Differential representation and epistemic inequality : how generative AI systems frame overdevelopment and ecological harm in Malta

Abstract

This study examines how generative artificial intelligence (Gen AI) systems represent overdevelopment and associated ecological harm in Malta. The analysis draws on 96 out puts from ChatGPT, Claude, Gemini, and DeepSeek, coded at the meaning-unit level using a comparative qualitative approach and a hybrid deductive–inductive framework. Representations varied across systems and prompt conditions. Consequence-oriented prompts foregrounded habitat loss, biodiversity decline, and ecosystem degradation, while causal prompts made speculative development pressures, regulatory weaknesses, and construction-driven economic growth more visible. These between-family differences are treated primarily as task-conditioned variation rather than, by themselves, as evidence of epistemic inequality. Actor visibility also varied within comparable prompt conditions. The Planning Authority and environmental NGOs appeared relatively frequently, whereas local councils, EU institutions, and citizens/residents showed more pronounced prompt-specific gaps within the expected-actor framework. The study therefore distinguishes differential representation from epistemic inequality: the former concerns variation in the actors, ex planations, and policy responses made visible, while the latter is used more narrowly for prompt-specific uneven visibility of expected actors beyond what follows from the prompt task. The findings position Gen AI systems as components of epistemic mediation rather than neutral channels of information.