A Second Modernism

MIT, Architecture and the ‘Techno-Social’ Moment

Edited by Arindam Dutta with Stephanie Marie Tuerk Michael Kubo Jennifer Yeesue Chuong

1 ALISE UPITIS Two or More Architectures

Computers and Design at MIT until 1963

In 1962 and 1963, Christopher Alexander method of solving problems of engineering was a research affiliate at the MIT Civil design; the next two reports detail the Engineering Systems Laboratory (CESL). computer programs that primarily Alexander At the time he was completing his PhD in wrote for implementing the method.4 architecture from the Graduate School of This dual focus on developing new computer Design at Harvard University, about a year systems for engineering design while before publishing his dissertation as Notes simultaneously investigating design methods on the Synthesis of Form. At the time of its was not uncommon in the context of MIT publication, Progressive Architecture heralded engineering at the time. Discussing a research that it “could revolutionize the approach to endeavor termed the Computer -Aided Design architectural design.”1 The journal Industrial project--a joint venture of the MIT Electronic Design stated it was “one of the most Systems Laboratory, Department of Electrical important contemporary books about the Engineering, and the Design Division, art of design.”2 The Journal of the American Department of Mechanical Engineering Institute of Planners predicted that “it may initiated in 1959--Steven Coons stated: one day prove to be a landmark in design methodology.”3 Notes has now gone through Out of the investigation will five editions and is in its 17th printing. come the design for a man- It seems Alexander’s entry to CESL machine organism to accomplish was through a course he took in the MIT the design process in a way far Department of Civil Engineering during easier than has ever before been the spring of 1961, “Transportation Route possible; but as by-products will Location.” Between March 1962 and June come new computer techniques 1963, he produced four CESL research and an enriched understanding of reports; the first three were co-authored the creative thought process.5 with Marvin Manheim, an instructor in the department. The first two reports they A dual concern was also manifest in the head produced use issues in highway engineering of CESL, Charles Miller. Under Miller the as a point of departure for proposing a general Photogrammerty Laboratory was renamed the

1 R. H. Mutrux, “Revo- Planners XXXI, no.1 (February Systems Laboratory, Cambridge, 5 Steven Engineering, School of Engineer- lutionary Approach to Design,” 1965): 84. MA. 1962); Alexander and Man- Anson Coons, “An Outline of the ing, Massachusetts Institute of Progressive Architecture XLVI, 4 Alexan- heim, “The Use of Diagrams Requirements for a Computer- Technology, 1963), 5. no. 5 (May 1965): 208. der and Manheim, “The Design in Highway Route Location” Aided Design System,” Proceed- 7 Ibid, 7. 2 Malcolm of Highway Interchanges” (Re- (Research Report 62-03, M.I.T. ings of the May 21-23, 1963, 8 Chris- J. Brookes, “Books,” Industrial search Report 62-01, M.I.T. Civil Civil Engineering Systems Labo- spring joint computer conference topher Alexander, Notes on the Design 12, no.3 (March 1965): Engineering Systems ratory, Cambridge, MA. 1962); (New York: ACM, 1963), 300. Synthesis of Form (Cambridge, 20. Laboratory, Cambridge, MA. Alexander, “HIDECS 3” (Re- 6 Charles MA: Harvard University Press, 3 Edward 1962); Alexander and Manheim, search Report 63-27, M.I.T. Civil Miller, , “Man-machine com- 1964). 20. J. Kaiser, “Book Reviews,” Jour- “HIDECS 2” (Research Report Engineering Systems Laboratory, munication in civil engineering,” ( 9 Ibid., 80. nal of the American Institute of R62-02, M.I.T. Civil Engineering Cambridge, MA. 1963). Cambridge, Mass., Dept. of Civil 10 Alexan- 516 Civil Engineering Systems Laboratory in 1960, the true engineers set in motion the new name signifying the lab’s change in extensive information processing focus under his leadership from investigating activities by a vast array of survey methods using analog photographic information handlers. Draftsmen and projection technologies to using digital and technicians of various kinds computers to process spatial data. But produce information displays.7 he also believed most existing attempts at engineering computer programs failed The computers members of the Computer- to embrace the revolution possible to civil Aided Design project deployed in the early engineering methods afford by new digital 1960s was often the TX series, developed at computers, instead mindlessly encoding Lincoln Laboratory. The computers Alexander existing design methods and theories: “The deployed were from the IBM 700/7000 series. most important by-product of attempts to use This paper investigates how these early computers is the extent to which such activity computers’ hardware enabled and constrained is now forcing us to critically examine the the development of computer programs for true role and function of the engineer and the which the problem the program posed to nature of the engineering process [emphasis solve was a design problem, and in turn how original]”6 this dialectic informed how the process of Miller’s interest included advocating for design was conceptualized. “problem-oriented programming languages” that he more accurately reflected the Alexander and his method engineering design process, yet his model for the understanding of how engineering In Notes, Alexander specifies that design is to design operated was the hardware of digital solve the problem of securing a “fit” between computer comprising information store, form and context. An example Alexander transmitter, and processor. The process of gives is when a new piece of metal must fit design is taken to imitate the very model by flush against an existing piece. The new piece which Miller tries to imitate it: of metal serves as the form and the context is the already existing piece. Fitness in this Large amounts of raw information instance is flatness. What is “required” of are fed in by surveyors of milling the metal form is that it must be flat.8 various types. The information A misfit is then a bump on its surface. When is reduced, processed, stored, a misfit exists a requirement is not met. and later combined with other Requirements are “the individual conditions kinds of information which are which must be met at the form-context stored in various kinds of files boundary in order to prevent misfit.”9 maintained by the organization. Alexander and Manheim’s method for Significant decisions made by approaching the design of highways follows der and Manheim, “The Design p.6-7 Research in Mid-Century,” Social Program for the Hierarchical thesis, Department of Architec- of Highway Interchanges” (Re- 12 Ibid, “Ap- Studies of Science 29, no. 5. Decomposition of a Set which ture, MIT, 1966. search Report 62-01, M.I.T. Civil pendix 2,” 174-191. (Oct., 1999): 685-718, 693 has an Associated Linear Graph, 19 MIT, Re- Engineering Systems 13 Walter 16 Claude (Cambridge, MA: Dept. of Civil port of the President 1968, Mas- Laboratory, Cambridge, MA. Gropius, The Scope of Total Archi- Elwood Shannon and Warren Engineering, Massachusetts In- sachusetts Institute of Technology 1962), 4. tecture (New York: Collier Books, Weaver, The Mathematical The- stitute of Technology, 1962), 7. Bulletin 103, no. 4 (December 11 Alexan- 1943 [1970]), 56 ory of Communication (Urbana: 18 Nicholas 1968): 32 der and Manheim, “HIDECS 2” 14 Ibid, 83 University of Illinois Press, 1949), Negroponte, “The Computer 20 Serge (Research Report R62-02, M.I.T. 15 Philip 18-22. Simulation of Perception during Chermayeff and Christopher Alex- Civil Engineering Systems Labo- Mirowski, “Cyborg Agonistes: 17 Alex- Motion in the Urban Environ- ander, Community and Privacy ratory, Cambridge, MA. 1962), Economics Meets Operations ander, HIDECS 2: A Computer ment,” unpublished master’s (Garden City, NY: Doubleday & 517 ALISE UPITIS

