Individualized computers: intellectual or clerical?

Early work in interactive computing took place within military-funded research laboratories during the 1960s. There, engineers designed machines to be used by people like themselves: autonomous, technical workers. Whereas existing computers were typically operated by clerical workers, these engineers reimagined computers as tools for intellectual work. J.C.R. Licklider and Douglas Engelbart, researchers who heavily influenced future generations of HCI professionals, believed that computers should perform routine, clerical work in order to empower intellectual workers.

Historians and HCI practitioners often remember Licklider, a psychologist and research funding agency director, as a progenitor of HCI.Footnote ⁸ According to this narrative, his canonical paper on ‘man–computer symbiosis’ presented a research programme that was distinct from AI. In the paper, Licklider explained that ‘it seems worthwhile to avoid argument with (other) enthusiasts for artificial intelligence by conceding dominance in the distant future of cerebration to machines alone’. In the meantime, he argued, computers should exist in ‘symbiotic’ relationships with ‘men’.Footnote ⁹

However, scholars have paid less attention to a practical problem that motivated Licklider's paper. He was exhausted with paperwork. Lacking something ‘comparable to a time-and-motion-study analysis of the mental work of a person engaged in a scientific or technical enterprise’, Licklider performed one on himself. Licklider found that ‘my “thinking” time was devoted mainly to activities that were essentially clerical or mechanical … my choices of what to attempt were determined to an embarrassingly great extent by considerations of clerical feasibility, not intellectual capability’. The busy scientist and director complained that 85 per cent of his ‘“thinking” time’ was devoted not to ‘thinking’ but ‘getting into a position to think’.Footnote ¹⁰ The solution he proposed: ‘Computing machines will do the routinizable work’ while ‘men will set the goals, formulate the hypotheses, determine the criteria, and perform the evaluations’.Footnote ¹¹

Licklider's distinction between clerical and intellectual work resonates with longer patterns in the history of computing. Lorraine Daston has demonstrated how calculation from the nineteenth century onwards went from being seen as an intelligent skill to the mindless activity of ‘mechanical’ workers after the mathematician Gaspard de Prony applied Adam Smith's division of labour to calculation projects in revolutionary France.Footnote ¹² Simon Schaffer has shown that debates around the best ways to measure and mechanize mental labour in the late nineteenth century corresponded with the emergence of English intellectuals as a ‘specific class formation’. This group distinguished routine mental work from their own ‘discretionary’ intelligence.Footnote ¹³ A division of labour between mathematicians, engineers, bureaucrats and scientists on the one hand and clerks, (human) computers, and computer operators on the other can be seen in almost all computing work through the mid-twentieth century.Footnote ¹⁴

In 1962, Licklider brought his vision of computing to the Information Processing Techniques Office (IPTO), the wing of the US Department of Defense's Advanced Research Projects Agency that would become the major funding agency for computer science research through the 1960s.Footnote ¹⁵ Under Licklider's inaugural direction, the IPTO also began supporting the research of Stanford Research Institute (SRI) engineer Douglas Engelbart in 1963.

Like Licklider, Engelbart thought that intellectual work could be improved through engineering, and that computer systems were the best means to accomplish this end.Footnote ¹⁶ Engelbart dedicated much of his research at SRI in the 1960s to improving the interface between users and computers, famously co-inventing the computer mouse with Bill English. Sociologist Thierry Bardini's monograph on Engelbart's career explains how he and his colleagues aimed to build systems for people like themselves, people Engelbart called ‘knowledge workers’.Footnote ¹⁷ Their image of the user was a reflexive and idealized version of themselves. They deemed knowledge workers autonomous and creative – willing and able to sink time into learning complicated input systems. Engelbart shared with Licklider the view that computing machines should be subordinated to knowledge workers, inviting others to imagine the computer as ‘a completely attentive, very patient, very fast symbol-manipulating slave’.Footnote ¹⁸

When the office machinery company Xerox established PARC in 1970, many of the research staff it hired came directly from Engelbart's lab.Footnote ¹⁹ The team's new research environment, however, brought changes. Xerox aimed to develop a large market for personal computers as generic office machinery, an essential component of the company's ‘office of the future’. As a result, their user was no longer a knowledge worker, but a ‘naïve user’. As Bill English remembered it,

