vom Scheidt, Ron (2025) When AI Talks About Nature: Ideological Bias in ChatGPT’s Environmental Discourse Across Priming Conditions. Bachelor thesis, Global Responsibility & Leadership (GRL).
|
PDF
2025GRLCapstone-ThesisRvS.pdf Download (3MB) | Preview |
Abstract
Biodiversity conservation is a global imperative, yet debates over how to balance economic and ecological priorities remain deeply polarised. Large language models (LLMs) like ChatGPT now play a significant role in shaping public discourse, raising concerns that their outputs may reinforce ideological divisions through biased or primed responses (Kaneko et al., 2024). While prior research has addressed LLM biases in domains such as gender, race, and politics, there appears to have been no systematic investigation into how prompt priming influences LLM outputs in biodiversity-related discussions yet. This study examines the extent to which ChatGPT-generated responses reflect or amplify political and ideological biases in biodiversity discourse, with a focus on the effects of prompt priming. Using a controlled experimental design, both GPT-4.1 and GPT-4o models were prompted under five ideological conditions. Responses to the validated 24-item Likert-scale Environmental Attitudes Inventory (Milfont & Duckitt, 2010) and corresponding open-ended questions were analysed using a combination of quantitative (ANOVA, Kruskal-Wallis, regression) and linguistic (LIWC-22) methods. Results reveal robust, systematic effects of both priming direction and intensity on model outputs, affecting not only stated attitudes but also linguistic features such as analytic style, emotional tone, and social framing. Furthermore, model architecture influenced the degree and nature of these shifts, with notable differences between GPT-4.1 and GPT-4o. These findings highlight the sensitivity of LLMs to prompt context and underscore the importance of transparency and bias mitigation in their deployment for public-facing environmental communication. The study contributes to ongoing discussions about the ethical and political implications of generative AI in shaping environmental and policy debates.
| Item Type: | Thesis (Bachelor) |
|---|---|
| Name supervisor: | Verkhodanova, V. |
| Date Deposited: | 10 Jun 2025 08:20 |
| Last Modified: | 10 Jun 2025 08:20 |
| URI: | https://campus-fryslan.studenttheses.ub.rug.nl/id/eprint/654 |
Actions (login required)
 |
View Item |