the ideas within Tom Kilburn, Tommy Flowers, How- rather than summarizing the team’s ard Aiken, and . Those progress has been much debated. Tur- who focus on designs, rather than ac- ing produced his own ACE design only tual functioning machines, can and do after reading and being influenced by make the case for Charles Babbage and this document, though his approach John von Neumann. A colleague once diverges in several interesting respects joked to me that we should identify and from von Neumann’s. honor the earliest computer never to be claimed as the first computer. Arguments For Turing The story behind all those “firsts” As historians followed this progression goes like this. From the late 1930s to of machines and ideas they found few the mid-1940s, a number of automatic mentions of Turing’s theoretical work computing machines were built. Their in the documents produced during the inventors often worked in ignorance of 1940s by the small but growing com- ◆ ◆ ◆ ◆ ◆ each other. Some relied on electrome- munity of computer creators. Turing is chanical relays for their logic circuits, thus barely mentioned in the two main JOCCH publishes papers of while others used vacuum tubes. Sev- overview histories of computing pub- significant and lasting value in eral machines executed sequences of lished during the 1990s: Computer by instructions read one at a time from Campbell-Kelly and Aspray, and A His- all areas relating to the use of ICT rolls of paper tape. Thanks in part to a tory of Modern Computing by Ceruzzi. in support of Cultural Heritage, series of legal battles around a patent Much of the overstatement of Tur- seeking to combine the best of granted on the ENIAC these machines ing’s role, in newspaper articles or by computing science with real dominated early discussion of the his- participants in online discussion, is tory of computing and their creation based on simple misunderstandings. For attention to any aspect of the has been well documented. example, a series of Colossus computers cultural heritage sector. The “modern” or “stored program” was used by the British for wartime code- computers from which subsequent breaking work. These were the first elec- ◆ ◆ ◆ ◆ ◆ computers evolved were defined by tronic digital computers to work prop- two interrelated breakthroughs. On an erly. People often assume, incorrectly, engineering level, computer projects that Turing must have designed Colos- of the late 1940s succeeded or failed sus because he worked at the same se- www.acm.org/jocch based primarily on their ability to get cret facility doing closely related work. large, fast memories to work reliably. I could fill many columns doing www.acm.org/subscribe The first technology proposed, by Eck- nothing more than skewering ridicu- ert who oversaw the engineering of lous things written about Turing, many ENIAC at the University of Pennsylva- of them by people who ought to know nia, was the mercury delay line. Freddy better. We will learn more by looking at Williams, working on the computer the best-supported, most careful argu- project at Manchester University, was ments in favor of the idea that Turing the first to successfully store on a invented the computer. The philoso- cathode ray tube. These were the two pher Jack Copeland has been one of the dominant high-speed memory tech- most passionate and industrious boost- nologies until the mid-1950s. ers of Turing’s role in recent years, un-