Where the user imagined by Engelbart's lab was a specialized and autonomous knowledge worker, at PARC the user was ‘nontechnical’, even ‘naïve’. No longer an engineer, she was a secretary.Footnote ²¹

The secretarial user

While at defence-funded laboratories like SRI, computer designers and engineers often centred scientific research and military command and control. In contrast, when these designers moved to the research and development wings of office machinery companies in the 1970s, they placed text-editing at the centre of their research programme. Secretaries – quintessential text editors – became generic ‘users’ in PARC research in the early 1970s. AIP researchers studied the ‘manuscript-editing task’ and used data from secretarial subjects to create their first model of the user. They applied Simon and Newell's information-processing psychology to formulate this model. However, tensions emerged between the subjects of AIP's research and the ‘user’ that these subjects represented.

PARC established AIP with the aim of developing commercial applications for Herbert Simon and Allen Newell's information-processing psychology.Footnote ²² Newell, who was then a consultant for PARC, proposed the research group in January 1971, one year after PARC's founding. Newell explained that the emerging ‘psychology of cognitive behavior’ developed by Simon and himself could be applied to the design of office technology at PARC with ‘substantial payoff (in dollars)’.Footnote ²³ While they are often remembered as progenitors of AI, Simon and Newell initially resisted identifying their work on the Logic Theory Machine as ‘artificial intelligence’. However, by the late 1950s they accepted this description due to its growing rhetorical power.Footnote ²⁴ While Simon and Newell's work in AI and AIP's proposed research were based on information-processing psychology, Newell thought that AIP could repair a gap in the information-processing psychology research programme that AI had neglected. Newell regretted that his graduate students who were ‘deeply interested in artificial intelligence’ obtained no knowledge of cognitive psychology.Footnote ²⁵ He hoped that AIP could amend that lack of interest and bring cognitive psychology into computer science.

Newell's idea for an applied-psychology unit came to fruition after he brought his doctoral students Thomas Moran and Stuart Card to PARC in 1974. The two Carnegie Mellon University students would conduct research towards their respective dissertations in computer science and psychology at PARC under Newell's direction. PARC established AIP that year. Newell served the project as a consultant for about three to four days each month.Footnote ²⁶ As he had envisioned, the project aimed to create an information-processing model of the personal computer user that designers could refer to without having to conduct user trials. They believed this would save PARC time and money as it prototyped new machinery.

AIP researchers did not initially see their work as explicitly drawing from artificial intelligence. Moran mentioned in 1975 that the name ‘AIP’ was selected as a ‘pun’, which may indicate a perceived association with AI.Footnote ²⁷ However, as with Simon and Newell's other work, AIP retrospectively applied this terminology. At a January 1986 Association for Computing Machinery conference on the History of Personal Workstations, Card and Moran explained that the idea for AIP ‘was to draw concepts from cognitive psychology and artificial intelligence to create an applied cognitive science of the user’.Footnote ²⁸

AIP and PARC agreed that the psychology of the ‘editing task’ would be the project's first area of study.Footnote ²⁹ The ‘editing task’ encompassed a common relationship between a secretary and her employer. The executive would dictate to the secretary, who would type on a typewriter. She would then pass the copy to the executive, who would mark it up for revisions and return it to the secretary to retype. The process would continue until the pair had produced a finished draft. With Xerox's prototype machines, however, the secretary would not have to completely retype the draft with each revision but only edit the digital file to reflect the executive's corrections. With the time saved, the executive could assign his secretary to other tasks.

AIP researchers’ focus on secretaries reflected Xerox's commercial response to US businessmen's concerns in the 1960s and 1970s. Xerox wanted PARC to develop new machinery that would dominate their ‘office of the future’, just as their copier had for much of the company's history. Secretarial text editing was important in this vision. The labour process of document preparation, and the stenographer or secretary's place in it, had troubled management experts in the United States since the early twentieth century.Footnote ³⁰ Thomas Haigh explains that management experts at this time saw the traditional secretary as ‘the enemy of efficiency’.Footnote ³¹ In 1974, the American Management Association reported on secretaries that ‘it's very hard to establish procedures and controls over what she does’.Footnote ³² Many business leaders in the 1970s briefly believed that ‘word processing’ was the solution to this problem. While word processing came to refer to a type of computer software by the 1990s, in the 1970s it meant replacing ‘the one-man, one-secretary, one-typewriter idea’ with a factory-style production of texts.Footnote ³³ Word-processing centres would supervise a mass of typists using specialized equipment under one roof.