the same logic: The set of vertices and the set of links together define a Every problem of designing a physical graph; that graph, by virtue of form has these two fundamental the correspondence between characteristics: vertices and requirements, and the correspondence between links 1. There are certain requirements and interactions of requirements, the form must meet represents, for the purpose of this analysis, the structure of the 2. Many of these requirements design problem.11 conflict with one another The objective of any design Take Alexander’s example of a tea kettle. process is to find a form which One requirement might be “must heat as

manages to meet all the rapidly as possible.” Call this requirement m1. requirements, in spite of the Another requirement might be “must be as

conflicts. We call the process of light as possible.” Call this m2. Now call L the inventing any such form “solving set of interactions between requirements. the design problem.”10 Alexander treats the set L as a set of connections or links with each link joining

The method these two characteristics two elements mi in set M. If two elements demand has three steps: (1) Create a planar interact a link is coded 1 and 0 if they do not.

graph; (2) Transform the graph into a tree In this example m1 and m2 interact, so l1 (the

structure; (3) Diagram the design. link between m1 and m2) would be coded 1. The two sets M and L are then used to

(1) Create a planar graph Figure 1. Graph to be decomposed. From The design method states that each Christopher Alexander and Marvin L. Manheim, requirement is to be represented as a HIDECS 2: A Computer point in two-dimensional space. When two Program for the Hierar- chical Decomposition of a requirements interact this relationship is Set which has an Associ- represented as a line between each point that ated Linear Graph (1962). represents a requirement. Specifically, each point represents a vertex of a planar graph, meaning the graph’s edges intersect only at their endpoints, while each line between two points is considered a link between two vertices (Figure 1):

