Abstract
Computational agents based upon the belief-desire-intention (BDI) architecture generally use reactive rules to trigger the execution of plans. For various reasons, certain plans might be preferred over others at design time. Most BDI agents platforms use hard-coding these preferences in some form of the static ordering of the reactive rules, but keeping the preferential structure implicit limits script reuse and generalization. This paper proposes an approach to add qualitative preferences over adoption/avoidance of procedural goals into an agent script, building upon the well-known notation of conditional ceteris paribus preference networks (CP-nets). For effective execution, the procedural knowledge and the preferential structure of the agent are mapped in an off-line fashion into a new reactive agent script. This solution contrasts with recent proposals integrating preferences as a rationale in the decision making cycle, and so overriding the reactive nature of BDI agents.
Mostafa Mohajeri Parizi
PhD Candidate
Giovanni Sileno
Assistant Professor
Tom van Engers
Full Professor (FDR)
I conduct research on AI & Law, with a particular focus on normative reasoning. Having a track record in AI & Law research going back to 1983, I have worked both on knowledge-driven as well as data-driven AI approaches.