UKRI AI CDT in Dependable and Deployable AI for Robotics (CDT-D2AIR)

Jessica Ciupa

Project
Neuro-symbolic AI Applications in Ethical and Moral Dilemmas: Constraints and Verification

Summary
This research project guided by Dr. Vaishak Belle and Dr. Katya Komendantskaya investigates the design, implementation, and verification of neuro-symbolic AI frameworks for ethical and moral decision-making in safety-critical autonomous systems. Spanning reinforcement learning, formal verification, computational ethics, and moral psychology, the work develops principled methods for encoding ethical constraints, drawn from deontological, consequentialist, and virtue-based traditions, into interpretable AI architectures, including Reward Machines, probabilistic logic programming, and neural network controllers. A distinctive thread runs through the project: the need to verify that autonomous agents remain aligned with human moral preferences, whether through formal robustness properties, neuro-symbolic verification pipelines, or cross-cultural social choice modelling. Alongside these technical contributions, the project critically examines the human stakes of AI-mediated moral decision-making, exploring how repeated delegation of difficult choices to autonomous agents may erode moral identity and the impulse to self-author decisions, and probing LLM prosociality through species-neutral behavioural paradigms borrowed from non-human primate research. Together, these threads constitute a coherent interdisciplinary programme integrating formal methods, moral philosophy, empirical psychology, and machine learning to build autonomous systems that are not only ethically reliable but whose alignment with human values can be rigorously demonstrated.

Lead Institution
School of Informatics, University of Edinburgh

Supervisors
Vaishak Belle (University of Edinburgh), Ekaterina Komendantskaya (Heriot-Watt University)