A Design Science and Empirical Study: Developing and Evaluating a Fine-Tuned Small Language Model for Climate Misinformation Detection

This research proposal outlines a study to address the growing issue of climate misinformation, which poses a significant threat to global sustainability efforts. The proposed research aims to leverage a pragmatic and resource-efficient approach to artificial intelligence by developing and evaluating a fine-tuned Small Language Model (SLM) for the purpose of detecting and classifying climate-related disinformation. The research is motivated by the documented shortcomings of Large Language Models (LLMs), including their high computational cost, susceptibility to bias, and a widespread gap between market hype and real-world performance. By adopting a combined Design Science and Empirical research methodology, this study will first create a specialized SLM artifact and a high-quality, domain-specific dataset. It will then empirically evaluate the model's efficacy by comparing its performance against a general-purpose LLM. The expected contributions include a practical tool to assist journalists and researchers, a novel methodology for applying resourceefficient AI to a critical social problem, and a robust discussion of the ethical implications of using smaller, more trustworthy models as a counter-measure to the current trend of large-scale, opaque AI systems.