The word-processing vision of the early 1970s was short-lived. Management experts soon came to see personal computers as a more affordable and widely applicable alternative. Along with long-term declines in component costs and an increase in processor power leading into the 1980s, US corporate executives believed that more people could use an individualized computer to produce documents, including secretaries and middle-level managers. They identified this arrangement as ‘office automation’.Footnote ³⁴ PARC led the wider business community in embracing this office automation vision by about a decade. Haigh points out that while PARC's researchers were not management experts, the centre was funded on Xerox's expectation that individualized computerization would change the office – and sell their products.Footnote ³⁵ Within this context, PARC researchers began studying text editing.

Soon after AIP was established in 1974, the project ran experiments recording their colleagues editing a manuscript on the POET text editor, run on PARC's MAXC.Footnote ³⁶ They asked subjects to speak aloud to themselves to narrate their thought process and reactions to the machine. AIP adopted this technique from Newell and Simon's research in problem solving at the RAND Corporation in the mid-1950s.Footnote ³⁷ AIP transcribed data from the videos into five columns: time, the subject's verbal utterances, where the subject was looking, the subject's keystrokes, and other miscellaneous events and comments.Footnote ³⁸ AIP researchers codified these transcripts into sequences of perceptual, motor and mental operations such as ‘choose’, ‘describe’, ‘look-at’ and ‘say’ at a grain ranging from half a second to five seconds. By analysing computer use into operations using the methods of Simon and Newell's information-processing psychology, Card and Moran promised to predict the time to complete any computer task that could be described in such operations.

The boundaries of AIP's research resembled some prior PARC research on secretaries. In May and August of 1973, before AIP's founding, computer scientist Larry Tesler conducted two studies on secretarial text editing at PARC with PARC secretary Sylvia Adams.Footnote ³⁹ As Tesler explained, his research was guided by PARC's goal to ‘offer a text editor that can be learned quickly and used effectively by users of varying technical backgrounds, including secretaries’.Footnote ⁴⁰ However, secretarial text editing was different from other forms of text editing. It ‘involves transcription of another person's writing, as opposed to literary composition or computer program development’, and ‘usually is guided by copy marked up by another person’.Footnote ⁴¹ Like AIP, Tesler's experiment included the person editing the text and excluded the person who had dictated it or corrected it. Furthermore, like AIP, Tesler observed Adams working with PARC's prototype system. And finally, both series of experiments reduced secretarial work to its most routine aspect: typing a manuscript produced by someone else.Footnote ⁴² However, Tesler interviewed his subject about her evaluation of the machine. This differed from AIP's quantitative approach, which broke down secretaries’ work into discrete steps, resonating more with the methods of time and motion studies.

The affinity of AIP's approach with time and motion studies was a source of disagreement between Moran and Card. By mid-1975, Moran had become frustrated with the group's approach. In May, he wrote to AIP to express his ‘vague, but growing, discontent with the current drift of some of AIP's research and stated goals’.Footnote ⁴³ The preceding March, Card had proposed establishing a catalog of operators along the lines of a time and motion study.Footnote ⁴⁴ He shared scans of industrial-engineering literature that inspired him. This included some which analysed workers’ behaviour into ‘Therbligs’, eighteen elementary classifications of motion created in the early twentieth century by Lillian Gilbreth and Frank Gilbreth, followers of the founder of scientific management Frederick W. Taylor.Footnote ⁴⁵ Card remarked elsewhere, ‘A good all-around measure … is how long it takes to get the job done’.Footnote ⁴⁶ He sought to create models that optimized for performance time – other values like intellectual creativity or ease of use were not captured here.

Moran did not believe that this was the best approach for the project. He contrasted Card's ‘industrial engineering’ approach, which tried to classify all behaviour into operators to predict performance, with the ‘human factors’ approach, which tried to develop a psychological theory to inform design choices, rather than exhaustively predict all behaviour. ‘A basic tension in this Applied Psychology game’, Moran explained, was the problem of ‘being psychologists or being time-study hacks’.Footnote ⁴⁷ Moran called for Card's industrial-engineering approach to be ‘abandoned’. Furthermore, ‘AIP should temper its “megalomodelomania”: the compulsion to build a complete (in some sense) model of the user in some task domain’.Footnote ⁴⁸ Despite Moran's criticism, the construction of generic user models would remain a central goal for AIP through the early 1980s.