Company, 1963). topher Alexander and Marvin of Systems which have an Asso- pdf (accessed October 23, 2009); 24 Chris- 21 IBM, IBM L. Manheim, HIDECS 2: A ciated Linear Graph (Cambridge, Martin H. Weik, “IBM 7090,” A topher Alexander and Marvin L. Electronic Data-Processing Ma- Computer Program for the Hier- MA: Dept. of Civil Engineering, Third Survey of Domestic Elec- Manheim, HIDECS 2, 19-23. chines Type 704 Manual of Op- archical Decomposition of a Set Massachusetts Institute of Tech- tronic Digital Computing Sys- 25 eration: Preliminary Edition, Form which has an Associated Linear nology, 1963). tems, Report No. 1115 (Aberdeen Christopher Alexander and 24-6661-1 (New York: IBM, 1956), Graph (Cambridge, MA: Dept. of 23 IBM, Proving Ground, MD: Ballistic Marvin L. Manheim, HIDECS http://www.cs.virginia.edu/bro- Civil Engineering, Massachusetts IBM 709 ...A Powerful New Data Research Laboratories, 1961), 2, G 1.1; IBM, FORTRAN II As- chure/images/manuals/IBM_704/ Institute of Technology, 1962); Processing System (New York: 548-457, http://www.ed-thelen. sembly Manual (FAP), Form IBM_704.html (accessed Novem- Christopher Alexander, HIDECS IBM, 1957), http://archive.com- org/comp-hist/BRL61-ibm7070. C28-6235-2 (Poughkeepsie, ber 14, 2009). 3: Four Computer Programs for puterhistory.org/resources/text/ html#IBM-7090 (accessed Octo- NY: IBM, 1963), http://archive. 22 Chris- the Hierarchical Decomposition IBM/IBM.709.1957.102646304. ber 23, 2009). computerhistory.org/resources/ 518 define a linear graph G(M,L) in which the equally the diagram of neutron- requirements or elements in M are nodes scattering problem in atomic and the interactions or elements in L are links bomb design, the flow chart of a between nodes.12 computer program, and (inverted) the organization chart of a military (2) Transform the graph into a tree structure unit or a multi-division corporation - the better to justify the transfer Bauhaus founder Walter Gropius remarked of mathematical techniques.15 Figure 2. Tree struc- that “the training of an architect should be ture showing subsets of requirements with concentric rather than sectional.”13 Jonas The next step in the generalized method of least interaction. From Itten’s representation of the Bauhaus Alexander’s is to convert this graph into a Christopher Alexander, “Determination for curriculum from 1923 is a series of concentric hierarchical structure or ‘tree’. This conversion Components of an Indian circles. The “Basic Course” occupies the is achieved through a step by step process, Village,” in Conference on Design Methods: perimeter, moving inward with “study,” the first step of which is to partition the entire Papers Presented at the “materials,” and finally “Building” in the set of vertices into two subsets for which Conference on Systematic center circle. He elaborates that “modern the least number of links connect the two and Intuitive Methods in Engineering, Industrial architecture is not a few branches of an old subsets. Design, Architecture and tree - it is a new growth coming right from the Each of these two subsets is then Communications (Lon- don, September 1962). roots.”14 partitioned into two further subsets, each of which is in turn connected with the least number of links. This stepwise process of partitioning continues until the initial set of vertices has been decomposed into the most independent subsets possible. The criterion used to partition the vertices into subsets is derived from “information-theoretic considerations”--specifically, it is a measure of “information transfer” adopted from Claude Shannon’s formula for the expected information content transmitted through a communications channel. 16 Alexander’s tree is instead a mathematical This hierarchical decomposition results tree (Figure 2). As economic historian Philip in a tree structure which in turn provides Mirowski notes of the transferability of this an ordered process for solving the design theoretical construct: problem at hand: “The tree specifies which requirements are to be considered together The extensive form of the play of and the order in which different groups a game as a ‘tree’ could resemble of requirements are to be combined and text/Fortran/102679279.05.01. Redmond and Thomas M. Smith, of Computing 5, no. 4 (October (Cambridge, Mass: Research GRAPH, 1989), 44. acc.pdf (accessed October 28, From Whirlwind to MITRE: The 1983): 341 Laboratory of Electronics, Massa- 32 John A. 2009); IBM, FORTRAN I, II, R&D Story of the SAGE Air 28 Douglas chusetts Institute of Technology, McKenzie, TX-0 Computer His- and 709: Customer Engineer- Defense Computer (Cambridge, Ross, “A Personal View of the 1999). tory, 19. ing Manual of Instruction, Form MA: MIT Press, 2000), 86. Personal Workstation,” HPW ‘86 30 Ibid 33 Jan Hurst R23-9518-0 (Poughkeepsie, NY: 27 Morton Proceedings of the ACM Confer- 31 Jan Hurst et al., “Retrospectives II,” 44. IBM, 1959), http://archive.com- M. Astrahan and John F. Jacobs, ence on The history of personal et al., “Retrospectives II: The 34 Wesley puterhistory.org/resources/text/ “History of the Design of the workstations, 21. Early Years in Computer Graphics A. Clark, “The Lincoln TX-2 Fortran/102679237.05.01.acc.pdf SAGE Computer—The AN/FSQ- 29 John A. at MIT, Lincoln Lab, and Harvard,” Computer Development,” Papers (accessed October 28, 2009). 7,” McKenzie, TX-0 Computer His- SIGGRAPH ‘89 Panel Proceed- Presented at the February 26-28, 26 Kent C. Annals of the History tory, Technical Report no. 627 ings (Boston, MA: ACM SIG- 1957, Western Joint Computer 519 ALISE UPITIS

considered.”17 graphics for aid in urban design. During his master’s Negroponte was able

(3) Diagram the design to access the facilities of Project MAC at Figure 3. Stage in MIT’s Computation Center. MAC was an early combining subsets of requirements with least In Notes, once a designer has obtained the pioneer in time-sharing (through CTSS and interaction into diagrams. decomposed subsets of requirements, her or MULTICS) and also supported the interactive From Christo-pher Alex- his goal is to create “constructive diagrams” computer-aided design system Kludge. But ander, “Determination for Components of an Indian or drawings by hand that mutually defines before at least the fall of 1963 when Project Village,” the physical form of a design and respects its functional requirements. Each decomposed subset is to correspond to one such drawing. These drawings are then combined, moving up levels of the tree hierarchy, to give the complete design (Figure 3).

Alexander and his computer

1963, the year Sutherland and Roberts completed their doctoral work, computing courses had yet to be taught in MIT’s School of Architecture and Planning. Three years later computing remained a nascent interest, although a master’s thesis in architecture— “The Computer Simulation of Perception during Motion in the Urban Environment” by Nicholas Negroponte—would serve as harbinger for the school’s research.18 By 1968 computers were gaining a foothold with at least six professors in the school developing computer applications.19 This count includes newly-appointed Negroponte and Leon Groisser for efforts in their newly-formed Architecture Machine Group on URBAN 5 to permit experimental interactive computer- aided urban design through remote consoles linked by satellite. It also includes William L. Porter for his DISCOURSE series system for processing and manipulating information and

