Understanding consumer navigation behaviour.

This article sets the scene for interface design by first placing the human computer interface within the context of human factors, software engineering, and related subjects, and then examining some justifications for why it is necessary to spend time and money designing human-computer interfaces. The main aim of this article is to present knowledge of the theory, models, and methods relevant to human-computer interaction and skills for designing better human-computer interfaces. The authors have attempted to place interface design into a framework of software development by drawing on methods from systems analysis and design as well as ideas in human-computer interaction. The article, also initiates the debate within the Human-Computer Interaction (HCI) community about scientific and engineering approaches to the subject.

HUMAN-COMPUTER INTERFACE DESIGN

Design of the human-computer interface is part of a wider subject area of Human-Computer Interaction (HCI). This, generally, but not exclusively, fits within Computer Science and is becoming an established part of the research. However, HCI is an evolving subject and its boundaries are still hazy and fuzzy.

Some History

HCI has formed from a coalescence of interests and knowledge drawn from computer science and psychology, although it has also drawn upon linguistics, sociology, applied psychology, ergonomics and management science. Historically it started life as the Man-Machine Interface (MMI), in the early 1970s; see, for instance, James Martin's hook on dialogue design (Martin, 1973). The MMI was recognised in the UK's Alvey research programme as a sector alongside software engineering, intelligent knowledge-based systems, and hardware. Further interest was stimulated by recognition in the European Communities' (EC) Esprit research initiative and the Japanese fifth-generation project. As a result MMI became an established area of research endeavour and was renamed, to the politically more correct "Human-Computer Interaction." The track record in industrial practice has been less encouraging. Most industrial structured development methods make little reference to HCI, although more recent versions (e.g., SSADM version 4, O bject oriented analysis; Coad & Yourdon, 1991) do have sections on the human-computer interface, and some reference is made to graphical user interface design.

Interface Design and HCI

HCI design is the engineering process of designing interactive computer systems so they are efficient, pleasing, easy to use, and do what people want. Interface design is related to the software engineering part of computer science. HCI research covers a broad field ranging from the environment in which the interactive system is situated, the effect of the interface on people, both individuals and groups, to techniques, methods, and tools to help designers build interactive systems. HCI design subdivides into (a) background theory and knowledge upon which design is based; (b) issues related to the design process such as methods, principles, and techniques; (c) tool support for the design process; and, (d) finally, quality assurance of the product! Hence, the core issues are:

* Understanding the essential properties of people which affect their interaction with computers.

* Analysing what people do with computer systems and their interfaces; understanding users' tasks and requirements.

* Specifying how the interface should function, how it should respond to the user, and its appearance.

* Designing interfaces so that users' needs are fulfilled and the system matches users' characteristics.

* Development of tools to help designers build better interfaces.

* Evaluating the properties of human-computer interfaces and their effect on people to ensure good quality.

* HCI, in the small (programming in the small), deals with detail of interaction such as interactive devices, widgets, or formal specification of restricted dialogue sequences. This addresses how to design the detail of interaction.

* HCI, in the large, is concerned with what to design and how an application should help people achieve the work. This perspective has a common motivation with subjects such as systems analysis and requirements engineering.

HCI Artefacts

This approach concentrates on the product rather than the design process. The idea is that well-engineered HCI products should incorporate good ideas. These ideas if described and formalised, can be seen as theories of HCI, that is, they explain and predict properties of good design within the scope or purpose for which an artefact was designed. Thus, HCI could progress by understanding how to design classes of artefacts rather than using universal models or principles. This has been developed as the task-artefact cycle (Carroll, Kellog, & Rosson, 1991). The artefact or product is evaluated and more generalisable truths are extracted by "claims analysis." These truths are then fed back into the pool of HCI knowledge. New artefacts are designed by first understanding their purpose through a task analysis. Theory and principles may be imported from cognitive psychology to inform design. The artefact is then built and evaluated going through a cycle of iterative improvement. One problem with this approach is def ining the scope of an artefact and how generalisable the claims can be. For instance, hypertext can be seen as a good and successful HCI artefact. Claims can be extracted about why this should be so, but how far these claims can be applied to the design of related artefacts (e.g., information retrieval products) is unclear.

