Colloquium Artificial Intelligence - Annet Onnes, University of Utrecht

Titel: User-centric Bayesian Network Construction for Hybrid Intelligence

Abstract:

Bayesian networks (BNs) are known as interpretable models that allow combining uncertain and rule-based knowledge in one representation, by compactly representing a probability distribution using a graph. The manual construction of these types of models, however, requires knowledge from both domain experts, as well as insights from experts in BNs as models. In this talk, I will discuss several of my research projects that have the shared aim to increase the input of domain experts and decrease the liaising required by BN experts in manual BN construction. The resulting construction methodology allows domain experts to construct BNs, with minimal help from modelling experts, to help monitoring AI systems in Hybrid Intelligence settings, where human teammates are the domain experts modelling what desirable collaborative behaviour looks like. By standardising construction methods and relying on knowledge from users, models can be constructed and adjusted to monitor in varying contexts.