Conference (New York: ACM, puter Engineering: A DEC View of Sutherland, “Sketchpad, A Man- nications of the ACM, Volume 4 puter Metaphor of Mind,” History 1957), 145. Hardware Systems Design, eds. Machine Graphical Communica- Issue 3 (March 1961), 147. of Psychology, 2, no.1 (February 35 Charles J. C. Gordon Bell, J. Craig Mudge, tion System,” unpublished PhD 43 Ibid., 15. 1999): 64n125. Bashe, Lyle R. Johnson, John H. John E. McNamara (Bedford, dissertation, Department of Elec- 44 George A. 46 Letter Palmer, and Emerson W. Pugh, MA: Digital Press, 1978), 126-7. trical Engineering, MIT, 1963, p.3. Miller, Eugene Galanter, and Karl from Jerome Bruner to Mc- IBM’s Early Computers (Cam- 37 Ibid., 129 40 Ibid., 2. H. Pribram, Plans and the Struc- George Bundy, November 4, bridge, MA: MIT Press, 1986), 38 Paul E. 41 Ibid., 25. ture of Behavior (New York: Holt, 1960, Box 3, Papers of Jerome 184. Ceruzzi, A History of Modern 42 Douglas Rinehart, and Winston, 1960). Bruner, 36 C. Gor- Computing (Cambridge, MA: MIT Ross, “A generalized technique 45 Hunter Correspondence, don Bell et al., “The PDP-1 and Press, 1998), 128. for symbol manipulation and Crowther-Heyck, “George A. 1961-1962 (HUG 4242.5), Harvard Other 18-bit Computers,” in Com- 39 Ivan numerical calculation,” Commu- Miller, Language, and the Com- University Archives. 520 MAC began really taking hold, an architect in means it processes electrical impulses that Cambridge hoping to use a computer could caused electronic circuits to turn on or off. basically seek out the TX series or a batch- On-off states of circuits represented numbers processing IBM series. Christopher Alexander in binary or base two. Parallel means instead sought an IBM. of transmitting one bit at a time, its 36-bits Alexander’s programming from the late (defining the size of operations it could 1950s until 1963 was performed on a series undertake) were transmitted simultaneously of IBM mainframes. He utilized an IBM over a bus (controlling the passage of 704 at the MIT Computation Center when electrical signals) with one channel for each programming for Community and Privacy, bit. It had magnetic core storage (planes of a book he co-authored with architect Serge small rings of ferromagnetic ceramic with Chermayeff while he was affiliated with the wires threaded through their core whose Joint Center for Urban Studies of MIT and magnetism corresponded to a 0 or 1) which Harvard.20 First issued in 1954 the 704 was took 12 μs (millionths of a second) from a stored program, parallel digital computer which to retrieve stored data .21 It was the (Figure 4, Figure 5). Data and instructions first production machine with magnetic core were stored in the same memory. It was the storage. An IBM 709 at the MIT Computation opposite, in a sense, of a program controlled Center and an IBM 7090 at the Smithsonian computer that was “hard-coded,” for which Astrophysical Observatory in Cambridge every new program to be run manually set were used for his programming at CESL switches and patches were needed to control and for Notes.22 7090 was a transistorized signals and transmit data. A digital computer version of the vacuum tube-driven (used to Figure 5. Architecture of the 704.

47 Harvard Way People See,” Perceptual 48 Plans is 51 Ibid., 80. Artificial Intelligence (New York: University, Center for Cogni- and Motor Skills 19 (1964): 235- also discussed in Christopher 52 Alexan- Basic Books, 1993),26-50; Paul tive Studies, Annual Report 253. Alexander also served as Alexander,“Information and an der and Manheim, “HIDECS 2” N. Edwards, The Closed World: 1961/1962 (Cambridge, MA: research assistant for an under- Organized Process of Design.” (Research Report R62-02, M.I.T. Computers and the Politics of Harvard University, 1962) 15, graduate course “Psychological New Building Research, Building Civil Engineering Systems Labo- Discourse in Cold War America 18. Christopher Alexander, “The Conceptions of Man” taught Research Institute (1961): 115- ratory, Cambridge, MA. 1962), (Cambridge, MA: MIT Press, Origin of Creative Power in Chil- by George Miller and Jerome 124. p.6-7 1996), 124. dren,” British Journal of Aesthet- Bruner. Harvard University, Cen- 49 Miller, 53 Daniel 54 A. Newell ics 2, no.3 (July 1962): 207-226; ter for Cognitive Studies, Annual Galanter, and Pribram, Plans, 21- Crevier, “The First AI Program: and J.C. Shaw, “Programming Christopher Alexander and A. W. report 1960/1961 (Cambridge, 39. Defining the Field,”AI: The Tumul- the Logic Theory Machine,” F. Huggins, “On Changing the MA: Harvard University, 1961), 7. 50 Ibid., 81. tuous History of the Search for Papers Presented at the Febru- 521 ALISE UPITIS

perform logical operations) 709. With 50,000 instructions into a deck of punched cards Figure 4 (at right). IBM 704 Data Processing transistors and a memory cycle of 2.18 μs, which computer technicians would input to System. the 7090 was time five-times faster than the the mainframe. The computer would execute 709.23 one program at a time and technicians would Alexander wrote a series of programs return a printout when the job was finished. written using the FORTRAN (short for IBM For all three issues, Alexander input data and Mathematical Formula Translating System) programs by way of a stack of punch cards Assembly Program (FAP), the assembler and received output in the form of assembly for the 709 and 7090. Similar to a compiler, listings or printed outputs of FAP assembly an assembler translates the mnemonics (see Figure 4, Figure 5). of symbolic language, like FORTRAN, Programming in FAP demands close into machine instructions which are in the understanding of the computer’s architecture form of binary numbers. Alexander termed and operations, and his mode of data his FAP programs HIDECS (Hierarchical preparation mapped readily to the machine’s Decomposition of Systems) and they were functionality. Alexander represented executed under control of the Fortran Monitor interactions between requirements as a

Figure 6. Layout for IBM 700/7000 series punch cards.

