Decision Support Systems
DECISION SUPPORT SYSTEMS
Decision support systems (DSS) are tools meant to assist in human decision-making (Turban and Aronson 2001). In an increasingly complex and rapidly changing world where information from human, software, and sensor sources can be overwhelming, DSS tools can serve as a bridge between the social and technical spheres. DSS tools offer support based on formal, technical approaches, but do so within a context that is often largely socially mediated.
Most DSS tools are assembled out of hardware devices and software constructs. The hardware devices, in the early twenty-first century, are dominated by digital computers and peripherals such as sensors, network infrastructure, and display and alerting devices meant to interact with these. Historically, many DSS were hard-wired to solve a specific task; control systems in nuclear power plants are an example. DSS hardware is increasingly dominated by physically distributed systems that make use of wired and wireless networks to gather and share information from and with remote sources (Shim et al. 2002). The convergence of remote sensing, sensor-networks, and distributed computational grids using the Internet as a foundation in the late 1990s–early 2000's reflects this trend.
The software, or algorithmic, component of DSS derives from historical research in statistics, operations research, cybernetics, artificial intelligence, knowledge management, and cognitive science. In early monitoring decision support systems the algorithms were typically hard-wired into the system, and these systems tended to be unchanging once built. Software-based decision support allows for multiple approaches to be applied in parallel, and for systems to evolve either through new software development or via software that "learns" through artificial intelligence techniques such as rule induction (Turban and Aronson 2001).
When used appropriately, DSS tools are not meant to replace human decision-making—they are meant to make it more effective (Sprague and Watson 1996). DSS tools do this by presenting justified answers with explanations, displaying key data relevant to the current problem, performing calculations in support of user decision tasks, showing related cases to suggest alternatives, and alerting the user to current states and patterns. In order to be a support rather than a hindrance, these tools must be constructed with careful attention to human cognitive constraints. As a result, DSS design is a prime area of human-computer interaction and usability research. In many cases, DSS tools make use of adaptive software interfaces; depending on the situation, different contents will be displayed on the interface, so as not to overwhelm the user with secondary or irrelevant information.
Decision Support Tools
Decision support tools fall into two broad classes: those that operate at the pace of the user (for example, to support planning decisions) and those that operate at or near the pace of real-time world events (such as air traffic control systems). The decision-making domain can be further divided into situations in which the system can be completely and accurately defined (in other words, closed and formal systems) and those where this is not feasible, desirable, or possible. The former is not normally considered a prime situation for decision support because a formal situation can be addressed without human intervention, while the latter requires the hybrid human-machine pairing found in DSS. In the case of open systems, heuristic approximations (rules of thumb) are needed in lieu of formal models; these may also be needed in cases in which a formal model exists but cannot be computed in a reasonable amount of time.
Systems that operate at the pace of the user provide support for such tasks as planning and allocation, medical and technical diagnosis, and design. Typical examples include systems used in urban planning to support the complex process of utility construction, zoning, tax valuation, and environmental monitoring, and those used in business to determine when new facilities are needed for manufacturing. Such tools include significant historical case-knowledge and can be transitional with training systems that support and educate the user. Formal knowledge, often stored as rules in a modifiable knowledge base, represent both the state of the world that the system operates on and the processes by which decisions transform that world. In the cases where formal knowledge of state and process are not available, heuristic rules in a DSS expert system or associations in a neural network model might provide an approximate model. DSS tools typically provide both a ranked list of possible courses of action and a measure of certainty for each, in some cases coupled with the details of the resolution process (Giarratano and Riley 2005).
Systems that operate at or near real time provide support for monitoring natural or human systems. Nuclear power plant, air traffic control, and flood monitoring systems are typical examples, and recent disasters with each of these illustrate that these systems are fallible and have dire consequences when they fail. These systems typically provide support in a very short time frame and must not distract the user from the proper performance of critical tasks. By integrating data from physical devices (such as radar, water level monitors, and traffic density sensors) over a network with local heuristics, a real-time DSS can activate alarms, control safety equipment semi-automatically or automatically, allow operators to interact with a large system efficiently, provide rapid feedback, and show alternative cause and effect cases. A central issue in the design of such systems is that they should degrade gracefully; a flood monitoring system that fails utterly if one cable is shorted-out, for example, is of little use in a real emergency.
History of Decision Support Research and Tools
As indicated above, DSS evolved out of a wide range of disciplines in response to the need for planning-support and monitoring-support tools. Management and executive information systems, where model and data-based systems dominated, reflect the planning need; control and alerting systems, where sensor and model-based alerting systems were central, reflect the monitoring need. The original research on the fusion of the source disciplines, and in particular the blending of cognitive with artificial intelligence approaches, took place at Carnegie-Mellon University in the 1950s (Simon 1960). This research both defined the start of DSS and also was seminal in the history of artificial intelligence; these fields have to a large degree co-evolved ever since. By the 1970's research groups in DSS were widespread in business schools and electrical engineering departments at universities, in government research labs, and in private companies. Interestingly, ubiquitous computer peripherals such as the mouse originated as part of decision support research efforts.
By the 1980s the research scope for DSS had expanded dramatically, to include research on group-based decision making, on the management of knowledge and documents, to include highly specialized tools such as expert-system shells (tools for building new expert systems by adding only knowledge-based rules), to incorporate hypertext documentation, and towards the construction of distributed multi-user environments for decision making. In the mid-1980s the journal Decision Support Systems began publishing, and was soon followed by other academic journals. The appearance of the World Wide Web in the early 1990's sparked a renewed interest in distributed DSS and in document- and case-libraries that continues in the early twenty-first century.
