Ethical assessment of large language model-generated advisory text on designing the built environment for health
Mohammad Javad Koohsari, Becky P.Y. Loo, Jing Zhao, Jiuling Li, Ying Long, Yi Lu, Koichiro Oka, Andrew T. Kaczynski
Developments in the Built Environment, 27, 101007
Large language models (LLMs) can generate advisory text on modifying built environments to support health. This study examined the ethical properties of a recent LLM generating text on built environments to support health. The prompts covered six health-related pathways in higher-income, lower-income, and mixed-income neighbourhoods. Overall, 180 answers were coded against four ethical criteria. Non-maleficence was satisfied in all answers. Lower-income contexts were rarely offered weaker proposals than higher-income contexts. Reference to collective participation and transparent oversight appeared in 70-90% of answers without a budget constraint, but only 30-50% under one. The LLM more consistently met minimum expectations for harm avoidance and distributive justice than for collective participation and transparent oversight. These findings suggest that current LLM outputs may reproduce some baseline ethical conventions in urban design discourse but are less reliable on procedural concerns. LLM-generated outputs should therefore be interpreted cautiously within existing built environment decision-making processes.