System, the operating system that managed square data matrix of 1s and 0s. A 1 in the ith the central control, memory, and so on. row and jth column indicates an interaction Each 7-series Alexander used operated or link between requirements i and j and a by batch processing, which means those 0 indicates a lack of interaction. Each row of who wanted to make use of a computer’s the matrix then served as a machine word resources typically translated their data and or binary vector. There was also a certain

ary 26-28, 1957, Western Joint (New York: ACM, 1963), 299-304. ing boundary values. Steven A. Proceedings of the ACM Confer- biosis,” 7. Computer Conference (New York: 57 Steven Coons, Surfaces for Computer- ence on The history of personal 63 Licklider, ACM, 1957), 232n3. Anson Coons, “An Outline,” 300. Aided Design of Space Forms, workstations, 22. ”Man-Computer Symbiosis,” 5; 55 Ibid., 58 Coons MAC-TR-41 (Cambridge, MA: 61 J.C.R. Engelbart, Augmenting Human 239. pioneered during the 1960s tech- Project MAC, Massachusetts Licklider, ”Man-Computer Intellect, http://dougengelbart. 56 Steven niques for describing surfaces in Institute of Technology, 1967). Symbiosis,” IRE Transactions on org/pubs/augment-3906.html. Anson Coons, “An Outline of the computer graphics to describe 59 Suther- Human Factors in Electronics, Requirements for a Computer- how a “patch”--the interior of land, “Sketchpad,” 1. HFE-1 (March 1960), 4. Paul N. Aided Design System,” Proceed- a three-dimensional surface 60 Douglas Edwards, The Closed World, 226. ings of the May 21-23, 1963, defined by four boundary curves- Ross, “A Personal View of the 62 J.C.R. spring joint computer conference -can be constructed by interpolat- Personal Workstation,” HPW ‘86 Licklider, ”Man-Computer Sym- 522 523 ALISE UPITIS

ontological continuity between the data matrix into the final goal of overall least information of 1s and 0s and these digital computers. transfer was met (see Figure 17). Each separate punch card for input Alexander’s programs remain as assembly contained one row of the matrix across listings or printed outputs of FAP assembly 72 columns; if the matrix contained (see Figure 15). Consider an output line additional columns, these were punched generated by HIDECS 2: on an additional card (see Figure 14). The Figure 7. FAP printout of HIDECS 3. From Alexan- 709/7090 had a word length of 36 bits, so for 0004 -634 00 4 00021 INPAR SXD IR4,4. der, HIDECS 3, 26. computational simplicity Alexander separated the cases in which the matrix was of order 36 or less (has 36 by 36 or fewer cells) and the cases in which the matrix is of order 37 or more. A different sequence of subprograms was used for each case. The maximum matrix size was of order 252, which is dependent upon computer storage space allocations. 24 Each program Alexander ran was comprised of three groups of subprograms. The first subprogram, Preliminary, read in and generated parameters and set variables to control loops in other subprograms. The second, Data, read a binary data matrix, checked it for inconsistencies, and structured it for analysis. The third, Analysis, decomposed the matrix and was executed through seven further subprograms. These subprograms decomposed matrices into subsets of least interaction or least information transfer. On the whole, multiple sub-routines served to feedback processes within a hierarchy of sub-routine levels. The goal at each level of analysis was to apply a series of repetitive comparative checking until the state of the program was equal to the goal state of least information transfer. Results were then stored and control was transmitted to a next level sub-routine. This continued

524 INPAR is the first subprogram executed in It not only reminds how the architecture of the grouping of Preliminary subprograms the computer determined the structure of and inputs control parameters. The details Alexander’s programs, it is also a signifier of of the line not signifcant other than to say how Alexander in 1962 using the IBM 7-series the numbers to the left refer to binary words mainframes was not able to intervene in produced by the assembler given in octal the operation of a program so that it could (since octal or base 8 allowed a complete respond and change its course accordingly. machine word of 36 bits to be represented By way of contrast, in the fall of 1952 the using only 8 digits) and includes the location Whirlwind digital computer, located on the of each instruction and the binary word second floor of the Barta Building at MIT, had assigned to the location; the the right is the 3 16-inch cathode-ray tubes or CRTs-- the related symbolic data.25 displays for television receivers throughout The output above is a representation the 20th century--to input instructions and data of the machine’s operations and locations to the machine and output computations and that permit the program to run, while these operations records. More than 20 such CRTs same instructions generate the output itself. were installed by late spring of 1954. Nearly Project Whirlwind, 1954. Left to right: Jay W. For- rester, Norman H. Taylor, Gus O’Brien, Norman Draggett, Charles L. L’Orderman (on ladder).

525 ALISE UPITIS

all the scopes were held inside consoles graphical information such as geographical that contained various switches, buttons, maps locating radar towers (Figure 8).26 as well as a “light gun”--used to detect light The system, under the auspices of pulses from electrons excited in the CRT’s Division 6 of Lincoln Laboratory, was a phosphor coating-- worked to register data prototype for SAGE, acronym for Semi- to the computer that would reconfigure the Automatic Ground Environment. SAGE was computer’s course of action instantaneously a project to create a US nation-wide network or in real time. This elaborate input-output of digital computers linked to radar and system was part of an effort sponsored by the anti-aircraft weapons to track and intercept US Air Force towards improved air defense. enemy aircraft automatically in real time. The The data displayed on the scope was radar SAGE production machine was contracted data, or data generated from the reflections to IBM, in fact, and it was through working of radiated radio waves as they are deflected with Whirlwind engineers that IBM was back after hitting an object or target, and the able to incorporate innovations such as core Charactron CRTs used were able to show memory in their 704, although they did not

Figure 8. Operator uses special light gun to target potential intercept coordi- nates, circa 1959.