38 communications of the acm | january 2014 | vol. 57 | no. 1 viewpoints

leashing a book on Turing’s ACE com- historians. While he provides foot- Turing’s work was not completely puter, another on Colossus, a collection notes to support these assertions they unknown in the 1940s. There is, for of Turing’s work, a website full of archi- are often to interviews or other sourc- example, reliable evidence that von val Turing documents, and a series of es written many years after the events Neumann was aware of the now-fa- journal articles. His work continues the concerned. For example, the claim mous paper and shared Turing’s in- influential legacy of logician Martin Da- that Turing was interested in building terest in the underlying mathematical vis, whose Engines an actual computer in 1936 is sourced questions it addressed. of Logic presented the universal Turing not to any diary entry or letter from the Where one might leap into fantasy machine as the crucial advance behind 1930s but to the recollections of one of is by asserting the cluster of ideas con- the modern computer. Turing’s former lecturers made long tained in von Neumann’s 1945 “First A painstaking and easily accessible after real computers had been built. Draft” are merely a restatement, or summary of the case for Turing comes Like a good legal brief, his advocacy is at most an elaboration, of Turing’s is “Alan Turing: Father of the Modern rooted in detailed evidence but pushes earlier work on computability. Judge Computer” published by Copeland and the reader in one very particular direc- for yourself, by placing side by side Diane Proudfoot in an online journal tion without drawing attention to other Turing’s 1936 “On Computable Num- edited by Copeland.2 This claims that possible interpretations less favorable bers…” and “First Draft of a Report on the “fundamental conception” embod- to the client’s interests. the EDVAC.” They are easy to find with ied in the “First Draft Report” came Google, though you might want to pour from Turing, and that von Neumann Theory vs. Practice yourself a fortifying beverage first as himself “repeatedly emphasized” this. Arguments of this kind raise funda- neither is particularly easy reading. Copeland also believes that “right from mental issues about the connection The former is a paper on mathemat- the start” Turing was interested in of theory and practice. Are abstract, ical logic. It describes a thought experi- building an actual computer based on theoretical insights more fundamen- ment, like Schrödinger’s famous 1935 the conceptual mechanism described tal than pragmatic, engineering-based description of a trapped cat shifting be- in his 1936 paper. This extends a recent advances? Must theoretical break- tween life and death in response to the trend, seen for example in George Dy- throughs precede and guide practical behavior of a single atom. Schrödinger son’s book, to write about the teams ones? For a computer scientist, in par- was not trying to advance the state of working to build computers in the late- ticular, it is easy to assume that Tur- the art of feline euthanasia. Neither 1940s as if they launched their projects ing’s theoretical work was as centrally was Turing proposing the construction primarily to build practical realizations important to the computer designers of a new kind of calculating machine. of Turing’s abstract machine. of the 1940s as it later becomes within As the title of his paper suggested, Tur- Copeland is deeply knowledgeable computer science. There is also some- ing designed his ingenious imaginary about computing in the 1940s, but as a thing undeniably attractive in the story machines to address a question about philosopher approaches the topic from of a lone genius who anticipates the the fundamental limits of mathemati- with a different perspective from most rest of the world by many years. cal proof. They were structured for chnica l Library s of th e AR L Te

rmy Photo from th e archi ve U . S A rmy Photo Two programmers wiring the right side of the ENIAC with a new program.

january 2014 | vol. 57 | no. 1 | communications of the acm 39 viewpoints

John von Neumann with the IAS computer circa 1951 (left); cover page of von Neumann’s “First Draft of a Report on the EDVAC” (right).

simplicity, and had little in common ously. The second, the “von Neumann cheaper, smaller, more reliable, and with the approaches taken by people Architecture Paradigm,” set out the ba- more flexible than their predecessors. designing actual machines. sic structure of the modern computer: ENIAC, the first general-purpose elec- Von Neumann’s report said nothing special-purpose registers on which all tronic digital computer, used almost explicitly about mathematical logic. It operations were performed and from 18,000 vacuum tubes. The more tubes described the architecture of an actual which data was exchanged with main a machine held the more expensive planned computer and the technolo- memory, separation of arithmetic it was to build and, as they eventually gies by which it could be realized, and functions from control functions from burn out, the less reliable. Its immedi- was written to guide the team that had memory units, only one action per- ate successors held 1,000 or 2,000 tubes already won a contract to develop the formed at a time, and so on. The third, yet could handle problems of greater EDVAC. Von Neumann does abstract the “Modern Code Paradigm,” con- logical complexity and were easier to away from details of the hardware, cerns the nature and capabilities of its program. This efficiency made pos- both to focus instead on what we would instructions. For example, instructions sible the construction of computers now call “architecture” and because were expressed as through a small vo- in cash-strapped Britain following the the computer projects under way at cabulary of operation codes followed war, and made computers affordable the Moore School were still classified by argument or address fields. These and useful enough that they were rap- in 1945. His letters from that period were held in the same numbered mem- idly turned into commercial products are full of discussion of engineering ory cells as data. While executed by de- and applied to business tasks as well as details, such as sketches of particular fault in a particular sequence, the ma- scientific computations.

