The Epistemic and Practical Reasoner


Introduction

The Epistemic and Practical Reasoner is the result of my MSc graduation project in Cognitive Artificial Intelligence at Utrecht University, The Netherlands. The program was demonstrated at the Second International Conference on Computational Models of Argument (COMMA'08), held in Toulouse, France, from 28 to 30 May 2008.

Abstract

The Epistemic and Practical Reasoner is an implementation of an argument-based practical reasoning system. Based on Dung's abstract argumentation system [4] instantiated with arguments in the form of trees of chained defeasible inferences, it combines some recent theories of argument-based practical reasoning.

Prakken [6] claims that in certain contexts, reasoning about beliefs is sceptical while reasoning about action is credulous. This idea is formalised in e-p-semantics, which combines grounded sceptical semantics for epistemic reasoning and preferred credulous semantics for practical reasoning.

Bench-Capon and Prakken [2] present a formalism that combines the practical syllogism with Prakken's [5] mechanism of accrual of arguments. The practical syllogism has two forms: a positive one (if action a realises p, and p is desired, then action a is also desired) and a negative one (if action a prevents q, and q is desired, then action a is not desired). Together, they deal with possible negative side-effects of actions (the two example arguments rebut each other). To deal with alternative ways to reach a goal, a new type of conflict between arguments is defined: if both action a and action b realise p and p is desired, the two arguments for desiring a and desiring b are conflicting. Accrual combines different arguments with the same conclusion into one argument. One application of accrual in practical reasoning is the accrual of positive (wanted) effects of a particular action on the one hand, and negative (unwanted) effects on the other hand, but in general accrual can be applied to any kind of defeasible reasoning. To ensure that the largest possible accrual is used, non-maximal accruals are undercut.

The program is written in Java 6 and has a graphical user interface. Input to the program consists of a belief base of formulas, stated in a propositional language with a single modality D standing for desire, and a query formula. The user can select different options, such as the semantics to use (grounded, preferred credulous or e-p), and whether the practical syllogism and accrual are to be used. After parsing the input, the program constructs arguments from the belief base and engages in an argument game to determine the status of arguments for the query formula. Three games are implemented: the G-game for grounded sceptical semantics, the P-game for preferred credulous semantics, and the GP-game for e-p-semantics. The resulting network of arguments and moves is presented in two ways: an XML file that conforms to an extended version of ASPIC's reification [1] of the Argument Interchange Format proposed by Chesñevar et al. [3], and a graph representation for human readability.

References

[1] ASPIC AIFXML Schemas. http://aspic.acl.icnet.uk/.
[2] Trevor J.M. Bench-Capon and Henry Prakken. Justifying actions by accruing arguments. In P.E. Dunne and T.J.M. Bench-Capon, editors, Proceedings of the First International Conference on Computational Models of Argument (COMMA06), number 144 in Frontiers in Artificial Intelligence and Applications, pages 247-258. IOS Press, 2006.
[3] Carlos Chesñevar, Jarred McGinnis, Sanjay Modgil, Iyad Rahwan, Chris Reed, Guillermo Simari, Matthew South, Gerard Vreeswijk, and Steven Willmott. Towards an argument interchange format. The Knowledge Engineering Review, 21(4):293-316, 2006.
[4] Phan Minh Dung. On the acceptability of arguments and its fundamental role in nonmonotonic reasoning, logic programming and n-person games. Artificial Intelligence, 77:321-357, 1995.
[5] Henry Prakken. A study of accrual of arguments, with applications to evidential reasoning. In Proceedings of the Tenth International Conference on Artificial Intelligence and Law, pages 85-94, New York, 2005. ACM Press.
[6] Henry Prakken. Combining sceptical epistemic reasoning with credulous practical reasoning. In P.E. Dunne and T.J.M. Bench-Capon, editors, Proceedings of the First International Conference on Computational Models of Argument (COMMA06), number 144 in Frontiers in Artificial Intelligence and Applications, pages 311-322. IOS Press, 2006. (Corrected version, May 14, 2007).