Keynote Lectures

Relations, Events, and Explanation
Giancarlo Guizzardi, Free University of Bolzano-Bozen, Italy and University of Twente, Netherlands, Netherlands

Self-Reflective Hybrid Intelligence: Combining Human with Artificial Intelligence and Logic
Catholijn Jonker, Delft University of Technology, Netherlands, Netherlands

Probabilistic Graphical Models: On Reasoning, Learning, and Revision
Rudolf Kruse, Otto von Guericke University Magdeburg, Germany, Germany

Abstract
In Computer Science and, in particular, in Artificial Intelligence, ontologies have been deemed for the past 30 years as artifacts capable of supporting knowledge exchange and reuse, and semantic interoperability. However, if they are to play this role, ontologies must be more than just logical specifications, and ontology engineering must be primarily about Ontological Analysis. Ontological Analysis, in turn, is about employing formal (i.e., domain-independent) theories in the philosophical sense, as well as the analytical tools put forth by Ontology (with a capital ‘O’) to unpack the semantic content of representations. One particular type of Ontological Analysis is Truthmaker Analysis, which, in turn, is strongly connected to the notion of Metaphysical Grounding and Explanation. In this talk, I will (i) present an ontological theory of Relationships and Events, and show that they are the natural truthmakers of Relations. I will then show how this theory can be used to systematically unpack (i.e., explain) the semantic content of domain ontologies, and how this reveals important and subtle aspects of domain modeling that can have a decisive impact on semantic interoperability.

Abstract
By creating forms of human-AI intelligence we aim to support the individual and social wellbeing of people. We illustrate this with examples of Human-AI decision making, and human-robot teamwork. With the aim in mind of creating responsible AI we depart from the position that we aim to enhance or augment human intelligence, not to replace humans. The implication of this is that from the start we have to design for the co-activities that human and AI will undertake and pay special attention of the interdependencies between human activities and those of the AI. We then go to the next level by discussing how such hybrid intelligent systems can perform self-reflection on their performance, and based on that adapt their activities and measurable criteria for behaving responsibly.
Thus we discuss problems and potential solutions of keeping AI and in particular Machine Learning algorithms in bounds.

Abstract
Probabilistic Graphical Models are of high relevance for complex industrial applications. The Markov network approach is one of their most prominent representatives and an important tool to structure uncertain knowledge about high dimensional domains. The decomposition of the underlying high dimensional spaces turns out to be useful to make reasoning in such domains feasible. Compared to conditioning a decomposable model on given evidence, the learning of the structure of the model from data as well as the fusion of several decomposable models is much more complicated. The important belief change operation revision has been almost entirely disregarded in the past, and in this context the problem of inconsistencies is of utmost relevance for real world applications. In this talk it is shown that there are efficient methods for solving such problems. A successful complex industrial application in automotive industry is presented, in which probabilistic graphical models are used for modelling the various belief change operations.