Modern decision-making increasingly relies on complex algorithmic and predictive systems. Yet as these models transition from theoretical frameworks to societal deployment, they face a critical challenge: ensuring scientific rigour while remaining explainable, socially responsive, and ethically grounded. This session explores the powerful convergence of mathematical statistics, Bayesian computation, spatial modelling, explainable AI (XAI), and the philosophy of AI ethics to address these pressing socio-technical challenges.

Statistics provides the foundational language for formal inference, calibration, and uncertainty quantification. Within this tradition, Bayesian methods offer a robust framework for reasoning under uncertainty and encoding prior domain knowledge into complex hierarchical models. When extended to spatial statistics, these methods account for geographic and socioeconomic context — a necessity, since real-world data and its societal impacts are rarely spatially homogeneous. Yet mathematical rigour alone does not guarantee justice or trust. This session highlights how XAI and actionable AI ethics transform opaque predictive systems into transparent, accountable tools. By bridging predictive performance with interpretable, uncertainty-aware reasoning, the five contributors collectively demonstrate how interdisciplinary synergy can build AI systems that prioritise both scientific excellence and social justice.

The session features five interconnected talks. Gandhi Jafta opens by introducing a framework for Bayesian spatial individual fairness, focusing on uncertainty-calibrated algorithmic auditing. Embedding individual fairness within a spatially explicit posterior inference paradigm, the B-Fair framework moves algorithmic auditing beyond point-estimate fairness metrics toward principled, uncertainty-aware assessments that respect geographic and demographic heterogeneity. Prof. Janet van Niekerk follows with a deep dive into computational feasibility, exploring INLA and Bayesian computation for fairness-relevant spatial models. By leveraging the Integrated Nested Laplace Approximation, she demonstrates how large-scale spatial fairness models can be estimated rigorously and efficiently. Shifting focus to structural inputs, Prof. Inger Fabris-Rotelli examines the critical determinants that shape spatial models themselves, interrogating how choices in spatial autocorrelation structure, covariate specification, and boundary definition propagate into downstream fairness assessments.

The session then pivots toward human-centred interpretation and policy. Dr Iena Derks presents a taxonomy of explainable Bayesian networks in socio-technical decision-making, demonstrating how structured explanation — spanning model transparency, reasoning traceability, and decision justification — generates the trust necessary for responsible deployment in high-stakes domains. Finally, Prof. Emma Ruttkamp-Bloem grounds the session's mathematical contributions in foundational philosophy, addressing the quest for actionable AI ethics and the imperative to move abstract principles into concrete, virtue-guided practice.

Collectively, this session embodies the unifying spirit of the ISI by bridging statistical computation (INLA), applied statistical methodology (algorithmic auditing, Bayesian networks), and foundational ethical theory. By addressing how these disciplines intersect to protect the public interest, the session offers broad interdisciplinary appeal to spatial statisticians, data scientists, AI researchers, and policymakers alike.