526 regularly incorporate the specialized input- computer with 3,600 transistors, and tests on output devices such the light gun into their the TX-0 quickly showed that transistors had standardized mainframe line. 27 several advantages over vacuum tubes, such The specialized console features allowed as greater durability, less power consumption, Douglas Ross to use Whirlwind to develop and longer life expectancy. In 1957 a 10-inch SLURP (Servo Lab Utility Routine Programs) electrostatic deflection CRT was installed in 1956, ”the first completely interactive with a 7x7 inch display screen and 512x512 software development environment, with spots that could be addressed, designed by many programming, language development, Ben Gurley; the year after a solid state (rather debugging, and display tools.”28 In other than vacuum tube) version of the light pen by words, it was the first interactive system Wesley Clark was added,.30 for a computer engineer to solve the design The TX-0’s combination of hardware and problems of the computer itself. The light gun peripherals enabled several novel programs was adapted into a more streamlined light in which users could engage interactively pen by Wesley Clark in the context of Lincoln’s using graphical or non-alphanumeric inputs Figure 9. TX-0 Console. next computer, the TX-0. and outputs. However, a problem of design is best addressed by the description of the the Flow Chart Display program--like SLURP a means of solving a problem of the computer itself--given by John T. Gilmore, one of TX-0’s programmers:

The programmer would produce an acetate transparency of his flow chart and place it on the display scope. Then, as each block was activated it glowed beneath the chart for some arbitrary time before moving on to the next block or stopping at an open switch for examination. Like Whirlwind, TX-0 or Zeroth The programmer had real time Transistorized Computer, was a parallel digital control of logical switches computer with magnetic core memory. It was within the flow chart and could developed under Ken Olsen at Lincoln Lab, control the speed of the cyclic with 64 K (thousand bits) of memory, a large process and stop at open capacity for the late 1950s, with 18 bit words switches to examine parameters (Figure 9).29 It initially functioned as a test and data.31

527 ALISE UPITIS

slowly than the central processor. In addition, In 1958 the TX-0, although striped down different I/O equipment operated at different to 4K of its memory, was relocated to the speeds. In response IBM introduced a second floor of Building 26 above the MIT Data Synchronizer for the 709 computer.35 Computation Center which then housed It contained two I/O channels that would the IBM 704. Unlike the IBM, the user was help with this disparity of operating speeds, also the computer operator: a user could but a channel is actually a small computer. personally program, prepare and input tape, Adding another computer to the central and operate the computer--24 hours daily.32 processor added to the computer’s cost and Flow Chart Display was a facet in this vision, complexity. The designers of the TX-2 chose supporting “an interactive situation where to bypass the use of channels to allow rapid programmers actually worked at a console I/O directly from device to core memory. and made changes right there on the spot.”33 This was achieved through a minimum The TX-2, developed in 1958 and for which buffering scheme with direct transfers to TX-0 really was a testbed, contained 22,000 memory, which means temporary data transistors and word length was doubled to storage was used to help compensate for 36 bits. Instruction size (word addressing) differential speeds of I/O equipment and was 18-bit so it had an address space of 218 central processor (see Figure 9).36 Such 36-bit words or approximately one megabyte direct memory access (DMA) subsequently of storage—sort of an ocean of memory defined the architecture of the microcomputer for 1958. Program control determined the and is built into microprocessors used in specific form of machine operating, from a full contemporary personal computers. This 36-bit computer to four 9-bit computers. This choice was also a great cost saver. As variability allowed more parallel operations creators of TX-2’s commercial version, Digital and improved efficiency in memory usage Equipment Corporation’s PDP-1 by, once since the TX-2 could store data with different remarked: “A single IBM channel was more word lengths. The TX-2 it had 32 instruction expensive than a PDP-1.”37 The PDP-1 originally counters that could be assigned to different sold for $120,000.38 computer users who would then compete Reduced cost and complexity for I/O also for operating time instruction to instruction.34 meant users could readily adapt the TX-2 to It was an innovation in time sharing, before their particular programming needs. In the the Compatible Time Sharing System (CTSS) early 1960s several individuals mobilized began operating on an IBM 7094 at the MIT Lincoln’s TX-2 towards endeavors in the new Computation Center. Computer-Aided Design project, but in the One of the reasons that input and output interest of space I will focus on one such (I/O) equipment was often limited on IBMs graduate student. Ivan Sutherland mobilized in the 1950s and early 1960s is that this Lincoln’s TX-2 in his doctoral research as a equipment tended to operate much more candidate in electrical engineering at MIT. He

528 titled the dissertation “Sketchpad, A Man- This is how Sutherland Machine Graphical Communication System,” summarized its workings: and two fields that benefited were interactive graphics and computer-aided design. A Sketchpad user sketches directly

Figure 10. Ivan Sutherland at TX-2 display. From Ivan Sutherland, “Sketchpad: A Man-Machine Graphical Communication System,” 329

Not only was Sketchpad the first on a computer display with a “light computer program where a user could pen.” The light pen is used both to create and copy shapes (in the form position parts of the drawing on of straight line segments and circular the display and to point to them arcs and their groupings), but present- to change them. A set of push day graphics programs also utilize buttons controls the changes to be several other functionalities pioneered made such as “erase,” or “move.” through Sketchpad, including: snap to Except for legends, no written end point, definition copy, geometric language is used.39 constraints, and the ability to stretch or reposition shapes while a user In other words, everything in Sketchpad observes the operation. was executed without a mouse or keyboard,

