Gunter Bombaerts

Gunter Bombaerts

Work on morality in large language models (LLMs) has progressed via constitutional AI, reinforcement learning from human feedback (RLHF) and systematic benchmarking, yet it still lacks tools to connect internal moral representations to regulatory obligations, to design cultural plurality across the full development stack, and to monitor how moral properties drift over the lifecycle of a deployed system. These difficulties reflect a shared root. Morality is installed in a model at training time. I propose instead a morality-as-a-system framework, grounded in Niklas Luhmann's social systems theory, that treats LLM morality as a dynamic, emergent property of a sociotechnical system. Moral behaviour in a deployed LLM is not fixed at training. It is continuously reproduced through interactions among seven structurally coupled components spanning the neural substrate, training data, alignment procedures, system prompts, moderation, runtime dynamics, and user interface. This is a conceptual framework paper, not an empirical study. It philosophically reframes three known challenges, the interpretability-governance gap, the cross-component plurality problem, and the absence of lifecycle monitoring, as structural coupling failures that the installation paradigm cannot diagnose. For technical researchers, it explores three illustrative hypotheses about cross-component representational inconsistency, representation-level drift as an early safety signal, and the governance advantage of lifecycle monitoring. For philosophers and governance specialists, it offers a vocabulary for specifying substrate-level monitoring obligations within existing governance frameworks. The morality-as-a-system framework does not displace elements such as constitutional AI or RLHF it embeds them within a larger temporal and structural account and specifies the additional infrastructure those methods require.

Gunter Bombaerts, Bram Delisse, Uzay Kaymak

Large language models (LLMs) increasingly mediate human decision-making and behaviour. Ensuring LLM processing of moral meaning therefore has become a critical challenge. Current approaches rely predominantly on bottom-up methods such as fine-tuning and reinforcement learning from human feedback. We propose a fundamentally different approach: embedding moral meaning processing directly into the architectural mechanisms and frameworks of transformer-based models through top-down design principles. We first sketch a framework that conceptualizes attention as a dynamic interface mediating between structure and processing, contrasting with existing linear attention frameworks in psychology. We start from established biological-artificial attention analogies in neural architecture design to improve cognitive processing. We extend this analysis to moral processing, using Iris Murdoch's theory of loving attention (sustained, just observation that enables moral transformation by reseeing others with clarity and compassion) to philosophically discuss functional analogies between human and LLM moral processing. We formulate and evaluate potentially promising technical operationalizations to embed morality in LLM architectures and frameworks. We acknowledge the limitations of our exploration and give three key contributions. (1) We conceptualize attention as a dynamic system mechanism mediating between structure and processing. (2) Drawing on the Murdoch notion of loving attention, we outline technical pathways for embedding morality in LLMs, through modified training objectives, runtime weight adjustments, and architectural refinements to attention. (3) We argue that integrating morality into architectures and frameworks complements external, constraint-based methods. We conclude with a call for collaboration between transformer designers and philosophers engaged in AI ethics.