models and their perfor- chine could jump out of sequence and According to Copeland, “the fun- s mance characteristics. the destination of this jump could be damental conception of the stored- The phrase “stored program con- modified as the program ran based on program universal computer” was Tur-

cept” has sometimes been used to the state of the computation. ing’s. Von Neumann merely “wrote the ania A rchi ve encapsulate the content of the “First Taken together, von Neumann’s first paper explaining how to convert c Draft” report, but this underplays its cluster of ideas guided the construc- Turing’s ideas into electronic form.” e for s Ce nt e r, I nstitut actual impact by implying it held just tion of computers that were much But what actually would have been one big idea. In fact it provided a wealth different about von Neumann’s “First of intertwined ideas and details. In my Draft” report if Turing had never writ- current work with Mark Priestley and The universal ten his now famous paper? My answer on L ev y A rchi ve

Crispin Rope I have found it useful to that question is: nothing (with the of th e U ni ve rsitylv sy e nnsy of P to separate these into three main ar- Turing Machine possible exception of the neuron nota- eas.b The first, the “EDVAC Hardware has appealed to tion he appropriated to describe logic Paradigm” described an all-electronic gates, whose creators cited Turing). binary computer with a much larger theorists from Copeland has gone so far as to ar- memory than anything ever built previ- the 1950s onward. gue the basic idea of a single machine Whit e and L of th e S h el by sy that could do different jobs when fed e e C ourt pe r I mag ); Pa ( IAS tudy b “Reconsidering the Stored Program Con-

cept,” forthcoming in IEEE Annals of the His- c See http://www.huffingtonpost.com/jack-cope- anc e d S v A d tory of Computing. land/what-apple-and-microsoft-_b_3742114.html e p h C ourt Photogra

