ETAG a Formal Model of Competence Knowledge for User Interface Design

This book is about a particular solution to the problem of how to design usable computer systems. Apart from this chapter, it may be read as a short history of the research efforts with respect to ETAG, which is the main subject of the book as well as the tool that is proposed to help create usable computer systems. Since answers are meaningless without the questions they address, this chapter discusses the questions underlying this research and, perhaps even more important, the reasons for asking these particular questions: the theoretical context which gave rise to them. This is especially important since a generally accepted approach to study and design in Human-Computer Interaction (HCI) is lacking. Moreover, the field might rather be characterised as a loosely organised collection of competing and often incompatible approaches. First, the chapter provides a general account of the subject matter of cognitive ergonomics, stressing the 'cognitive' part of it, and distinguishing it from related fields of research. Thereafter, the chapter focuses on creating usable computer systems as the primary aim of cognitive ergonomics, and it explains why this is not an easy task. Finally, the chapter defines cognitive ergonomics more precisely, as a science and engineering trade, that is concerned with user interfaces, and more specifically, their design. Guided by these three main characteristics of cognitive ergonomics, the major questions put forward in this thesis are derived and their background is discussed. 1.1 Cognitive Ergonomics Cognitive ergonomics is the study of human behaviour that is mediated by cognitive tools and devices. Cognitive tools are natural or artificial tools which require and determine the human ability to process information. The purpose of cognitive ergonomics is to adapt such cognitive tools and their usage so as to improve human information processing in terms of improved efficiency, fewer errors and accidents, and increased well-being. To this end it is necessary that cognitive ergonomics is able to analyse how such tools are used, to synthesise recommendations for design, and to evaluate uses and recommendations. An important area in cognitive ergonomics is the study of Human-Computer Interaction. Cognitive ergonomics is based on the premise that human interaction with computer devices is essentially a matter of knowledge representation and information processing, or: cognitive behaviour. Interacting with a computer system takes place by means of physical interaction: pressing buttons on a keyboard changes the physical state of the computer system which is fed back to 2 Chapter 1: Cognitive Ergonomics and UI Design the user by means of intensity changes of the light on the display unit. This description may be accurate, but it is as irrelevant as describing driving a car in terms of opening valves and pulling cables. It would rather be more sensible to describe interacting with a computer system in terms of writing a book with a word processor or calculating turnovers using a spreadsheet. To acquire 'common' human goals like writing a thesis or calculating the day's turnovers by means of a computer (or, for that matter any other tool, like a typewriter, pencil and paper, etc.), it is necessary to recursively subdivide and translate goals into commands for the word processor or spreadsheet. Since computer systems are not too sophisticated with respect to supporting human goals, the reformulation and translation processes of human goals and the interpretation of results of command invocations are necessarily human cognitive tasks. In cognitive ergonomics, interacting with a computer is assumed to involve different stages of human behaviour, such as formulating an intention and executing an action, and different levels of activity, such as the intention to improve a text and the intention behind physically executing a particular command (Norman, 1984). In cognitive ergonomic terms, users have to apply their knowledge about the computer system, in the form of a mental representation, to find the difference between a current state of affairs and a goal state within the user's task domain. Once established, the user has to devise a plan to diminish the difference, and reformulate the plan into the commands and command arguments of the computer system used. After issuing the commands, again, the user has to apply the knowledge about the system to transform feedback data from the computer into meaningful information about the success or failure of reaching the goal state within the task domain. As an example, consider making a paper copy of a report. In a paper office, one would acquire the report from its file, bring it to the copy machine to make a copy, and try not to forget to store the archive copy back in its place. In a 'paperless' office, it is necessary to know that reports are stored as computer files and that, in order to make a paper copy of a file, it has to be sent or copied to a printer device. The instruction how to copy a file varies between computer systems: dragging an icon representation of the report to a printer icon, typing in the name of the print command and the filename, etc. When the printer is not within visual or auditory reach, additional command specifications may be necessary to acquire information about the progress of the print command. Since computer usage primarily involves acquiring, transforming and applying (human) knowledge it may be clear that the basic thesis or central premise of cognitive ergonomics states that investigations should focus on cognitive factors in order to improve HumanComputer Interaction. The focus on cognitive factors distinguishes cognitive ergonomics from traditional or 'classical' ergonomics. As the name suggests, cognitive ergonomics