New DFG-funded project "CASPER – Conditionals and ASP for Expert Reasoning"
The overall goal of this project is to devise a hybrid knowledge representation framework based on answer set programming (ASP) and plausible conditionals with ranking semantics (OCF-conditionals) in which technical (default) knowledge and plausible expert knowledge can be combined to support creative technical tasks like warehouse planning in a professionally and cognitively adequate way. Due to the need for involving humans in many places, logistics is an ideal (academic) application domain that raises multiple challenges regarding the modelling of domain-specific knowledge in highly dynamic environments which are typical and relevant also for many other application areas. Both ASP and OCFs (ordinal conditional functions) are well-known for their good logical properties and their suitability to model human knowledge and reasoning. Moreover, ASP is one of the most successful implementations of nonmonotonic and uncertain reasoning. We advance the state of the art of the OCF methodology by developing network-based inference algorithms that improve the efficiency of this elegant reasoning framework. Having both approaches in one framework allows for taking characteristics of one framework into account suitably in the respective other so that synergies arise, e.g., by assessing the plausibility of ASP solutions via OCF-conditionals, or taking ASP semantics as a base logic for OCF-conditionals. Jointly they will provide an expressive formal framework that advances significantly the realm of methods of today‘s knowledge representation in Artificial Intelligence. Moreover, we take characteristics of human reasoning into account when setting up the knowledge bases to improve the cognitive adequacy of the knowledge modelling. To ensure professional adequacy, an interactive modelling environment is provided so that experts can be involved in many stages of the modelling process. A demonstrator system for finding suitable logistic layouts and configurations of warehouses in a holistic way is implemented as a proof of the interactive concepts to be developed in this project: enabling the user to customize knowledge in the hybrid framework, making basics of knowledge representation and reasoning processes intelligible by explanations, resolving conflicts in an informative way, and controlling the quality of solutions according to professional standards.
Our project addresses several fundamental research issues in knowledge representation, proposing in particular innovative approaches to hybrid reasoning, plausible network-based inference, and cognitive modelling. Embedding this in a comprehensive knowledge representation task for an academic application, also involving novel types of explanations and interactions ensuring robust consistency and a professionally and cognitively adequate modelling, allows for showing the relevance of the methods and techniques to be developed for advancing explainable and intelligible AI for real-world applications.
Project leaders Prof. Dr. Gabriele Kern-Isberner (Professorin im Ruhestand, TU Dortmund)
Staff members Andre Thevapalan & Marco Wilhelm (beide TU Dortmund)
Cooperation partner for the logistics part Prof. Dr. Michael ten Hompel (TU Dortmund)