Balancing GreenAI: Efficiency and Impact of LLMs in Climate-Vulnerable Communities
Keywords: LLM, Green AI, climate change, efficiency, environmental racism
TL;DR: Large language models (LLMs) advance NLP but have high environmental costs, disproportionately impacting low- and middle-income countries, emphasizing the need for efficiency, transparency, and accountability in their development.
Abstract: Large language models (LLMs) are revolutionizing natural language processing (NLP), but their creation comes at a significant environmental cost. This research investigates the carbon emissions produced by pre-training BERT-based language models and situates these findings within the broader context of global carbon emissions. Climate change disproportionately affects low- and middle-income countries (LMICs), so we weigh LLMs' impact within the context of these disadvantaged communities. We explore methodologies for estimating the carbon footprint of model training. We contemplate trade-offs between model efficiency and potential bias, considering how such side effects could exacerbate existing inequalities, particularly in LMICs. Furthermore, this research emphasizes the necessity of transparency and accountability in NLP. LLMs should be developed with clear purposes and a focus on both efficiency and mitigating potential harms, particularly in climate-vulnerable communities.
Archival Submission: arxival
Submission Number: 28
Loading