ZLAIRE LNGAI 2026 Logics & Argumentation for New-Generation AI

Keynote Speaker

Keynote Speaker

Kees van Berkel

Talk Title

Argument Calculi: What Happens When Defeasible Rules Become Explicit Objects of Reasoning?

Portrait of Kees van Berkel

Biography

Kees van Berkel is Assistant Professor AI Ethics at the Institute for Logic and Computation, TU Wien. His research intersects symbolic AI, philosophy, and ethics, including topics such as logical reasoning with norm systems, defeasible reasoning, formalized metaethics, explanations in AI, and AI auditing methods for value-alignment. He is a member of the Research Ethics Committee of TU Wien, the cluster of excellence Bilateral AI (on ethical AI systems and Explainability) [2], and the Center for Artificial Intelligence and Machine Learning [3].

Abstract

What Happens When Defeasible Rules Become Explicit Objects of Reasoning? This question is answered by Argument Calculi (AC), a proof-theoretic companion to logical argumentation. Over the past decades, logical argumentation has proven to be a unifying approach to the characterization of nonmonotonic logics (i.e., formal defeasible reasoning) in symbolic AI. Through the construction of arguments and counterarguments, logical argumentation keeps defeasible reasoning close to human reasoning practices.

In linking premises and conclusions, defeasible rules (or, warrants/defaults/levers) are used, which however often remain implicit in such arguments. By making defeasible rules explicit objects of reasoning -- yielding arguments such as "given context Δ, the set of defeasible rules Γ are jointly inapplicable" -- Argument Calculi give rise to interesting conceptual as well as technical results.

In this talk, I will introduce the conceptual background of Argument Calculi, a highly modular rule-based proof-theoretic framework for the construction of arguments, and go through a series of examples. I will illustrate how this highly modular framework can capture classes of nonmonotonic logics by exchanging some of the derivation rules employed in these calculi. These calculi are particularly useful for knowledge representation and reasoning tasks. I will go through a series of studies that investigate the application of these calculi, touching upon normative, doxastic, value-based, and legal reasoning, as well as explainability characteristics of these calculi. This talk is mainly based on joint work with Josephine Dik [1], Andrea Sabatini [2], Christian Strasser [3,4], and Zheng Zhou [4].

Selected References

  • [1]Josephine Dik (2026). Reasoning with Values in Norm-based Argumentation. Proceedings of the International Conference on Artificial Intelligence and Law (ICAIL 2026).
  • [2]Sabatini, A. (2026). Proof-search for normative and doxastic reasoning and its use in logical argumentation. In Proceedings of the International Conference on Principles of Knowledge Representation and Reasoning (Vol. 23, No. 1, pp. 633-643).
  • [3]Straßer, C. (2022). Reasoning with and about norms in logical argumentation. In Computational Models of Argument: Proceedings of COMMA 2022 (pp. 332-343).
  • [4]Straßer, C., & Zhou, Z. (2024). Towards an Argumentative Unification of Default Reasoning. In COMMA (pp. 313-324).