Engineering principles. These follow the concept of civil and mechanical engineering design. If principles and rules can be devised, then designers should have predictive guidance for certain problem situations. For instance, bridge designers know how certain classes of bridge design are good for specific problems and how materials in bridge pillars will react to stress and load. Could HCI designs be amenable to similar treatment? Unfortunately, HCI is itself still dependent on an underlying science, psychology, which is far from complete. It is hard to derive design rules from rival theories of cognition. Worse still, computer systems are built in a bewildering variety of domains, so it is difficult to specify a problem class. Nevertheless, design principles can help, especially if they are given with scooping rules and caveats to explain when they may and may not apply. The engineering approach also advocates methods, which bind principles and guidelines together into an agenda of issues, which must be atte nded to.

HCI AND OTHER DISCIPLINES

One particular discipline, ergonomics, has made a considerable contribution to human-computer interaction. Ergonomics, also called human factors in the USA, is a branch of applied psychology which aims to improve the design of machines for people. Many of the recommendations from ergonomics can be applied to computers, in terms of both hardware, for example, how Visual Display Units (VDU) terminals should be constructed and positioned, and software, for example, how the appearance of VDU displays is controlled. While several ergonomic issues will be described, constraints of space mean that many hardware and environmental issues, such as ergonomics of office design, lighting, seating, and so forth, will not be covered.

Another important influence on HCI has been cognitive psychology. This is a branch of psychology, which deals with theories, models, and experimental investigations on mental phenomena such as memory, reasoning, problem-solving and perception. Cognitive science uses computers extensively and shares many of the same motivations as Artificial Intelligence (Al), that is, understanding how people and machines can think and learn. HCI draws upon cognitive psychology for models and theories of human-computer interaction. Psychology is an important parent science for HCI, as it supplies knowledge about people. Designing computer systems for people needs to take human cognitive abilities and limitations into account. Linguistics and computational linguistics share an interest in communication with HCI. Natural-language dialogues draw upon work in computational linguistics, while other influences come from discourse models of human-human communication.

Sociology also contributes knowledge to HCI, although so far its influence has been less important than psychology. Computer-supported cooperative work (CSCW) is a sub-area of HCI concerned with group work and computer tools to support such activity. Sociology contributes theories of group action and communication to help CSCW design, although, as with cognitive models, the influence of theory on design is often hard to find. Sociology has also contributed to the methodology of HCI in the large and requirements engineering by hermeneutics or ethnomethodology. Whether such methods and approaches become part of HCI or mainstream software engineering remains to be seen.

HCI and Computer Science

HCI permeates many parts of computer science. Its closest relatives are software engineering and subjects concerned with software design, for example, systems analysis, specification, and design. It is also related to Al and knowledge-based systems, as many advanced interfaces require intelligent processes.

The methodological part of HCI addresses many issues also found in software engineering. Requirements analysis is a common concern, while task analysis and systems analysis show many similarities. Both analyse functions and data structures, and object orientation influences both subjects. Specification is also an area of convergence, as software engineering formalisms have been used to specify the behaviour of interactive dialogues. It is essential that HCI become an integral part of software engineering, in its formal context and in pragmatic methods for industrial systems development such as Structured Systems Analysis and Design Methodology (SSADM). To meet this goal, most of the design method presented in this book is placed within the context of SSADM, which, while not perfect, can claim to be the market leading method for commercial systems development in the UK.

Artificial intelligence shares HCI's interest in cognitive psychology. Both disciplines construct models of reasoning, memory, and problem-solving, although their approaches and perspective differ. Al is motivated towards building machines, which can exhibit human-like qualities in language, vision, reasoning, and learning, whereas HCI's interest is in using knowledge-based systems to help interaction. This happens in natural-language interfaces, intelligent dialogue management, adaptable user interfaces, and intelligent agents.