40 communications of the acm | january 2014 | vol. 57 | no. 1 viewpoints different instructions can be traced to Computer Science vs. Computing tion adopted precisely because it hid all Turing. But Charles Babbage had that Our urge to believe the computer proj- the messy issues of architecture and en- idea long before, and as mentioned ects of the late 1940s were driven by a gineering needed to make any real com- earlier, several computers controlled desire to implement universal Turing puter function. Hardware and software by sequential instruction tapes had al- machines is part of a broader predis- are interchangeable to the theorist, but ready been constructed with no influ- position to see theoretical computer not to the historian. ence from Turing and were well known science driving computing as a whole. to von Neumann before he wrote his If Turing invented computer science, Further Reading report. EDVAC went far beyond this to which is itself something of an oversim- store a program in addressable inter- plification, then surely he must have in- Aspray, W. nal memory rather than on a sequen- vented the computer. The computer is, John von Neumann and the Origins of Modern Computing. MIT Press, 1990. tial instruction tape. To suggest this in this view, just a working through of A thorough and careful survey of von advance came from Turing is odd, as the fundamental theoretical ideas rep- Neumann’s many contributions to early the machine Turing described had no resented by a universal Turing machine computing, including his work on the “First internal writable memory and took in that it is universal and stores data Draft of a Report on the EDVAC.” its instructions from a tape. Von Neu- and instructions interchangeably. Copeland, J. Turing: Pioneer of the Information Age. mann brought a concern with logic This line of thinking blurs the fun- Oxford, 2013. A concise summary of and preference for minimal, general- damental distinction between build- Copeland’s work on Turing’s ideas and their purpose mechanisms to the design of ing something and modeling it. Cope- legacy. He has produced related volumes EDVAC but he did not need Turing to land shows that as early as 1949 von on Turing’s planned ACE computer and the teach him that. He was a mathematic Neumann alluded to Turing’s abstract wartime Colossus work. genius with a deep pragmatic streak model of computation as an interest- Hodges, A. Alan Turing: The Enigma (Centenary Edition). and an astonishing track record of pro- ing proof that automata with a certain Princeton University Press, 2012. ductive collaborations across a huge “minimum level of complexity” could An updated edition of the monumental range of fields. simulate each other’s functioning. Yet biography that originally put Turing on Turing’s 1936 paper lacks many finding an abstraction useful or pro- the road to broader fame. novel and fundamental features found vocative as a model of a particular real Lavington S., Ed. in the “First Draft” such as addressable system does not imply the design of the Alan Turing and His Contemporaries: Building the World’s First Computers. British memory locations. Neither did Turing real system was patterned on the ab- Informatics Society, 2012. A concise and describe instruction codes followed straction. An abstraction, ultimately, is clearly written expert history, honoring by arguments, the building blocks of useful because of what it leaves out. Turing’s accomplishments and placing computer programs. The suggestion To focus on historical computers them in the context of British computer developments during the 1940s. that the EDVAC design was merely a primarily as embodiments of logical conversion of Turing’s paper implies Levy, P. ideas, ignoring the trade-offs their cre- “The Invention of the Computer.” In Serres, M. these features are trivial, and the sin- ators made when faced with limited re- (Ed.) A History of Scientific Thought. Blackwell, gle important idea in each document sources and unproven technologies, is 1995. Concise and thoughtful in its summary is that code and data should be treat- to abstract away from the information of key early computers and their relationship ed interchangeably so programs can needed to understand their history and to technologies, applications, and Turing. modify themselves. Yet while Turing’s development. Progress in electronic en- Mahoney, M.S., Ed. Haigh, T. Histories of Computing. Harvard University paper showed one machine could, in gineering, particularly in memory tech- Press, 2011. Section three of this book, “The modern terms, emulate the function- nologies, created the circumstances in Structures of Computation,” is a provocative ing of another it never described a which it began to make sense to think selection of papers on the origins of machine altering its own instructions. about high-speed digital computers in theoretical computer science and its Furthermore, at the very end of the which instructions were stored elec- relationship to computation and simulation. “First Draft” von Neumann expressly tronically. In turn, ideas about the best References forbade EDVAC from overwriting the way to design these machines drove fur- 1. akera, A, Calculating a Natural World, MIT, 2006. operation fields in its instructions, ther progress in component technolo- 2. copeland, B.J. and Proudfoot, D. Alan Turing: Father of the modern computer. Rutherford Journal even though he relied on modifications gies and engineering methods. 4, 2011–2012; http://www.rutherfordjournal.org/ to their address fields to accomplish The universal Turing Machine has article040101.html. 3. daylight, E.G. Towards a historical notion of ‘Turing— basic operations such as conditional appealed to theorists from the 1950s on- The father of computer science.’ To appear in History branching. This address modification ward precisely because it abstracts away and Philosophy of Logic; www.dijkstrascry.com/ TuringPaper. was a very influential idea in the “First from the complexity of real computer 4. mounier-Kuhn, P. Logic and computing in France: Draft,” but was, of course, absent from A late convergence. International Symposium on architectures and decouples questions History and Philosophy of Programming; http://www. Turing’s paper as his machines did not of computability from those of design computing-conference.ugent.be/file/12. use addresses. In other words, the ca- and engineering. This has been enor- 5. Priestley, M. A Science of Operations. Springer, 2010. pability for unrestricted self-modifying mously useful for computing theorists, Thomas Haigh ([email protected]) is an associate code von Neumann is said to have cop- both technically and sociologically. Yet, professor of information studies at the University of ied from Turing is something Turing paradoxically, the world seems increas- Wisconsin, Milwaukee, and chair of the SIGCIS group for historians of computing. did not describe and von Neumann’s ingly eager to locate the origin of the design explicitly prohibited. computer in a mathematical abstrac- Copyright held by Author/Owner(s).

january 2014 | vol. 57 | no. 1 | communications of the acm 41