The Information Technology and Innovation Foundation (ITIF) has published a report challenging the narrative that AI’s energy consumption is dangerously high. The report suggests that such alarmist depictions are frequently misleading and overblown. ITIF argues that concerns like these have arisen with new technologies in the past, citing a 1990s-era Forbes report that claimed half the electric grid would soon be absorbed by the digital-Internet economy. Fast forward today, it is estimated that internet data centres consume only 1-1.5% of global electricity use.
However, ITIF acknowledges the challenge in calculating accurate figures for AI energy consumption and carbon emissions. Factors include CPU processing power, the energy costs of chip manufacturing and cooling, among others. In 2019, researchers estimated that training Google’s BERT model came with about 1,438 pounds of CO2 emissions, roughly equivalent to flying from New York to San Francisco roundtrip. In spite of such alarming figures, ITIF suggests that AI models and hardware are becoming more efficient and will continue to reduce energy usage.
Potential benefits cited by ITIF arguing for AI’s energy-saving potential include slowing improvement rates prompting focus on model efficiency, the rise of more energy-efficient AI chips, and AI’s capacity to eliminate higher carbon-emitting tasks (such as car driving). AI can also enhance the efficiency of utility systems, process climate change data, optimize logistics, and more, thereby reducing CO2 emissions.
While ITIF encourages energy transparency standards for better benchmarking, it warns that excessive regulation may result in AI models becoming less energy efficient. It cites examples of detractors using misleading energy data to hinder the advancement of AI and calls for a more balanced view on AI energy use, citing that alarmist attitudes and false information can persist, often going uncorrected. It concludes that while AI does consume energy, it is not to the extent that it poses a significant threat to the environment.
