Artificial intelligence and environmental sustainability: A balanced path to innovation and responsibility

Artificial Intelligence (AI) is a transforming technology which has measurable impacts on environmental sustainability. This study provides an overall evaluation of AI’s double environmental impact, both positive and negative, through combining a targeted literature review with a benchmarking environmental assessment, using Life Cycle Assessment (LCA), to evaluate the energy consumption, water withdrawals and greenhouse gas emissions associated with the key players in AI and their respective life cycle stages. For example, training a large model such as GPT-3 requires 1,287 MWh of electricity and produces 552 metric tons of CO2e. Since 2016, Google’s AI infrastructure has increased its water withdrawals by over 250% to 8.65 billion gallons per year. While AI can theoretically prevent 20% of energy losses and reduce fuel use by 15%, AI models can exacerbate resource demands. From 2020 to 2023, Microsoft’s energy use increased by 29.1%, and GHG emissions increased by four million metric tons mainly due to hardware production and supply chains. Mitigation needs more than operational fixes. This study proposes a systematic framework to ensure environmental accountability and sustainable AI growth.