A new study warns that artificial intelligence (AI) systems are producing significant emissions, posing serious environmental risks. Published in Frontiers of Environmental Science & Engineering, the paper highlights how the growing energy demand for training and running complex AI models is worsening emissions. The study points to OpenAI’s GPT-4, which uses 12 times more energy than its predecessor, as an example.
The report also reveals that while training these models uses substantial energy, the real concern lies in operating them. The energy consumed when running AI models is estimated to be 960 times higher than the energy needed for a single training run. This usage suggests that AI’s ongoing operations are an even greater source of emissions.
The research, led by Meng Zhang from Zhejiang University, calls for urgent change. “The exponential growth in AI capabilities mirrors a concerning rise in its environmental impact,” Zhang noted. The study urges the adoption of greener practices and sustainable industry standards.
AI-related emissions could potentially cost the sector over $10 billion annually. To mitigate this, the study recommends governments and regulators create standardised methods for measuring emissions and implement strict regulations to cap emissions.
This comprehensive research aims to empower policymakers with data essential for crafting effective environmental regulations. The findings underscore the critical need for sustainable practices within the AI industry to curb its rapidly expanding carbon footprint.