may be seen as a mere branch of ergonomics, with which it shares the goal of facilitating human performance through adaptation of the tools to human characteristics and preferences. On the other hand, cognitive ergonomics differs from 'classical' ergonomics in that the focus is not so much on externally measurable quantities, such as movements, forces and body measures, but on A Formal Model for User Interface Design 3 psychological phenomena, such as knowledge, perception, and planning; phenomena that, generally, allow for indirect measurement only. Cognitive ergonomics is also closely related to cognitive psychology, in that both investigate a mental phenomena. Whereas cognitive psychology often uses computers to study human mental phenomena for the sake of acquiring general theories about mental behaviour, cognitive ergonomics studies mental phenomena and applies theoretical knowledge in order to solve the practical problems related to using computers. The abstract nature of cognitive psychological knowledge generally precludes applying it to practical problems without first making additional assumptions that may undermine the validity of solutions. Cognitive ergonomic theories are more directly connected to the domain of application. Finally, cognitive ergonomics is related to computer science in that both study the use of computer systems, but, whereas computer science investigates the technical requirements for using computers, cognitive ergonomics studies the human and cognitive requirements for doing so. Whereas cognitive ergonomics is related to general ergonomics and cognitive psychology via the subject matter of the investigations, computer science and cognitive ergonomics may be seen as mutual clients, where cognitive ergonomics is responsible for the design of the user interface (the user machine) that sets cognitive constraints, and computer science is responsible for the design and creation of the software (the soft machine) which sets and implements technical constraints for the overall design of computer systems. The fact that computer science and cognitive ergonomics are clients of each other does not express anything about their relative importance. Although there is a growing awareness that computer system design should include usability aspects, at present, especially in software engineering practice, the technical constraints are still predominant. Cognitive ergonomics and ergonomics in general are still regarded as additional to technical programming skills, rather than the opposite. Later on, it will be argued that, from a human task performance point of view, a far more important, and perhaps a leading role, should be assigned to cognitive ergonomics. At the start of this section, cognitive ergonomics was defined as the study to improve the use of cognitive tools in terms of efficiency, errors and accidents, and well-being. It may be possible to be more precise about the field and purpose of cognitive ergonomics. For example, humanitarians might insist that well-being refers to all human beings involved and exclude weapons, technocrats might want to focus on work systems and exclude enabling tools, and rationalists might want to restrict attention to purposive tools and exclude toys and pleasure. These statements may be caricatures but they do exemplify the risk of losing generality by playing the language game to strictly. Despite that cognitive ergonomics is not served by a priori excluding application areas and a more precise definition will not be provided, the work that is discussed in this thesis is restricted to the purposive usage of tools by means of discrete task performance. Computer systems are most commonly used to support work systems which makes it an obvious choice in HCI to focus on users who perform tasks for the purpose of acquiring specific goals and, because of the way to issue commands to computer systems, to focus on performing discrete tasks. 4 Chapter 1: Cognitive Ergonomics and UI Design Discrete tasks put specific requirements on cognitive processes such as perception, memory and attention which may differ considerably from the requirements of the tasks in different areas of cognitive ergonomics such as steering vehicles or controlling industrial processes. As a consequence, tasks have a rather universal status within HCI but in other areas of cognitive ergonomics aspects like the structure and presentation of information, and learning and skilled task performance may be more important. The remainder of this chapter discusses the design of usable computer systems as one of the main subjects of cognitive ergonomics and what should be done in order to develop cognitive ergonomics into the science and the engineering practice of user interface design. Arguing that user interfaces should primarily be seen as the user's means to perform tasks, the user interface is defined as the knowledge that users need to successfully perform tasks with a computer system (Moran, 1981; van der Veer, 1990). In combination with the argument that cognitive ergonomics should focus on the development of theories to capture its scientific knowledge and methods to use this knowledge to solve practical design problems, the two main challenges of this thesis are formulated as: • the selection and development of a good representation for user interfaces • the development of a user interface design method based on this representation 1.2 Usable Computer Systems Designing usable computer systems is a major concern for cognitive ergonomics. In this paragraph it will be argued that regardless of the general difficulties to design and to design computer systems, designing usable computer systems is a more difficult task due to the combination of the technology which, at least, in principle does not impose restrictions on the design and the psychology of the human