HCI also shares interests with software architectures in computer science. Methods, techniques, and guidelines will not be effective unless they are embedded with computerised support tools. HCI so far has had little influence on the Computer-Aided Software Engineering (CASE) community. In the future HCI will become a standard component of such tools as well as software engineering methods and principles. Interfaces are rapidly becoming more sophisticated and the construction of user interface design environments and UI management systems is becoming a subject in its own right.

In conclusion, HCJ is a core component of the computer science curriculum. It is an essential complement of software engineering and systems analysis/design, and shares much with AI.

A case for the importance of HCI in computer systems design has been made from the economic penalties of poor design practice. Poor interfaces increase errors and stress. Systems with poor interfaces will be underused or rejected; furthermore, designers will soon become liable for bad interfaces under standards legislation. An outline of the subject matter of HCI is given. Approaches to theory and concepts of HCI start with the three-layer model of craft, engineering, and applied science. Other variations such as bridging models, the influence of psychology on design, and task artefact theory give different perspectives. HCI is an engineering type of discipline but it also embraces theoretical study and computer architectures for user interfaces.

The relationship of HCI to psychology and sociology is reviewed along with its place within computer science. HCI draws theory from psychology and to a lesser extent sociology. It is a core part of computer science and with software engineering constitutes the study of software design. HCI also shares many issues with artificial intelligence and uses knowledge-based systems in intelligent interfaces.

User Psychology

This topic gives an overview of cognitive psychology, which is relevant to human-computer interaction. It starts with how we perceive information from the environment with the senses of sight and hearing and then progresses to understanding the information we receive. Memory is then investigated; how information is coded and possibly stored, with the limitations of human memory. This leads on to mental activity and how one reasons and solves problems.

Understanding Users

Basic anatomy of the human processor. The human brain is composed of a vast number of nerve cells, estimates varying around 15 billion. Nerve cells are the basic elements of human information processing and memory.

Nerve cells communicate by electrical signals, although this is not the same as in electrical circuits; instead it is electrochemical communication. This means that the human brain is not strictly digital like computers; rather, it works by a mixture of digital and analogue computing. Nerve cells can send signals to each other to indicate a state change, thus creating the digital I/O states necessary for computation. Nerve signals (called impulses) travel along nerve cells but to transfer to the next cell they must cross an intercell gap. This gap is very small (2-5 microns) but so is the electrical voltage, which means it cannot jump the gap. Inter-cell transmission is by chemical means. This electrochemical activity means that the speed of nerve messages is slow compared with speeds of electrical signals in computers.

Electrical signals from nerve cells usually come as a series of bursts, each signal being a transient 010 state change. The effect they have on receiving cells is to either increase or decrease the tendency of a receiving nerve to fire, giving a graded analogue effect. Inter-nerve cell connections can therefore be inhibitory and damp down a cell's activity, or excitatory and stimulate it. As most nerve cells in the brain have an average of 150 connections, this composite influence can be very complex and is capable of much more complex coding than in digital electronics. A recent branch of computer science, neural computing, has developed to build artificial versions of nervous processors.

Nerve cells and their artificial equivalents in neural nets are organized in complex lattices and networks. The effect the cells have on one another is determined not only by the quantity of impulses they receive from other nerve cells but also by the sensitivity of their reaction. Some cells require many impulses to excite them sufficiently to fire, while others are tuned to hair-trigger levels. Tuning and different inhibitory or excitatory connections can imitate AND/OR logic gates and other familiar computer logic components. The use of feedback connections allows nerve cells to alter their connections according to patterns of input. This property allows the network to learn as new connections can be made and old ones suppressed.

Nerve cells in the brain are highly interconnected in a very complex network. The human wiring diagram is infinitely more complex and subtle than the most advanced microprocessor and is still poorly understood. Attempts to follow the real architecture of the brain at lower levels at present are unrewarding; therefore to further our understanding of the human machine an abstract model is used, that is an interpretation of how the logical processing units in the brain may work.

