Computer Assisted Reasoning for Decision Support
Most approaches to computerized decision support are based on a rationalist perspective that presumes a monolithic decisionmaker who is consciously aware of all relevant information (possibly including probability distributions to describe uncertainties), and who has explicit and consistant values or preferences regarding possible outcomes. Various technique exist for constructing an optimal decision for such idealized decision problems, including most notably the elegant mathematical framework of statistical decision theory.
Many real decision problems, however, pose pragmatic difficulties that clash with this idealized picture. For instance:
- Reasoning about decisions is a social process. Multiple interested parties, with divergent concerns, expectations, values, and expertise to share, must engage in a conversation, which if successful converges on a consensus decision.
- Reasoning involves the discovery of new information. The important knowledge and information that is available is often not fully known to any single individual, and is often not explicitly known to any individual. During the process of reasoning about a problem, new knowledge may be brought forward by members of the community. Some knowledge may be consciously held but only emerges through the stimulus of specific examples.
- Reasoning involves information of differing types. Organizations possess information across a wide range of types, from quantitative data, to qualitative knowledge based on human experience. All of this information may be useful in addressing a complex problem, yet it is difficult to express all of this information in any single mathematical or computational framework.
Decision-making is often political. Various stakeholders to a decision will frequently have divergent beliefs about the probabilities of various uncertainties, and about the values that imply preferences between different outcomes. Thus, there is frequently no possibility of creating consensus subjective probabilities or consensus cost or utility functions that could be used to solve for the optimal decision. Indeed, for any fixed method for computing a decision, participants will be prone to bias their assessments to support the answer they favor.
Classical approaches to decision support suppress these pragmatic issues, and in doing so, they frequently fail to address the central problems facing decisionmakers. In practice, the reasoning process itself needs to find ways to bridge differences among stakeholders, while utilizing all the knowledge that is available to address the problem.
Evolving Logic provides methods and technology that exploit the power of computer modeling to deduce new insights from available information in ways consistant with all of the pragmatic aspects of real decisionmaking. The social process of addressing complex problems produces arguments for or against alternative decisions. Evolving Logic combines the logic that members of the decision-making community use to articulate these arguments with the results of computational experiments that serve as exhibits supporting arguments for or against particular decisions. Evolving Logic's approach to decision support allows the logic of a particular argument to be translated into searches across universes of alternative experiments to provide (or attempt to discover) the crucial computational experiments that support and/or contest key arguments.
In the process of problem solving, people build models that allow them to deduce new insights from their existing knowledge. Every model embeds assumptions about the world, either explicitly as variable parameters, or implicitly in the way the problem has been represented. The conversation that takes place between decisionmakers in the context of these models is frequently about these assumptions. Supporting real world decisionmaking thus requires support for exploring the implications of alternative assumptions made in modeling experiments as needed by the reasoning process. Providing this support is Evolving Logic's mission.
>| return to News, Press and Publications section