task performance which is largely unknown and hidden from direct inspection. Being concerned with the design of usable artefacts is not specific to cognitive ergonomics. For example, for centuries people have been designing usable bridges, at first, using trial and error, and gradually shifting to genuine design methods that carefully consider all known relevant variables (e.g. Petrosky, 1996). What is new in HCI design is that computers are general purpose information processors and able to support human task performance in many different ways, which may not be obvious, or even compatible with each other. As such, the number of variables to consider is large, and when their interactions are included, the number becomes huge, and when considering that there are no such 'obvious' criteria for evaluating HCI designs as there are for designing bridges, the number becomes endless. In mathematical terms, for each problem that can be solved computationally there is an infinite number of possible solutions and, generally, a large number of feasible ones. In terms of computational support for human tasks, concern is not with solving single, isolated problems, but with interconnected problems. As such, when a software solution is found to solve a particular task performance problem, it may create new problems, and make yet other problems easier or harder to solve. In addition, when working on a problem, it may remain unknown for some time how particular solutions will influence solving other problems. A Formal Model for User Interface Design 5 For example, an undo facility is a means to solve the problem that users may perform unintended actions, but this solution may create synchronisation problems when data is shared among users, and it may create the problem of what to do with accidental use of the undo facility. In this respect, design is largely a matter of solving trade-offs between requirements, between solutions and between requirements and solutions (Norman, 1986). The individual problem solving aspects in supporting human task performance are not unique to HCI. Most trades that involve designing things for human use have to deal with discovering the unknown and trading-off requirements, partial solutions and interactions between them, and have to deal with human abilities, limitations, habits and preferences, and how these change over time. Unique to HCI is that both the problem, the material as well as the requirements and constraints are not well known. The problem is clearly: how to support human cognitive task performance. Intelligent cognitive behaviour, abstract information processing and dealing with technology are very young branches of human conduct. Abstract mathematical information processing is only a few thousand years old and it is only since the great wars that technology has become a household product. It should not come as a surprise that performing cognitive tasks is difficult and that little is known about how to support it. The problem has become particularly relevant during the last decennium now that the rapidly expanding use of digital technology is exponentially multiplying the number of cognitive tasks that the humble user has to cope with. The same lack of knowledge exists with respect to the material used in performing cognitive tasks: information. Little is known about what exactly information is, and how much information of what kind constitutes too little, optimal and too much information. Something is known about the difficulty of dealing with abstract and symbolic information in comparison to real life information. A main difference between computer and pre-computer information processing is that the symbolic information of the former is less, less-well and less-directly related with physical reality. In terms of Norman (1988), symbolic information processing lacks the so-called affordances that are abundantly present in information processing by physical means. A genuine hammer or bicycle provide many more clues about how to use them, in which circumstances, and for which purposes than any pictorial, auditory, textual or virtually real representation of them on a computer. The relation between computers and software, and empirical reality is completely accidental and depends much more on learning and memory. Finally, to establish and determine requirements and constraints for tasks with empirical or physical components is much easier to do than it is for information processing tasks. Requirements for tasks with a physical component such as building a bicycle can be derived from the empirically measurable properties of the task, the user and the context, such as the distance to travel, body strength, and the state of the road. Information processing tasks depend on cognitive psychological properties, which lack directly measurable, empirical equivalents, and since cognitive psychology deals with human information processing, it is a very young branch of scientific conduct that is difficult and knows little. 