HCI Issues in the Design of Decision Support Systems
Philip J. Smith and Norm Geddes

1. Scope
   The initial section will bound the coverage of this chapter. In particular, it will emphasize that the focus will be on active decision support technologies (i.e., systems where the computer is doing significant information processing), rather than on more passive information display or communication systems (which also support human decision making, but do so in a more passive manner as far as the computer processing itself is concerned).
2. Underlying Technologies
   • Artificial Intelligence
     • symbolic reasoning/knowledge-based systems
     • neural networks
   • Operations Research/Optimization Approaches
     • decision analysis
     • math programming
     • regression models
     • simulation discrete event and continuous simulations, prediction of output given an input)
     • control theoretic models (prediction of output given an input, determination of recommended inputs to meet situation)
   
   This section would describe these different technological approaches and discuss the different high-level tasks to which they are applicable. It would emphasize the important conceptual issues addressed in the application of these technologies from the perspective of the technology approaches that issue (such as explanation, reasoning about uncertainty, learning, user modeling and task specific issues such as the model of planning that underlies the technology).
3. HCI Issues This section will provide a high-level overview of the important HCI issues that must be addressed in the design and use of decision support systems. This discussion will include issues regarding the support of distributed work; the impact of different roles and responsibilities; information display and communication; explanation; conceptual vs. mental models (regarding system functioning as well as situation awareness in particular scenarios); user modeling; reasoning about uncertainty, different forms of human vs. machine expertise, etc.). It will also address the issue of comprehensiveness and the potential for brittle performance, focusing on the nature of the gaps that develop if the underlying decision support technology approach is unable to provide well-delineated boundaries to define its support capabilities. These issues will be discussed in terms of the impact of using the different technological approaches discussed in Section 2.
4. Case Studies
   Supporting Distributed Work: The FAA has an operational system called the Flight Schedule Monitor that has been in operation for 2 years and that supports collaborative decision making regarding the use of ground delays to control air traffic into specific airports when there is something causing reduced capacity. This software allows FAA traffic managers at the Systems Command Center in Washington to collaborate asynchronously with Airline Operations Control to set capacity limitations and to determine which flights to delay on the ground. The software supports information exchange regarding FAA and airline intentions and uses optimization techniques to predict the outcomes of alternative plans and to reallocate resources (arrival slots at airports). It won a major award from ORSA last year as an example of an outstanding application of OR for decision support.

   The developers have agreed to give us training on the system and access to screen captures and data on its performance. This should make this an exceptionally rich case study.

   From an HCI perspective, this case study will focus on issues associated with effectively supporting cooperative work, on concerns regarding the evaluation of alternative solutions where there are several difficult to measure performance criteria (safety and equity as well as system efficiency or throughput), on issues of gaming when the environment involves competition as well as cooperation, and on information display so that the users can understand and work effectively with each other through the medium of the software.

   Dealing with Complexity: Traumatique is an emergency medicine associate (Univ of Penn). This system is an example of a decision support system that deals well with a complex domain with many time critical interactions. It will be discussed in terms of the challenges of ensuring effective cooperative problem-solving between the human and the computer when there is time stress in dealing with the integration of numerous information sources that have interactions in terms of their implications for appropriate diagnosis and treatment of a patient.

   Design-Induced Error: This section will focus on two fully functional laboratory prototypes that have been used to study how brittleness in the performance of the decision aid influences user performance. One system, the Antibody Identification Assistant (AIDA), is a rule-based system that supports performance on an abduction/diagnosis tasks, while the other, the Flight Planning Testbed (FPT) uses linear programming to support performance in planning the routes of flight. Both have been used in detailed studies regarding how different system designs and alternative roles (the person as critic vs. the computer as critic) influence human problem solving in terms of information access, situation assessment and the use of specific problem-solving strategies.

   Reducing Staffing and Training Requirements: An important use of decision support systems is in reducing training time and staffing needs. A lot of this literature is in an area known as Electronic Performance Support Systems (EPSS). There are numerous case studies here from the real world, where EPSS has significantly shortened time to skilled performance. These will be contrasted with one another to illustrate the challenges in reducing training and staffing while trying to still maintain a high level of performance.

   Keeping the Humans Responsible: There are numerous examples of computer-based decision support systems and automation resulting in loss of human responsibility to the extent that major accidents occurred. Examples of this will be described, along with examples of humans making inappropriate interventions that overrode or contradicted the recommendations made by decision support systems. Two specific systems, the Pilot's Associate and Rotorcraft Pilot's Associate, will then be discussed in term of using their designs to discuss how a decision support system can support humans while guarding against error in complex real-time systems.