529 ALISE UPITIS

and the TX-2 certainly was not a desktop additional computer resources, such as computer. It not as large as Whirlwind’s the ability of the light pen to function “as a occupation of 2,500 square feet, but a still sort of analog computer to remove from consider floor space had to be allocated to it consideration all but a very few picture at Lincoln Lab. Sutherland adapted a console parts which happen to fall within its field or terminal for the TX-2 containing a CRT of view, thus saving considerable program display, light pen, and push buttons as well time”.41 as four knobs to rotate and magnify shapes The TX-2 had multiple sequence design, and a panel of toggle switches, controlling which allowed the display system to operate the display of various sorts of drawing-related at an independent speed while the remainder information. It was the TX-2’s flexible I/O of the computer could be executing other capacities that permitted such customizations operations. The CRT displayed dots in the and Sketchpad was indeed a system, as form of ten bits per axis. An individual image Sutherland states, not just computer code had about 3,000 dots. This is because flicker, (Figure 10). which occurred at 30 frames or fewer per Sketchpad’s light pen was a photocell second, could be avoided when dots were mounted in housing, about the dimensions of displayed 100,000 per second. Each dot a fountain pen, which connected to the TX-2 displayed was stored in a display table in TX- by a small coaxial cable. It had an amplifier 2’s main memory in the form of 36-bit words, sensitive to the first flash of a dot on a display as the TX-2 itself was a 36-bit computer. This system. As Sutherland summarizes its use: is in addition to a data table that contained information on the graphical elements. 20 If we point the light pen bits defined the coordinates of the spot on at the display system and the display. The program could identify which press a button called “draw,” spot in the display table was being identified the computer will construct a by the light pen being used on the screen. straight line segment which In turn the display table pointed back to the stretches like a rubber band data table that contained the information from the initial to the present on the graphical element the light pen was Figure 11. From Douglas Ross, “A generalized location of the pen…To close pointing to, so 16 bits were used to display technique for symbol ma- the figure we return the light the appropriate spot on the CRT by pointing to nipulation and numeri-cal pen to near the end of the the data table element that in turn activated a calculation,” Communica- tions of the ACM, Volume first line drawn, where it will point to be added to the display table. 4 Issue 3 (March 1961), “lock on” to the end exactly. A This clever use of allocating tasks 147. sudden flick of the pen terminates between the computer and specialized display drawing.40 hardware was indebted to the work on data structures by Douglas T. Ross, then active Independent peripherals could then provide in the CAD project. During the 1950s, Ross

530 Figure 12. TX-2 block led the MIT Servomechanism Laboratory’s diagram. From John M. Computer Application Group in developing Frankovich and H. Philip Peterson, “A Functional the Automatically Programmed Tool Language Description of the Lincoln (APT) for programming machine tools TX-2 Computer,” Proceed- ings of the Western Joint controlled by computers. Ross’s motivation Computer Con-ference, for investigating data structure practices in 1957 (New York: ACM, 1957): 147. programming was that tree structures or list structures—structures then being deployed by pioneers in artificial intelligence such as John McCarthy and Herbert Simon— are composed of elements with only two components, and hence the characteristic branches of a tree structure. Yet any element that is more accurately described with more than two components must nonetheless in list-structured computer languages be described as a tree of elements with a number of branching points. Each additional branching point represents a computer word, consuming storage and computer time. His innovation was that for programming to another element, also with an arbitrary purposes elements are often better described number of components.”42Ross gives a as a “plex” containing an arbitrary number wonderful illustration of the contrast between of components each of which can be, for list structure and plex (Figure 11). example, machine instructions, symbolic Sutherland specialized this insight of data, or machine addresses, or “a component Ross’s to create a “ring structure” or data of an element may be a ‘link’ or reference structure in which each element contained a reference to the next record with the last record pointing to the first. Sketchpad was then able to, for example, insert new or remove elements in a displayed drawing and merge different files of picture data. Furthermore, because “the large storage capacity of the TX-2 did not force storage conservation,” Sutherland programmed the ring structure to contain some redundancies thereby permitting more rapid program execution (Figure 12).43

531 ALISE UPITIS

tool for augmentation? Simulation/Augmentation Plans seeks to explain how a system-- computer, mechanical, or human--executes In 1960 psychologist George Miller co- “Plans” or goal-directed actions and the founded the Harvard Center for Cognitive structure that underlies this behavior. The Studies (HCCS) with psychologist Jerome authors propose that a Plan is realized by Bruner. Published in 1960, Miller, psychologist means of a hierarchy of sub-plans that take Eugene Galanter, and neurophysiologist the form of feedback loops. A sub-plan is Karl Pribram co-authored Plans and the referred to as a TOTE unit, acronym for Test- Structure of Behavior.44 Plans has been Operate-Test-Exit, and a Plan requires a seen as “paradigm-shifting” for American hierarchy of these units. In executing a plan, a psychology, signifying a rupture from system selects an operation to be performed. stimulus-response models and a behaviorist It then tests the result of that operation tradition.45 Bruner in turn sought support from against a goal state. If the goal state is not Harvard’s administration for Herbert Simon’s met, it mobilizes a feedback loop to repeat appointment to the HCCS--Allen Newell, the operation until the goal state is met, at J.C. Shaw, and Herbert Simon provided a which point the system transmits control to cornerstone of artificial intelligence through the next level in the hierarchy. One of the their attempts to simulate human thought on

digital computers. Bruner wrote: “Simon’s Figure 13. TOTE unit presence here would almost automatically hierarchy for hammer- facilitate some of the kinds of developments ing a nail. From Miller, Galanter, and Pribram, we have been talking about for several years Plans, 24. now” and continuing “we need him now more than ever.”46 Alexander was affiliated with the HCCS from 1960 to 1962, assisting Miller and Bruner in teaching and aiding with psychological experiments on how individuals parse, process and recall patterns.47 Alexander mobilizes Plans and the work of Newell, Shaw, and Simon a number of times in Notes and elsewhere.48 Between and beyond a hierarchical program process and the hierarchical tree that is sought as the hierarchical program’s result and design method to be followed the question I seek to ask is: is the design method a simulation, or a