Outstanding Technical Issues with Decision Support Tools
DSS tools, as described above, integrate data with formal or heuristic models to generate information in support of human decision making. A significant issue facing the builders of these tools is exactly how to define formal models or heuristics; experts make extensive use of tacit knowledge and are notoriously unreliable at reporting how they actually do make decisions (Stefik 1995). If the rules provided by domain experts do not reflect how they actually address decisions, there is little hope that the resulting automated system will perform well in practice.
A second, related, issue is that some systems are by their very nature difficult to assess. Chaotic systems, such as weather patterns, show such extreme sensitivity to initial (or sensed) conditions that long-term prediction and hence decision support is difficult at best. Even worse, many systems cannot be considered in isolation from the decision support tool itself; DSS tools for stock market trading, for example, have fundamentally changed the nature of markets.
Finally, both the DSS tools and the infrastructure on which they operate (typically, computer hardware and software) require periodic maintenance and are subject to failure from outside causes. Over the life of a DSS tool intended to, for example, monitor the electrical power distribution grid, changes to both the tools themselves (the hardware, the operating system, and the code of the tool) and to their greater environment (for example, the dramatic increase in computer viruses in recent history) mean that maintaining a reliable and effective DSS can be a challenge. It cannot be certain that a DSS that performs well now will do so even in the immediate future.
Ethical Issues
Decision support rules and cases by their very nature include values about what is important in a decision-making task. As a result, there are significant ethical issues around their construction and use (see, for example, Meredith and Arnott 2003 for a review of medical ethics issues). By deciding what constitutes efficient use in a planning support system for business, or what constitutes the warning signs of cardiac arrest in an intensive care monitoring system, these tools reflect the values and beliefs of the experts whose knowledge was used to construct the system. Additionally, the social obligation of those who build DSS tools is an issue. On the one hand, these are tools for specific purposes; on the other hand, many social and natural systems are so interrelated that, in choosing to build an isolated and affordable system, many issues will be left unresolved.
The ruling assumption of efforts to build DSS tools is that decision-making is primarily a technical process rather than a political and dialogical one. The bias here is not so much intellectual as informational: It may overestimate the usefulness of information in the decision-making process. Rather than more information, or ever more elaborate displays, people might need more time to reflect upon a problem. Coming to understand another perspective on an issue is a matter of sympathy and open-mindedness, not necessarily information delivery. Delivering detailed information, cases, and suggested courses of action to a single user is opposed to the idea of community-base processes. While placing these issues outside of the scope of a system design might be a useful design decision from a technical position, it is a value-laden judgment.
In fairness, the decision support literature does occasionally recognize that the public needs a better understanding not only of technology but also of science. There is often little appreciation, however, that decision support is an ethical and political process as much as a technical one—or that the flow of information needs to involve the scientist, the engineer, and the public. Exactly how the political process can be engaged for systems that must by their very nature operate in real time is an open question. Certainly the process of knowledge and value capture for such systems could be much more open than is currently the norm.
A second pressing issue regarding DSS tools is the degree to which the data, knowledge, sensors, and results of their integration represent a limitation on individual freedoms and/or an invasion of privacy. DSS tools based on expert-systems approaches actively monitor every credit card transaction made. Semi-automatic face recognition systems are widespread. Radio-frequency identification tags built into price tags on consumer goods allow consumer behavior to be monitored in real-time. Cell-phone records provide not only who a person was speaking to, but where they were at the time. Decision support tools for national security, market research, and strategic planning integrate information, apply rules, and inform decisions that affect human freedom and privacy every day.
Conclusions
DSS tools will only become more common in the future. The widespread reach of Internet connections and the dramatic decrease in the cost of sensors is driving the creation of decision support tools within governments and industries worldwide. It remains to be seen how these systems may impact on human lifestyles, freedoms, and privacy, and whether these tools can continue to evolve to handle the difficult questions facing decision makers in a complex and changing world.
ROBIN M. HARRAP
ROBERT FRODEMAN
SEE ALSO Choice Behavior;Decision Theory;Engineering Method;Information Society;Rational Choice Theory;Science Policy.
BIBLIOGRAPHY
Giarratano, Joseph C., and Gary D. Riley. (2005). Expert Systems: Principles and Programming, 4th ed. Boston: Thomson Course Technology. A focussed introduction to expert systems technology from the authors of the open source expert system tool CLIPS, widely used in decision support systems.
Meredith, Rob, and David Arnott. (2003). "On Ethics and Decision Support Systems Development." In Proceedings of the Seventh Pacific Asia Conference on Information Systems, 10–13 July 2003, ed. Jo Hanisch, Don Falconer, Sam Horrocks, and Matthew Hillier. Adelaide: University of South Australia. An examination of the methods of decision support systems in medical practice.
Shim, J. P.; Merrill Warkentin; James F. Courtney; et al. (2002). "Past, Present, and Future of Decision Support Technology." Decision Support Systems 33(2): 111–126. A systematic overview of the development of decision support systems.
Simon, Herbert A. (1960). The New Science of Management Decision. New York: Harper and Row. An early and highly influential look at the automation of decision making in business from the leading researcher of the era.
Sprague, Ralph H., Jr., and Hugh J. Watson. (1996). Decision Support for Management. Upper Saddle River, NJ: Prentice Hall. A broad overview of decision support systems, emphasizing systems used in business practice for financial and organizational planning.
Stefik, Mark. (1995). Introduction to Knowledge Systems. San Francisco: Morgan Kaufmann. A detailed examination of the human and software problems in constructing software-based knowledge systems.
Turban, Efraim, and Jay E. Aronson. (2001). Decision Support Systems and Intelligent Systems, 6th edition. Englewood Cliffs, Upper Saddle River, NJ: Prentice Hall. A broad overview of decision support research and artificial intelligence approaches that support decision making. Includes extensive case studies from the history of DSS tools.