AIP's first two years of research culminated in the circulation of their first PARC report, ‘The manuscript editing task: a routine cognitive skill’, in December 1976. Where Newell and Simon's earlier work had focused on humans encountering novel problems, Card and Moran extended their information-processing psychology to encompass routine clerical work. This paper evaluated ten models of a manuscript-editing task, ‘differing with respect to generality and descriptive detail’.Footnote ⁴⁹ All the models described the manuscript-editing task in terms of goals, operators, methods and selection rules (GOMS), a technique that would continue to be widely used in HCI through the late 1990s. The trio recognized the ‘kinship’ of their work ‘with the time-and-motion studies done by industrial engineers’ but they claimed that it was different. Rather than setting performance standards, their research sought to predict performance based on a theory of ‘skilled behavior’.Footnote ⁵⁰

The experiment described in this report blurred the distinction between the woman clerical worker and the masculine user that would come to dominate Xerox's vision for its machines. AIP collected the data used to create and evaluate the models from an experiment whose only subject was a ‘highly skilled female secretary–typist’ referred to as Subject 13 (see Figure 1).Footnote ⁵¹ Subject 13 was given a manuscript to edit which took about twenty minutes. As with their previous experiments, AIP video-recorded Subject 13 and the terminal display and analysed her work into operations. AIP spoke of Subject 13, the basis for their models and the only subject of the main experiment, using she/her pronouns. However, they spoke of the generic subject in the outline of experimental procedure and the ‘user’ it represented exclusively using he/him pronouns.Footnote ⁵²

Figure 1. ‘Physical Layout of the MS-Editing Task’. Stuart K. Card, Thomas P. Moran and Allen Newell, ‘The manuscript editing task: a routine cognitive skill’, December 1976, Carnegie Mellon University Archives, Allen Newell Collection, Box 72, Folder 5095, p. 3. Republished with permission of Carnegie Mellon University Archives and the author.

More tensions between their research subject and the ‘user’ surfaced when Card, Moran and Newell submitted the ‘Manuscript editing task’ article to Cognitive Psychology in 1978. Referee B praised the article for extending Simon and Newell's ‘theory of problem-solving to real-world tasks’ but was hesitant to accept it. Referee B questioned the group's description of manuscript editing as routine since ‘much editing, even on computers, involves deciphering the author's handwriting which in itself is a (routine?) problem-solving task for secretaries’. Furthermore, Subject 13's typing speed was unrepresentative: she ‘typed at 103 words per minute … which has limited generalization to most mortals, even secretaries’.Footnote ⁵³ The journal's editor, Earl Hunt, said that their work was important and accepted the article with significant revisions. Nonetheless, the referees were challenging some of AIP's base assumptions. Like Licklider, AIP researchers assumed that clerical work was ‘routine’. Furthermore, their attempt to depict secretaries as generic users failed to consider their special affinity with text-editing technology, notably the QWERTY keyboard. The researchers who moved from Engelbart's laboratory assumed that secretaries were ‘naïve’ while AIP saw them as generic. But as the referees here emphasized, secretaries were too technologically adept to count among ‘naïve’ or generic users.

The paper was eventually published in 1980 after Card and Moran included computer scientists as subjects in addition to secretaries.Footnote ⁵⁴ They retitled the paper from ‘manuscript editing’ to ‘computer text editing’. This reframing presented computer text editing not as feminine clerical labour but as a generic and gender-neutral activity. This was significant. Although they were far less common in US offices, programmers – typically male – were the other key users of text-editing software. As Haigh points out, text-editing software for programmers was ‘the direct technological ancestor’ for typists’ word-processing software.Footnote ⁵⁵ However, the difference between these two applications was ‘more cultural than technical’.Footnote ⁵⁶ In this case, it had more to do with the perceived difference in gender and technical skills between programmers and secretaries than with technical differences in the software. AIP's new model of the text-editing user collapsed this difference and de-emphasized its secretarial origins. Although AIP researchers had used the secretarial user image in two other internal reports, they left it out of their publication.Footnote ⁵⁷ In the following years, a new user image took shape.