532 Figure 14. Block diagram book’s central illustrations directly, although of BLDUP. From Alexan- rudimentarily, engages the process of design der, HIDECS 3, 12. through the example of hammering a nail until it is flush with a surface (Figure 13).49 The concept of behavior that Plans proposes is remarkably similar to the operations of Alexander’s computer programs (Figure 14). Its authors in fact readily acknowledge that Plans treats behavior as the execution of a computer program.50 They also propose that the “structure underlying behavior” can be described in terms of a computer program—specifically, the Information Processing Language (IPL) series developed by Newell, Shaw, and Simon.51 This rings with Alexander’s claim that the planar graph of verticies and links—in turn to be transformed into a hierarchical tree- -represents “the structure of the design problem” 52 Yet in something with a sufficient level of complexity, it is simpler to describe the structure by building it than describe its behavior. Alexander and Simon et all describe human behaviors mathematically and build a structure accordingly, that of Sutherland, Coons and others is to create a physical model and then compares it with observation by mapping. Newell and Shaw programmed what is widely considered the first program for artificial intelligence , the attempt to simulate human thought on a computer, in 1956, the Logic Theorist (LT), by using IPL-II on the Johnniac computer built at the RAND Corporation in Santa Monica, California.53 The Johnniac, named after computational theorist John von Neumann, was a stored- program, parallel-processing computer using

533 ALISE UPITIS

vacuum tubes with 40 bit words, 4096 words through novel display and peripheral of primary magnetic core storage, and 9216 hardware.56 words of secondary storage on magnetic As Coons explains: drums.54 LT was used for proving theorems in symbolic logic, to simulate. It was built from It is typical of the design process data structured according to lists--ordered that…iterations—from concept, sets of information that contain elements through analysis, evaluation of (variables and symbols) or further lists of the analysis, decision to modify elements--and LT commonly required 13 or 14 the concept, and finally to a new sub-routine hierarchies that together produced concept—form loops that are the IPL-II program. LT’s creators asserted traversed again and again, until that human problem solving is a hierarchical eventually the designer judges process, while at the same time LT’s sub- the design adequate to satisfy routine hierarchization was to some extent some scale or scales of value 57 required since it allowed the program to be judgments. Figure 15. compressed to a size more manageable by General list structure for Logic Theorist. From Al- 55 the Johnniac’s memory (see Figure 15). The language Coons mobilizes is len Newell and J. C. Shaw, Conceptually LT operated like the enmeshed in a discourse that drew from Programming the Logic Theory Machine, P-954 Plans of Miller, Galanter, and Pribram, or the discrete operational sequences of (Santa Monica: The RAND Alexander’s programs: as a hierarchy of computer processing and the iterations Corporation, 1957) 13b subroutines that, once begun, would run its course moving through its predetermined tree structure towards a predetermined goal. These subroutines could “branch”, or switch between different subroutines at certain points, but again this is a property of a tree structure. For the programs of Alexander and Simon the operations were batched, at least conceptually, for even if more than one program could be run during a given computer session there was no option for a human intervene with the program’s operations in real time. Steven Coons—who was made to speak in the introduction and on Sutherland’s dissertation committee—proposed an idealized computer system that can respond to and represent graphics and calculations

534 of programming’s feedback loops. These effortlessly with technological operations processes were widely represented through while Douglas Engelbart—best remembered the block diagrams and flowcharts regularly for inventing the computer mouse-- sought used by engineers to illustrate the operation an “integrated domain where hunches, cut- of a program or the structural flow of a and-try, intangibles, and the human ‘feel for computer hardware. 58 While discursively a situation’ usefully co-exist with powerful similar, the hardware is different, as when concepts, streamlined terminology and Sutherland asserts Sketchpad “opens up a notation, sophisticated methods, and high- new area of man-machine communication” powered electronic aids.”63 The innovators through the system’s ability “to converse in interactive graphics discussed above rapidly through the medium of line sought to develop computer systems that drawings.”59 Likewise when Ross states his would contribute to human efforts as a preference for man-computer “conversation” collaborating partner or “second self,” to use over “communication.” 60ing the expression of Sherry Turkle’s. Alexander’s Interchanges between humans and view of designing is remarkably different from machines being envisioned at MIT by many involved in early computer graphics at Sutherland and others, were closely aligned MIT. Requirements for Alexander take the with the image of “cooperative interaction form of computer data, minimizing the conflict between men and electronic computers” between them becomes the computer’s articulated by J.C.R. Licklider, a psychologist instructions to execute. His computational and mathematician who had contributed system does not work as a symbiotic to interface design on the SAGE project at system or cooperative coupling but as a Lincoln Lab.61 Licklider’s “Man-Computer means of automating the execution of logical Symbiosis” from 1960 opposes batch operations. Alexander executes an inverion of processing and procedural languages such as ideal and imitation, and the process of design FORTRAN to a cooperative coupling between is taken to imitate the mdoel by which he tries human and computer built largely on time- to imitate it. sharing and interactive graphics. “Certainly,” Licklider asserts “for effective man-computer interaction, it will be necessary for the man and the computer to draw graphs and pictures and to write notes and equations to each other on the same display surface.62 These speculative systems embodied technologies to seamlessly accommodate the more intuitive aspects. For example, Licklider believed contributions of “man’s intuitive judgment” will some day blend

535