Validity at the Forefront: Investigating Threats in Green AI Research

Green AI aims to make artificial intelligence energy-efficient and sustainable. Researchers have formulated new frameworks, methods, models and experiments aimed at the understanding and optimization of greenability in the AI arena. One of the emerging observations is that Green AI research has to cope with challenges, threats and limitations that are not always easy to overcome. In this paper, we investigate how validity threats (VTs) are reported in Green AI research. To this end, we address two research questions. RQ1 examines whether Green AI researchers identify VTs in their paper and what influences this practice. RQ2 categorizes the VTs mentioned in these papers, looking at their types and prevalence. We conclude with a list of takeaways and recommended actions aiming to enhance the understanding and management of VTs in the field.