Research

I study whether AI alignment can be evaluated only through agreement with human preferences and judgments, or whether it also requires attention to the moral grounds and principles behind those judgments. Current work examines rationale-level moral alignment in large language models and the boundary between legitimate AI assistance and the delegation of non-transferable human responsibility.

What kinds of AI development or deployment should not be pursued even when they are technically feasible? Through the Moral Red Lines project, I connect technical analysis of frontier systems with ethical and theological accounts of dignity, agency, work, responsibility, relationship and the common good.

Generative AI can improve immediate task performance while changing how people learn to think, remember, create and judge. I study the emerging evidence around cognitive offloading and intellectual deskilling, with particular attention to children, students and educational design that uses AI without surrendering the formation of independent capacities.

As conversational systems become companions, confidants and increasingly agentic partners, the central question is no longer only whether machines can simulate social intelligence. I examine how frictionless and asymmetrical artificial relationships may reshape human expectations of reciprocity, vulnerability, responsibility and genuine encounter.

My technical work develops intelligent systems that adapt to context and resource constraints at runtime. This includes dynamic neural-network adaptation, efficient mobile inference, approximate computing, on-device learning and federated learning, with an emphasis on real-world deployment and meaningful system-level evaluation.

I apply human-centred and resource-aware AI to domains where deployment constraints matter: wearable and smartphone health monitoring, UAV-based precision agriculture, sensing, and digital cultural heritage. Earlier work also included conversational agents, multispectral imaging and digital cultural heritage applications.