We have recently taken part in the OSETAI workshop which was part of the JOWO 2026 Conference in Vitória, Brasil.
OSETAI stands for Ontology and Semantics for Explainable and Trusworthy AI and it is an academic and professional workshop dedicated to the formal representation, characterization, evaluation, and auditing of explanations and trust in artificial intelligence systems.
The event was co-organised by our partners in the Horizon Europe project DigiPass CSA.
Explainability and the documentation of explanations are now increasingly considered key technologies to making modelling useful in practice. In fact, this has now become a key area of innovation, since substantial investments were made in AI models that are not explainable and are generally not trustworthy. Making such models more ready for use in critical environments would require providing explanations or grounds for trust; this will be possible in some cases, not in all, but it can be observed that funds and investments are increasingly directed toward that aim.
At the same time, the traditional, inherently interpretable modelling and simulation techniques, such as physics-based modelling, can now reassert themselves: A major argument in their favour has always been that they are explainable and trustworthy, especially for extrapolation outside the range of data that has been used for model parameterization.
A related epistemic issue concerns the explanations themselves, ranging from accounts and informational descriptions to causal explanations and, ultimately, judgments that support responsible decision making and trust. This includes ontological distinctions between validating individual components, claims, or behaviors and validating models or systems as coherent wholes, together with the semantics of how evidence, explanations, and trust propagate across these levels.
Metadata standardization and the alignment and harmonization of metadata standards for this type of annotation was among the workshop's themes. On the less technical, more foundational side, the discussion centered on what constitutes an explanation or interpretation, what is needed to make something trustworthy, and how these concepts can be characterized, and evaluated, given explicit ontological commitments, and connected to foundational and middle-level ontological frameworks.
Read the paper from BatCAT's very own Martin Thomas Horsch
PACE: A Neuro-Symbolic Framework for Plausible and Actionable Counterfactual Explanations