Human-In-The-Loop Large Language Models for Decision Support in Modern Power Systems: Capabilities, Risks, and Governance
Year: 2026
Type: article
Abstract: The integration of renewables, electrification, and distributed control systems has accelerated the digitisation of contemporary power systems, increasing operational uncertainty and complicating decision-making in safety-critical environments. While the use of Artificial Intelligence (AI) in the prediction and optimisation of systems is improving, the absence of human intuition, transparency, and accountability in fully automated decision-making is a challenge in power systems. The goal of this paper is to consolidate and structure recent technical literature, the decision science field, and AI governance to articulate the position of large language models (LLMs) in the decision-making aide role within Human in the Loop (HITL) systems. The research analyses LLMs to support decision-making at various levels of the power systems continuum, attempting to make sense of the technical, cognitive, and socio-organisational challenges, and, paradoxically, the value these systems are expected to provide (or justify) to power systems. The findings of the research indicate that such HITL configurations are instrumental in diminishing failure modes of AI systems (e.g., hallucination and automation bias) and accountability erosion in critical energy systems. The research illustrates that the key contribution of LLMs is not in automated control but in collaborative cognitive support systems that enhance human decision-making, particularly in conditions of uncertainty. From a governance perspective, loss underscores auditability, the allocation of clear responsibilities, and the institutional oversight that must be designed for the generative AI technology in question. For example, the paper concludes by describing a future research agenda that has empirically driven research, design of human–AI interfaces, and specific field governance frameworks for the responsible use of large language models in next generation power systems.
Subfield: Social Psychology
Field: Psychology
Domain: Social Sciences
Open Access status: gold
OpenAlex ID: W7128815893