SSA 2026

Formal argumentation provides powerful models for reasoning with conflicting information and plays an increasingly important role in Artificial Intelligence. While much research has focused on the formal properties of argumentation frameworks and semantics, an equally important question is how these models relate to human reasoning.

This tutorial introduces empirical cognitive studies of formal argumentation, focusing on the relationship between normative models of argument evaluation and descriptive accounts of human judgment. After a brief overview of abstract and gradual argumentation, we discuss argumentation principles, experimental methodologies, and recent findings on how people assess the strength and acceptability of arguments. We also examine factors that influence compliance with normative principles and discuss the implications of these results for the design of cognitively plausible and human-centered AI systems.

The tutorial is intended for students and researchers interested in the intersection of argumentation theory, cognitive science, and artificial intelligence.

Machine learning and knowledge representation and reasoning offer complementary strengths: while machine learning methods are flexible, data-driven, and often scale well to complex real-world settings, they typically provide limited formal guarantees and can be difficult to interpret. Knowledge representation and reasoning approaches, in contrast, offer explicit structure, formal semantics, and transparent reasoning mechanisms, but may face challenges regarding scalability, and adaptation to real-world application settings.

This tutorial will discuss how these two perspectives can be combined, using computational argumentation as a representative example of a knowledge representation and reasoning paradigm. We will inspect different ways in which the two areas can interact; this includes, for instance, using machine learning to support or approximate reasoning tasks in argumentation, using argumentation-based methods to explain or structure machine learning outputs, and learning heuristics or components for argumentation systems.

The tutorial is intended to provide an accessible overview of the area, highlighting key ideas, existing lines of work, and open challenges.

In Part 1 I briefly review the main philosophical approaches to truth, by way of then developing an approach inspired by the work of the American pragmatist and logician Charles S. Peirce. Specifically, I suggest that truth amounts to a normative injunction to inquire, so as to resolve uncertainty, in view of the instrumental utility that accrues from resolving uncertainty. Inquiry is to be understood in a broad sense, as norm governed inferential processes that are inherently dialectical, and that offer prescriptions for individual agent reasoning and multiple agents engaged in collaborative, distributed reasoning. Moreover, I suggest that the truth norm's injunction to engage in dialectical inquiry, is constitutive of recent predictive processing models of cognition. These essentially Bayesian models provide a unifying account of perception and action. They describe the brain's main function as effectively resolving uncertainty by seeking to minimise the 'errors' that arise when brain-generated predictions as to the sense data it expects at any given moment, conflict with incoming sense data.

Then, in Part 2 I suggest that we look to the study of non-monotonic logics when articulating the inferential norms that govern these processes of dialectical inquiry; in particular argumentation-based formalisations of non-monotonic reasoning characterised in terms of the dialectical exchange of arguments. Given the account of inquiry articulated in Part 1, we are then faced with two challenges. Firstly, integration of these dialectical formalisations of non-monotonic logics with probabilistic (in particular Bayesian) inference; one that is suited to modelling predictive processing account of cognition. Secondly, reformulating argumentative characterisations of individual agent non-monotonic reasoning in terms of normative constraints on speech acts, so as to obtain communicative accounts of distributed non-monotonic reasoning. I focus on progress towards meeting this second challenge, and argue for the importance of this 'dialogical turn' given some of the ethical challenges raised by contemporary developments in Artificial Intelligence.