n o uc ion
Abduction is inference to the best explanation[21, 10]. This reasoning method is very useful in accomplishing a variety of tasks such as diagnosis, identification, and recognition[23, 25, 28, 27, 29, 26, 32, 2, 12]. However, it has also been used for solving theory formation and theory decision problems[33, 9, 19]
We have developed a domain independent strategy for solving abductive problems. To further test out our abductive strategy, we have chosen to implement a theory decision making system. This system attempts to decide which theory is better, creationism or evolution. It attempts to explain a variety of facts and findings about life on Earth in terms of one of the two theories. This paper documents the abductive strategy used, the system built, and results of running the system.
i c - o in n n n uc i
Peirce is a tool for constructing abductive problem solving systems. Peirce is a tool in the
Integrated Generic Task Toolset, from which complex problem solvers can be constructed.[22,
18, 13, 17] The task for an abductive problem solver is to construct a composite (multipart)
hypothesis in order to explain the appearance (or absence) of findings presented to the
problem solver. A Peirce abducer attempts to construct the best" explanation based on a
set of plausible hypotheses.
Peirce is a shell which requires domain specific knowledge. Past domains which have been implemented by using the Peirce strategy have included blood bank antibody identification, speech recognition[7, 6, 8], and legal reasoning.
nowle ge Re uire for eirce
To use Peirce, specific types of domain knowledge must be supplied. Peirce must be given the set of findings that need explaining. Peirce must also be given a set of plausible hypotheses to work with (these hypotheses are generated by some other problem solver).1
The hypotheses must have been given some score in terms of plausibility (the tool RA, in the Toolset is suitable for scoring hypotheses based on the current situation or data present). Only plausible hypotheses are used by Peirce. In addition to some plausibility score, each hypothesis must be given information about what findings it could explain if it were true.
Another form of knowledge about a hypothesis is what interactions it might have with other hypotheses. These interactions come in the form of expectations (where one hypothesis would expect to see another hypothesis appear), implications (where one hypothesis implies another whether causally or deductively) and incompatibilities (where one hypothesis is mutually exclusive with another hypothesis). Hypothesis interaction knowledge is optional
1The Integrated eneric Task Toolset o ers tools to build agents which can generate hypotheses by classification and other methods.