The user reconstructed

AIP's first decade of research resulted in the 1983 publication of The Psychology of Human–Computer Interaction, the first book to use the emerging field's name in its title.Footnote ⁵⁸ The book also announced the launch of the journal Human–Computer Interaction, which included Moran as editor with Licklider, Newell, Card and other researchers on the editorial board. As PARC developed the Alto and then the Star – a notoriously commercially unsuccessful personal computer – Xerox realized that the kind of computers PARC wanted to build remained priced well beyond what an executive could justify purchasing for his secretary. At the same time, many male executives were wary of using keyboards, instruments that they regarded as feminine.Footnote ⁵⁹ As political scientist Jeanette Hofmann reflected in the following decade, the Xerox Star possessed ‘the only word processing program … ever to have been developed for men’.Footnote ⁶⁰ Hofmann's provocative observation drew attention to the changing gender of manuscript production work in the office that followed the introduction of personal computers.

While the user began at PARC as a secretary in the early 1970s, its gender and position in the office hierarchy had changed by the 1980s. Card, Moran and Newell's book depicted the user as a bald person sitting in front of an Alto (see Figure 2).Footnote ⁶¹ Even disregarding the white, cartoonishly ‘neutral’ human figure, the masculinity of this image compared to the diagram of Subject 13 is apparent. The then-ubiquitous QWERTY keyboard was missing from this image, which is striking given the importance of keystrokes to AIP's models.Footnote ⁶² This absence is also conspicuous given that the keyboard typically fit snugly between the legs of the display's stand (compare to Figure 1). The secretary reappeared in the book's ‘keystroke-level model’ in a negative sense: Card, Moran and Newell disaggregated the ‘average non-secretary typist’ in the crucial parameter of typing speed.Footnote ⁶³

Figure 2. ‘The human–computer interface’. Stuart K. Card, Thomas P. Moran and Allen Newell, The Psychology of Human–Computer Interaction, Hillsdale, NJ: L. Erlbaum Associates, 1983, p. 5. Republished with permission of the author.

In this short period of time, the group of researchers who would help create the field of HCI realigned their focus to fit a new arrangement of office work. During the 1960s, Licklider and Engelbart imagined users to be intellectual and technically skilled workers like themselves. When the researchers from Engelbart's lab arrived at PARC in the early 1970s committed to building (and selling) machines usable by non-engineers, they saw the secretary as a typical user. Secretarial subjects became the focus of Card and Moran's early research on text editing, sources of data about computer use that AIP used to construct models. However, as the personal computer came to be seen less as a clerical machine and more as a managerial tool, secretaries appeared less typical.Footnote ⁶⁴ They were also too technical to be considered ‘naïve users’. Card, Moran and Newell's book simultaneously finalized and obscured the transition they had undergone. As they explained, the user was explicitly not a secretary:

The replacement of ‘operation’ with ‘communication’ here mimicked the language used by Licklider and Engelbart. ‘Communication’ also evoked a relationship that was familiar to managers: dictation to a secretary.

At the same time, however, the computing tools that proliferated office work in the 1980s diverged strongly from the visions of these earlier engineers. Licklider wanted to free intellectual workers like himself from clerical labour. But with personal computers, managers found themselves doing more text-editing work, rather than less.Footnote ⁶⁶ Personal computers became something that was remarkably rare in the history of industrial capitalism: they became machinery for managers.Footnote ⁶⁷ HCI explained and accommodated this new arrangement by affirming the vision of computers as ‘augmenting the intellect’ while – in contrast to Engelbart – designing computers to be usable by non-technical managers.

Card, Moran and Newell's work was not uncontroversial. In December of 1981, before the book's publication, three computer scientists wrote with concern to Association for Computing Machinery Computing Surveys, questioning whether AIP's goal of ‘augmentation’ would result in ‘dehumanization’. They accused Moran, who had edited a previous Surveys special issue on psychology, of ‘controlling human behavior in order to subordinate it to the purpose of computer system performance irrespective of the goals that people wish to achieve with the help of the system’.Footnote ⁶⁸ While sidestepping the criticism, Moran acknowledged some ambivalence about the political ramifications of his research: ‘Scientific power does not guarantee good taste. A scientifically powerful user psychology could be used against, as well as for, the user’.Footnote ⁶⁹ Despite their attempt at a reorientation, the residue of industrial engineering was difficult to remove.