Sasha Luccioni, lead climate researcher at AI company Hugging Face, discusses how data storage and machine learning contribute to climate change and energy consumption. She notes that switching from a non-generative AI approach to a generative one can use 30 to 40 times more energy for the same task. Luccioni suggests providing information to allow people to make choices, such as choosing a more energy-efficient model, and advocates for "digital sobriety", questioning the need for new gadgets or AI uses.
Key takeaways:
- The International Energy Agency (IEA) predicts that the energy demand for data centers, cryptocurrency, and artificial intelligence could double by 2026, equal to the electricity used by Japan.
- One of the fastest-growing energy demands is from generative AI, with training large language models like OpenAI’s GPT-3 using nearly 1,300 megawatt-hours (MWh) of electricity, equivalent to the annual consumption of about 130 US homes.
- AI researcher Sasha Luccioni highlights the lack of transparency in AI deployment and the significant increase in energy use when switching from a traditional AI approach to a generative one.
- Luccioni suggests the concept of "digital sobriety" where consumers question the necessity of new gadgets or AI applications, and the development of an Energy Star rating for AI models to promote energy-efficient choices.