Speaker
Description
This study compares an AI-assisted workflow with a researcher-applied manual policymaker-style workflow, rather than an independently replicated policymaker process, for structuring scenario-based decisions on low-carbon material uptake in Stavanger affordable housing. An exploratory matched-process case study applied both modes to the same seven-document corpus, housing archetype, three pathways, seven criteria, scoring anchors and AHP-based weights. The AI-assisted workflow was developed from a modular prototype to a final integrated evidence-to-policy-brief process linking source, evidence, scoring, scenario and output records. Structural and computational checks confirmed workflow completion and arithmetic consistency, but not semantic accuracy or policy validity. Both workflows ranked the balanced phased transition first (77.88 manual; 77.90 AI-assisted). However, exact agreement occurred in only 47.6% of the 21 criterion-level judgements, and the full low-carbon pathway differed by 15.46 points. The findings show rank convergence under a shared analytical architecture alongside material differences in criterion judgement. The integrated workflow improved traceability and separated generative interpretation from deterministic aggregation, but the study does not establish GenAI superiority, reproducibility or policy validity. Independent analysts, stakeholder-derived preferences and repeated fixed-configuration testing are required for stronger validation.
| Topic Area | Computational Intelligence for Power Systems |
|---|