Design Models for Interactive Problem-solving: Context & Ontology, Representation & Routines
Keith A. Butler
Abstract
ACM Classification
Director of User Experience
We describe and illustrate a new framework for the design of interactive problem-solving based on recent research on the psychology of distributed cognition.
H5.m. Information interfaces and presentation (e.g., HCI): H.1.2 [Information Systems]: User/Machine Systems – Human factors; H.3.3 [Information Systems]: Information Storage and Retrieval Information Search and Retrieval; D.2.1 [Software Engineering]: Requirements/Specifications methodologies
Microsoft Global Site Management & Analysis Redmond, WA 98052 USA
[email protected] Jiajie Zhang Associate Dean for Research School of Health Information
Introduction
Sciences
Keywords
University of Texas at Houston
Representation effect, interactive problem solving, analysis methods, design methods, information architecture, workcentered design, work ontology, top-level routine
Houston, TX 77030
[email protected]
Copyright is held by the author/owner(s). CHI 2009, April 4�9, 2009, Boston, Massachusetts, USA. ACM 978-1-60558-247-4/09/04.
Keywords
Problem-solving is ubiquitous in human endeavors, and essential for success in business, government, engineering, and science. People even seem to get intrinsic satisfaction from doing a good job on solving a challenging problem, to the point it’s been a part of recreational games throughout history. Conversely, when problem-solving is undermined by awkward, unusable information systems the results can be frustration, productivity loss, and even disaster. What more should our industry do to support problemsolving for the pleasure and professional needs of our customers? We describe a framework to help answer that question for a variety of domains when the problem-solving is technical in nature. It is based on psychology research and the Theory of Distributed Cognition [3, 4, 5, 6, & 7]. This theory clarifies how
the ontology of a problem constrains the effectiveness of interactive problem representations, and in turn how representations constrain the cognitive strategies and procedures that are used to solve problems.
on the design. The following narrative describes how two key design artifacts, the Top-level Routine and the UI View satisfy them.
Problem Ontology Our framework, which we call work-centered design (WCD) provides a framework that incorporates and relates models that are important design-time artifacts: the ontology of the problem; integrated user procedures with machine procedures; and it accounts for their relationship in design to such powerful factors as work context and cognitive architecture. WCD facilitates overall design integration across disciplines via concurrent engineering methods.
Many approaches to understanding problem-solving have been studied in the psychology literature. The approach we exploit here is an extension of the abstract modeling technique recommended by Newell & Simon [6]. Our approach introduces a new design-time artifact: Ontology that is user-centered. It is modeled with objects, operations and their relationships, and
Our initial research on these design artifacts was validated on a complex scheduling system for aircraft [1]. This WIP elaborates the framework and illustrates it with a familiar problem. Figure 1 is organized around the external constraints on a design, shown in dark blue. These external factors are beyond the scope of the project to change, but need to be understood in terms important constraints and requirements they impose
Figure 1: Design framework
constraints. It serves several important purposes. Ontology provides a declarative model of the work entity that must be produced by the new system. Ontology is an abstract, implementation-independent description of that entity. It describes essential requirements independently of any technology or implementation, work procedures, or cognitive strategies. This entity defined in the ontology constrains Top-level Routines (TLR) by providing an abstract model of the product that the new system must produce. This makes work procedures more sensible by relating them to the states of the problem. The TLR must be effective to change the state of the work entity to the goal state, or the system will fail, regardless of cutting-edge technology, fancy features, or other technical merits.
isomorphs from ontology models [7, 8, 9]. Ontology can also used to define the problem space and generate corresponding information architecture. Consider the simple example of how Work Context and Work Ontology are related in Figure 2.
Work Context Ontology separates the inherent nature of the work from Work Context. Context imposes important constraints on the way work is allowed to be performed, which must be understood and addressed
Ontology also constrains the Representation in the UI View to be isomorphic to it. All the objects, relationships, constraints and states, must have a corresponding representation in the UI View. As we will discuss in more detail later, Zhang has developed a technique called representational analysis to generate effective visual
Figure 2: Modeling a purchase
for a successful design. These constraints include Physical Factors and Geographical Factors, Computing Architecture (infrastructure) and Organizational Rules that govern the acceptable ways the work can be performed (e.g., obtaining authorization, collecting taxes). Figure 2 depicts the familiar example of a cash purchase between a buyer and a seller when the Work Context is a yard sale and the item is a small item, such as a used tennis racket. In the upper part of figure 2 the Work Ontology Model provides a declarative model of the fundamental nature of the work. The ontology of a purchase has a buyer, a seller, a payment, and an item for sale. At the start of a purchase the buyer owns the payment and the seller owns the item. The fundamental operator of the purchase ontology is to change the ownership relations: e.g., ownership of the racket switches to the buyer, and ownership of the payment switches to the seller. It would not be a purchase without each of the four objects and the two ownership relations that switch. If any were missing it would be something else- a theft, a gift, etc., but not a purchase. The Context row in the middle of Figure 2 shows three different contexts to illustrate how they constrain Toplevel Routines. Representations #1 and #2 are based on different infrastructure and each enables different Top-level Routines that satisfy the constraints of remote, asynchronous context, but at different costs, while the Work Ontology remains constant. The yard sale context allows people to solve problems for a cashand-carry purchase because buyer and seller can
negotiate in-person, then exchange cash for the racket to complete the transaction. The state change for ownership relations is accomplished by exchanging physical possession. Work is always performed (and studied) in some context that constrains the procedures for performing the work. A typical yard sale constrains buyer and seller to be in the same place at the same time, so beyond negotiating a price there is little problem solving that is required. If the context changed, however, from face-to-face to remote and asynchronous, with separated locations and different times, it would pose problems for cash-and-carry procedures. Buyers and sellers need different representations that enable procedures to overcome these constraints, such as mail-order catalogs or shopping web sites. Understanding the relationship between Work Context and Work Ontology is valuable for design because it distinguishes requirements that are intrinsic to the problem from those that may vary with context. On the other hand, this tells us little about how to accomplish the work. For that must understand how context constrains procedures.