Cognitive Models

These models have been devised by psychologists to explain human mental activity by an analogy with computer processing. It is important to remember that models are only an abstraction; the final story of how the human machine works will be much more complex. Cognitive models, however, are useful because they illustrate the advantages and limitations of the human machine. In interface design these qualities need to be taken into account. In the following sections perception and cognition will be explored using an information processing model based on research at Xerox by Card, Moran, and Newell (1983).

Another more recent model, Interactive Cognitive Sub-systems will also be used to explain human information processing from a different viewpoint. All models require input from the outside world. This concerns perception, the process of receiving information from the outside world, while cognition is the mental activity we describe in everyday terms, as reasoning, problem-solving, thinking, and learning. The boundary between the two is blurred because, as we receive information, we also interpret it.

THINKING AND PROBLEM-SOLVING

Thinking, reasoning, and problem-solving are all human mental activities which process data derived from our senses and long-term memory. Problem-solving is something we do every day of our lives when we come up against the unexpected. It may be defined as "the combination of existing ideas to form new ideas." An alternative view focuses on the cause. Problems arise when there is a discrepancy between a desired state of affairs and the current state of affairs and there is no obvious method to change the state. Problem-solving progresses through several stages. The names of stages vary between authorities on the subject, so the following scheme is a generalisation:

1. Preparation or formulation: the goal state is defined and necessary information for a solution is gathered.

2. Incubation or searching: anticipated solutions are developed, tested, and possibly rejected, leading to more information gathering, and development of alternative hypotheses.

3. Inspiration: the correct solution is realised.

4. Verification: the solution is checked out to ensure it meets the goals and is consistent with the information available.

To illustrate how problem solving may work, another model will be employed. The Goals Operators Methods Selection rules (GOMS) model of Card et al. (1983) owes its heritage to the General Problem Solver Model (GPSM) of Newell and Simon (1972).

Problem-Solving Models

The GOMS model is composed of a set of goals and sub-goals organised in a conceptual problem container, called the problem space. During the problem searching phase, goals are broken down into a sub-goal network; searching then proceeds by traversing the network and testing hypotheses at each node. At each sub-goal node data are read into short-term memory, evaluated and then stored back into long-term memory as searching progresses to the next sub-goal. Eventually, if the search network has been well constructed and all the facts are available to be evaluated, and the subgoals pass the tests, the final node is reached resulting in the problem solution. This operation is the familiar reasoning strategy of problem solving by steps, that is, if X is A, then Y is probably B, which means that Z must be true, and so forth. It uses top down decomposition of the problem, a technique also adopted by structured analysis. However, not all problems can be approached in such a sequential manner.

Other components of the model are: (a) operators, which describe the actions necessary to reach the goal, and (b) methods, which group actions to arrive at the strategy or approach to the problem. Actions ultimately are low-level cognitive primitives defined as perceptual, mental, and motor in the extension of GOMS called the keystroke level model (Card et al., 1983). This allows time for simple "unit tasks," for example, deleting a word, to be calculated from the sequence of perceptual, mental, and motor actions, as these have set times of 70-100 ins. However, this low-level usage of GOMS can only predict error-free performance and getting the detailed sequence of actions right is difficult and time consuming.

Models of Interaction

So far this topic has been considering human information processing. However, interface design concerns computer software as well as wider issues of the computer system environment. Therefore a model of interaction is necessary to help focus design issues. Norman and Draper's (1986) model has been one of the most influential models of people interacting with computers. It is simple, yet it encapsulates the key points about human-computer interaction. The essence is that interaction occurs in cycles of user and system actions. These create the dialogue between the user and the computer. Between the two, as Norman points out, there are two gulfs, which have to be bridged by the interface as shown in Figure 1.

* The gulf of execution: when the user is faced with the question of what to do next. To help bridge this gulf the interface has to suggest and prompt the user about possible actions.

* The gulf of evaluation: this occurs after a user action and the consequent system response when the user has to figure out what has happened. Design here has to help by making the effects of user action explicit and giving a clear representation of the system state.