1.3 Essentials of Cognitive Ergonomics 6 Chapter 1: Cognitive Ergonomics and UI Design In the previous paragraph it was argued that the combination of human cognitive task performance, the computer as a general purpose information processor, and the lack of knowledge about human cognitive behaviour and how it should be supported, is unique to HCI. This paragraph states what cognitive engineering is, or should be, in practical terms. To address the specific problems and aims of HCI, cognitive ergonomics might be defined more precisely as: the science and engineering practice of user interface design. Although it is a short definition, it mentions several important and interesting aspects: • cognitive ergonomics is a science and engineering practice. • cognitive ergonomics is about user interfaces. • cognitive ergonomics is a design trade. Stating that cognitive ergonomics is a science and engineering practice means that it is not black magic or art that depends on individual characteristics such as experience, talent, wisdom and skill, but that it is an endeavour that yields, and should yield systematic knowledge and methods. As a result, cognitive ergonomics can be taught and, as a prerequisite, research should focus on establishing a firm knowledge and methodological foundation before moving on to other interesting questions or applications. Cognitive ergonomics is about user interfaces because user interfaces provide human beings with the means to use cognitive tools and devices. In principle, cognitive ergonomics is not restricted to the use of computer systems just as user interfaces are not restricted to computer systems but since this study is situated in the area of human-computer interaction, cognitive ergonomics is taken to be about user interfaces as the means to create usable computer systems. Usability is not a characteristic of the functionality or the software of a computer system, but a characteristic of how users interact with computer systems to perform tasks in the user's work domain. This means that the term "user interface" in cognitive ergonomics is, and should be, extended beyond the most common definition as a piece of software. More specifically, user interfaces should be defined as: all the knowledge users need to perform tasks with a computer system (Moran, 1981; van der Veer, 1990). Cognitive ergonomics is a design trade since its aim is to improve, rather than gather knowledge about, human-computer interaction. Design methods should be the main concern of cognitive ergonomics because only design creates user interfaces, and because design methods, used implicitly or explicitly, are what all design processes have in common. Analysis, prediction, and coding are techniques to supplement design methods but do not create user interfaces by themselves. Since design methods are the only systematic way to improve user interfaces, they are an obvious candidate to provide structure to the field of cognitive ergonomics. 1.3.1 Cognitive Ergonomics is Science and Engineering This paragraph discusses the requirement that cognitive engineering be a science and an engineering practice. With respect to their development it draws parallels between the development of software engineering and cognitive ergonomics, and assessing cognitive A Formal Model for User Interface Design 7 ergonomics in terms of the minimum requirements for being a science and engineering discipline it is argued that too much attention is spent on fashionable research subjects at the cost of research that aims to create the scientific knowledge base as an organised set of facts, tools and methods. As such, this thesis is about what is necessary and not about what is most popular. First, cognitive ergonomics is a science and engineering practice. The question is not whether cognitive ergonomics is an academic subject, or to what extent it is a pure, empirical or applied science but, rather, whether there is something necessarily artful, mystic or creative about user interface design that precludes it from being approached in a predominantly methodological way. On the one hand, questioning the scientific nature of cognitive ergonomics is only of academic interest, and makes little sense because it won't help solve any practical problems. On the other hand, asking to what extent cognitive ergonomics is an art that depends on creativity does make sense because answering this question determines to what extent problems can be solved methodologically and without the need to resort to individual insights, experience, and expertise. When computers are regarded as theatre and user interface design is regarded as an art (Laurel, 1990, 1991) then the quality of user interface design is implicitly accepted as depending on the intuition, the skill and the talent of the individual artist. When, on the other hand, user interface design is regarded as cognitive engineering (Norman, 1986; Rassmussen, 1987), it is accepted as something that may be abstracted from the individual applicant, and explicitly passed on and improved, for instance by means of education. A similar reasoning applies to the field of computer programming, which may be regarded as an art (Knuth, 1968, etc.) but also as a type of engineering, as it is most commonly regarded and, even more stringent, as a discipline (Dijkstra, 1976). The developments in cognitive ergonomics seem to closely resemble the early developments in software engineering at the time when computers were commonly regarded as mysterious devices and the exclusive domain of computer specialists. In order not to waste valuable hardware resources and to increase programming productivity, attention went to the selection of talented programmers (e.g. Perry and Cannon, 1966) and individual differences (e.g. Curtis, 1988a). With respect to the task of operating computers, research focussed on programming languages and language constructs (e.g. Sime et al., 1977; Green et al., 1980), program style guides (e.g. Kernighan and Plauger, 1974) and guidelines (e.g. Shneiderman, 1980). The implicit rejection of programming as an art or individual craft made it finally possible to shift the focus of research towards the psychological aspects of programming and program understanding (e.g. Weinberg, 1971; Brooks, 1977; Curtis, 1988b) and creative problem solving (e.g. Guindon and Curtis, 1988). Currently, facilitated by the consensus around object-orientation the field seems underway towards the notion of the "software factory" with standard reusable system architectures (CORBA, 1999; DCOM, 1998), domainand software representations (UML, 1999), and methods for process control and maintenance (CCTA, 1998; Roa et al., 1996; Paulk et al.,