Top-Level Routine The Top-Level Routine (TLR) is central to the success of any interactive system. It is made up of User Procedures and of Machine Procedures, which must work in coordination to change the state of the work entity to the required goal state. In our meta-model the User Procedures are as much a part of the design as the Machine Procedures because they must work in coordination. The TLR is also constrained by the Work Context. If the TLR fails to achieve the goal state for a given context, then the main purpose of the system will not be realized. Technical capabilities, such as secure web sites that provide asynchronous communication
between buyers and sellers, and trusted systems for payment and delivery, are examples of how technology can enable a TLR that overcomes the constraints of a given Work Context. Machine Procedures are constrained by the Computing Architecture, which is the part of the Work Context environment in which the machine system performs. User Procedures are constrained by the Cognitive Architecture [2], which is part of the user’s person, and by the UI View, which is a Representation of the problem that appears in the user interface (UI View). The Cognitive Architecture is the durable characteristics of information processing of the human user. It changes slowly, if at all, and is a important source of constraints on the User Procedures in response to the UI View and the Work Context. The main sources of added value of an interactive problemsolving system are from improved User Procedures that reduce human effort and from Machine Procedures that overcome obstacles of Work Context, or that reduce or amplify human effort.
Representations There are two Representations of the Work Ontology: the Coder’s View and the User Interface (UI) View. The UI View is an external Representation of the work domain that the system provides to the user. In WCD the UI is treated as a representation of the problem to the user, whether or not it was designed with that in mind. The Coder’s View may have different content than the UI View, but it has big impact on the UI View because the Machine Procedures generate it. The design of the UI View has several important constraints: it must be isomorphic with the Work Ontology; and it must induce the required User Procedures, which are constrained by the Cognitive
Architecture. We view the UI of a problem-solving system as the representation of that problem that the user must deal with: in order to be sensible and usable the UI must provide an effective, interactive representation of the problem, based on the principles of the representation effect [3, 5, 6, & 7]. External representations are only effective if they redistribute cognition to achieve more powerful problem solving through better insight, providing cognitive affordance and easy awareness of status. But there can be many different representations of the same problem that are displayed in a wide variety of ways. Some of these isomorphic representations may produce drastically different psychological effects, despite being logically equivalent. When isomorphic representations produce drastically different human understanding, strategies, and behavior, the difference is known as the representation effect [5, 6, & 7]. A given representation can enable some strategies and procedures while making it difficult to follow others. The representation effect implies that it will not be sufficient to assemble accurate problem data and give the user flexibility to discover effective views. Sixty years of psychological research on problem solving consistently shows that the initial representation encountered can constrain users’ strategic understanding of the problem to the point that they may fail to recognize the possibility of alternate views. An effective UI View must provide users with insight for the problem domain, with cognitive affordance to interact with the representation, and with visibility of
the state of the entity and of the effects of user controls.
Summary
provides important guidance for effective representations and a criterion for evaluating the combined procedures of users and their machine.
The WCD framework provides the clarity needed for complex design problems. WCD focuses on models of ontology and context separately to understand requirements, then designs procedures and supporting representations to satisfy them. The challenge for design is to induce complementary user behavior for the TLR by creating the needed representation. A major design-time advantage of Work Ontology is that it
Our most recent project for WCD is a notoriously difficult domain: online, technical self-support for consumer users of personal computers. To date summative testing on this new design has increased solution discovery by 33% over the currently deployed system. We will discuss the newest results in our poster session.
Citations [1] Butler, K.A.; Zhang, J; Esposito, C; Bahrami, A.; Hebron, R.; & Kieras, D. (2007). Work-centered design: a case study of a mixed-initiative scheduler, Proceedings of the SIGCHI conference on Human factors in computing systems, April 28-May 03, 2007, San Jose, CA.
[4] Newell, A. and H.A. Simon. Human Problem Solving. 1972, Englewood Cliffs, NJ: Prentice-Hall, Inc.
[2] Byrne, M. D. (2003). Cognitive architecture. In J. Jacko & A. Sears (Eds), Human-Computer Interaction Handbook. Mahwah, N.J.: Lawrence Erlbaum Associates. pp. 97-117.
[6] Zhang J. (1996). A representational analysis of relational information displays. International Journal of Human-Computer Studies, 45, 59-74.
[3] Kotovsky, K., Hayes, J. R., & Simon, H. A.; Why Are Some Problems Hard? Evidence From Tower of Hanoi; Cognitive Psychology, 1985, 17, 248-294.
[5] Zhang, J. & Norman, D. (1994) Representations in distributed cognitive tasks. Cognitive Science, 18, pp. 87-122.
[7] Zhang, J. (2001) External Representations in Complex Information Processing Tasks. A. Kent, ed., in Encyclopedia of Library and Information Science. Marcel Dekker, New York.