Interaction continues in a cycle of deciding what to do (formulate goal), user action, system reaction, and so on. However, things are a bit more complex than Norman and Draper's model suggests. First, how we plan what to do varies. Plans may come from procedural memory: we just retrieve the appropriate procedure and run it as a series of actions. Alternatively we may decide what to do according to the state of the environment we perceive. This is "situated action," in which mini-plans are executed according to the input triggers from the environment. Interaction with graphical user interfaces may well be situated according to interface display. There is still much debate about how plan or environmentally driven we are; the truth is probably that we are individually different, some people liking to do things in a well-planned manner, while others act more "on the spur of the moment."

Bridging the gulfs of execution and evaluation can be helped by use of analogical memory. This gives a context for interpretation in the interface metaphor. Design can also help by guiding users' attention to key messages. There are a set of issues, which map the basic cycle of interaction; for instance, for execution of a user action, design of prompts is important; and for evaluation, feedback messages are necessary. To help the user plan future goals, overviews and summaries are required. This leads to consideration of principles for HCI design.

Principles of Human-Computer Interaction

Models of human information processing and knowledge of cognitive psychology allow a set of tentative principles to be drawn up, although care has to be taken in applying principles in practice, because the context of design has a strong effect on the validity of generalisations drawn from psychology. Unfortunately, psychology does not lend itself to such a venture, as many explanations of human behaviour are still models and hypotheses, and in some areas little definite proof exists. However, some principles can be derived despite of this limitation, although they have to be supplemented by justifications to substantiate them based on interpretation in a context, as well as empirical evidence of usability.

Seven basic principles are proposed:

Consistency. This is similarity of patterns, which may be perceived in task, in presentation of information, and other facets of an interface design. Consistency reduces the human learning load and increases recognition by presenting a familiar pattern As we are pattern recognition machines, the more consistent patterns are, the less we have to learn, and the easier an interface will be to use.

Compatibility. This is the goodness of fit between the user's expectation and realty of an interface design. It's principle follows on from consistency to state that new designs should be compatible with, and therefore based upon, the user's previous experience. If this is followed, once again recognition is enhanced, learning is reduced, and the interface should be easier to use. Compatibility relates to the concept of users models; the essential concordance is between the user's mental model of the task and the task model embedded in software by the designer.

Predictability. The interface should always suggest the user, the possible action. This may be by way of messages, prompts, and so forth, or information may be contained in metaphors and icons in graphical user interfaces. Predictability has a subprinciple of observability, which states that the current state of the system should be made explicit to the user. Only by having information about what can be accomplished and the current state can users plan what to do next. An example would be highlighting a graphical object with handles' showing that it has been selected.

Adaptability. Interfaces should adapt to the user in several ways. The user and not the computer should be in control; so the interface adapts to the user's speed of work and does not enforce continuous attention. Also, the interface should adapt to individual user characteristics, skill levels, and so forth, as to do otherwise would offend the compatibility principle. Adaptability, however, must not be overdone, otherwise the consistency of the interface is reduced.

Economy and error prevention. This principle is based more on common sense than psychology. Interface designs should be economic in the sense that they achieve an operation in the minimum number of necessary steps and save users' work whenever possible. Short cuts in dialogues and setting defaults are examples of this principle in design. Error prevention is the inverse, trying to save the users' work by recovering from errors. Whenever possible, an interface should not allow the user to get into states causing damaging errors, for example, deleting all files without backups.

User control. The interface should function at the user's pace according to their commands and should not attempt to control the user. This principle is related to predictability, as users should be able to forecast what to do next from a system's current state. It has a sub-component: reversibility. This states that users should be able to backtrack at will when mistakes are made. Reversibility is manifest in "undo" commands.

Structure. Interface designs should be structured to reduce complexity, because humans process information by classifying and structuring it within a framework of understanding. One deals with complexity in the environment by imposing order on it and trying to automate solutions to problems. Classification, structuring of information, and skills are consequences of this propensity to organise and automate.

Structuring should be compatible with the user's organization of knowledge and not overburden memory. This leads to a sub-component of simplicity and relevance: information should be organized so only relevant information is presented to the user in a simple manner.

Principles are intended for overall guidance during design and as a set of criteria against which interfaces may be evaluated. To apply principles in the design process, they have to be translated into guidelines, which pertain to different aspects of a human-computer interface.

