Cognitive models are computer programs that simulate human performance of cognitive skills. They have been useful to HCI by predicting task times, by assisting users, and by acting as surrogate users. If cognitive models could interact with the same interfaces that users do, the models would be easier to develop and would be easier to apply as interface testers. This approach can be encapsulated as a cognitive model interface management system (CMIMS), which is analogous to and based on a user interface management system (UIMS). We present five case studies using three different UIMSes. These show how models can interact with interfaces using an interaction mechanism that is designed to apply to all interfaces generated within a UIMS. These interaction mechanisms start to support and constrain performance in the same ways that human performance is supported and constrained by interaction. Most existing UIMSes can and should be extended to create CMIMSes, and models can and should use CMIMSes to look at larger and more complex tasks. CMIMSes will help to further exploit the synergy between the disciplines of cognitive modeling and HCI by supporting cognitive models as users.

What is the role of a cognitive architecture in shaping a model built within it? Compared with a model written in a programming language, the cognitive architecture offers theoretical constraints. These constraints can be "soft," in that some ways of constructing a model are facilitated and others made more difficult, or they can be "hard," in that certain aspects of a model are enforced and others ruled out. We illustrate a variety of these possibilities. In the case of Soar, its learning mechanism is sufficiently constraining that it imposes hard constraints on models constructed within it. We describe how one of these hard constraints deriving from Soar's learning mechanism ensures that models constructed within Soar must learn a display-based skill and, other things being equal, must find display-based devices easier to learn than keyboard-based devices. We discuss the relation between architecture and model in terms of the degree to which a model is "compliant" with the constraints set by the architecture. Although doubts are sometimes expressed as to whether cognitive architectures have any empirical consequences for user modeling, our analysis shows that they do. Architectures play their part by imposing theoretical constraints on the models constructed within them, and the extent to which the influence of the architecture shows through in the model's behavior depends on the compliancy of the model.

Users who have worked with just one or two pieces of application software on a computer system are often faced with the need to use a new piece of software on the same system.Consistency between program interfaces is intended to make the new program easier to learn in this situation, but how "consistency" should be defined is not always clear.We present a model of analogical reasoning that describes how users rely on interface consistency to induce correct actions in a new situation.Versions of the model are implemented in ACT-R and Soar.The model yields a clearer and more principled understanding of design guidelines that recommend interface consistency.

Programmable User Models (PUMS) are tools based on psychological theory that enable interface designers to predict the usability of a proposed design.This paper presents a variant in which the PUM, implemented in Soar and incorporating the constraints of Display-based Task-Action Grammars, learns the task-action mapping by being guided by the designer during performance.We show that the more consistent and interactive the interface, the easier it is for the designer to teach the PUM the necessary taskaction mapping. PREDICTING INTERFACE LEARNABILITYA Programmable User Model (PUM, [14]) is a psychologically constrained architecture in which the interface designer is invited to program a simulation of a user.By making the user's knowledge explicit, PUMS

A variety of theoretical approaches to HCI have been proposed in recent years.HCI theories not only have to account predictively for the observed phenomena in interaction but should also be able to demonstrate their utility in improving the design quality of HCI products.The key question to be addressed in this panel is how HCI theories can, and possibly do, influence the design process.Three approaches will be examined.First is the emulation-based theory of design developed by John Carroll and his group at IBM, Yorktown heights.This proposes that designed artifacts (i.e.successful designs and systems) effective embody HCI theory as it is practiced in the field.Design is advanced by generation, evaluation and evolution of artifacts.The second approach, advocated by Richard Young and his colleagues at the Applied Psychology Unit in Cambridge, is to apply cognitive models and architectures to HCI design.This approach will explore how cognitive simulation models can be applied to the design process and the properties required by such theoretical-based models so that they are tractable and usable by designers.The third approach, proposed by John Long and his colleagues at the Ergonomics Unit, University College, London, proposes the development of prescriptive engineering design principles, both substantive and methodological, which in the long term would support practices seeking solutions to the general design problem of HCI.The principles would derive from an engineering conception of HCI, which in the short term supports the development of techniques and methods for effective design.The proponents will discuss how their approaches can support design practice and the ways in which theories are made available to the designer, either as artifact-products, predictive models, prescriptive guidelines and methods, or simply as designers' "tools for thought". THE PANELISTSJohn Carroll is manger of User Interface Theory and design at the IBM Watson Research Centre in Yorktown Height, New York.His research is in the analysis of learning, problem solving and language capacities that underlie endeavour and experience, especially human interaction with computers.

The knowledge analysis of a device and a task, when written in an external Instruction Language and translated into rules for a programmable cognitive architecture, enables a designer to predict conceptual errors in device usage. This kind of prediction lies outside the scope of GOMS-based models. The cognitive architecture, which is referred to as a “Programmable User Model” (PUM), incorporates a limited problem-solving capability based upon means-ends analysis and multiple problem spaces. The example presented, concerning a simple text editor, illustrates the application of a PUM and demonstrates that a correct description of local knowledge does not necessarily lead to correct behaviour. This can serve to alert the designer to difficulties with the usability of a proposed interface.

We assert that the product of user interface design should be not only the interface itself but also a rationale for why the interface is the way it is. We describe a representation for design based around a semi-formal notation which allows us explicitly to represent alternative design options and reasons for choosing among them. We illustrate the approach with examples from an analysis of scrolling mechanisms. We discuss the roles we expect such a representation to play in improving the coherence of designs and in communicating reasons for choices to others, whether designers, maintainers, collaborators or end users.

A Programmable User Model (PUM) is a psychologically constrained architecture which an interface designer is invited to program to simulate a user performing a range of tasks with a proposed interface. It provides a novel way of conveying psychological considerations to the designer, by involving the designer in the process of making predictions of usability. Development of the idea leads to a complementary perspective, of the PUM as an interpreter for an “instruction language”. The methodology used in this research involves the use of concrete HCI scenarios to assess different approaches to cognitive modelling. The research findings include analyses of the cognitive processes involved in the use of interactive computer systems, and a number of issues to be resolved in future cognitive models.

We offer an account of how users choose between alternative methods which take different times to accomplish the same task. Users offered a choice between two methods do not necessarily pick the faster. We argue that users reduce the complexity of the decision space by applying a 'simple compensation schema' acquired from everyday experience. Linear models of performance time enable us to predict how users will view the situation in terms of this schema, and how accurate assessment of the optimal choices within the schema-based assimilations can result in an apparent bias in favour of one method.

A scenario is an idealised but detailed description of a specific instance of human-computer interaction (HCI). A set of scenarios can be used as a “filter bank” to weed out theories whose scope is too narrow for them to apply to many real HCI situations. By helping redress the balance between generality and accuracy in theories derived from cognitive psychology, this use of scenarios (1) allows the researcher to build on empirical findings already established while avoiding the tar-pits associated with the experimental methodology, (2) enables the researcher to consider a range of phenomena in a single study, thereby directly addressing the question of the scope of the theory, and (3) ensures that the resulting theory will be applicable in HCI contexts.