Guidelines, in turn, are modified by the context of a particular application into design rules. Unfortunately, systems and people are complex and to issue a simple set of guidelines for all situations may be appealing but in reality would only be misleading.

A key point of user-centered design is to prevent overloading of human information processing facilities, in particular short-term memory. People deal with vast quantities of data from the environment by filtering them and abstracting interesting qualities from basic data. Hence, principles, which help memory and human reasoning abilities, are important.

Designs need to be considered in terms of the objectives of creating good human-computer interfaces; which raises the question of assessment.

The effectiveness of interface designs is measured in terms of usability, utility, efficiency, and ultimately user satisfaction. There are three basic concerns about the quality of an interface design:

* How well does it fulfill the users' objectives?

* How easy is it to learn and use?

* How effective is it in helping users carrying out some work?

CONCLUSION

A design should aim to provide users with what they require to fulfill their objectives. This concept is common to systems analysis and HCI design, that is the matching of user requirements to the facilities provided in the system. In HCI terms this is called task fit, providing the appropriate tool to carry out a required task. A system may be easy to use and learn, but if it does not do what the user wants, it will be useless. Task fit is a consequence of the compatibility principle, memory, and mental models, the users' expectation of reality and what they get.

Effectiveness is how well a computer system helps users achieve work at an acceptable cost. Cost can be incurred both on the computer side in programming and on the user side by mental effort, stress, and operation time. Good design is about minimising human costs within the available budget of computer development costs. A major component of effectiveness is efficiency. This is often measured in terms of how easy an interface is to learn and use combined with the inverse measure of how many mistakes are made. Generally it may be thought that there is a trade-off between ease of use and ease of learning, but evidence points the other way: interfaces should be easy to learn and easy to use. Efficiency is a consequence of the economy, consistency, and compatibility principles. Utility and satisfaction with an interface depend on how it helps us. This is partly a consequence of compatibility but it also relates to the creative side of task support design. A computer system should empower people, or, put more sim ply, help them to do their work with more creative enjoyment.

The concern for how much of an interface, and, hence, a system, is used is often ignored. There is a natural tendency for many computer products to add extra features with each successive release. A product, which starts out life as a simple application, can gradually evolve into a monster of considerable complexity. This imposes a learning penalty for new users in particular, who may only want a fraction of the overall functionality provided. Usability is also concerned with problems caused by excessive functionality. However, there are other reasons why many functions are never used, including poor task fit, poor training, and poor interface design. Users may be ignorant of, or can't be bothered to use, a facility, even though it may fulfill their task very well.

Ideally a system should be sufficient for users' needs and compatible with their experience. This will be based on previous experience of computer and noncomputer systems. It is the analyst's task to capture that knowledge and build the new system to be as compatible as possible with the users' expectations. Full compatibility may be technically impossible because of improvements to the logical design of the new system. Also, user models differ according to variation in individual experience; one single model cannot be completely compatible with each individual's view. Summing up, one concludes that the final design remains a compromise with inter-individual variations.

References

Card, S.K., Moran, T.P., & Newell, A. (1983). The psychology of human-computer interaction. Hillsdale, NJ: Lawrence Erlbaum.

Carroll, J.M., Kellog, W., & Rosson, M.B. (1991). The task-artifact cycle. In J.M. Carroll, (Ed.), Designing interaction: Psychology at the human computer interface, Cambridge, UK: Cambridge University Press (pp. 74- 102).

Coad, P., & Yourdon, E. (1991). Object-oriented analysis (2nd Ed). Englewood Cliffs, NJ: Prentice-Hall.

Martin, J. (1973). Design of man-computer-dialogues. Englewood Cliffs, NJ: Prentice-Hall.

Newell, A., & Simon, H. (Eds.) (1972). Goals problem solver model. New York: Cambridge University Press.

Norman, A.D. & Draper, S.W. (Eds.), (1986). User centered system design: New perspectives on human-computer interaction. Hillsdale, NJ: Lawrence Erlbaum.