AI is hungry for energy. Can it really support climate action?

Much has been written and demonstrated about how artificial intelligence (AI) improves efficiency and saves money across sectors. It helps monitor, identify and reduce energy wastage to lower the carbon footprint of industries.

 But AI itself is an energy guzzler, particularly generative AI (GenAI), which can generate anything from text, images and videos with just a few lines of prompts. Cloud computing has led to the creation of thousands of data centres across the world. The International Energy Agency (IEA) has said that data centres use 1-1.5 per cent of global electricity use. “The combination of rapidly growing size of models and computing demand are likely to outpace strong energy efficiency improvements, resulting in a net growth in total AI-related energy use in the coming years. Although AI itself can help reduce energy use in data centres, the rapid and mainstream adoption of AI chatbots like OpenAI’s ChatGPT and Google Bard are likely to accelerate growth in energy demand for AI,” said IEA in a report.

“5G, the Internet of Things and the metaverse are likely to increase demand for low-latency computing, increasing demand for edge data centres. User devices such as smartphones – increasingly equipped with ML (machine learning) accelerators – are set to increase the use of ML with uncertain effects on overall energy demand.” IEA said. Data centres and data transmission networks are responsible for 1 per cent of energy-related greenhouse gas emissions.

GenAI will place further demand for energy as it depends on data training algorithms. “Recent data from Meta and Google indicates that the training phase only accounts for around 20–40 per cent of overall ML-related energy use, with 60–70 per cent for inference (the application or use of AI models) and up to 10 per cent for model development (experimentation). Google estimates that ML accounted for 10-15 per cent of its total energy use in 2019-2021, growing at a rate comparable with overall energy growth (+20-25 per cent per year over the same period),” said the IEA report. [Meta is the owner of Facebook, Instagram and Whatsapp.]

An article in ‘MIT Technology Review’ cited a study that calculated the energy cost of computing. “Generating images was by far the most energy- and carbon-intensive AI-based task. Generating 1,000 images with a powerful AI model, such as Stable Diffusion XL, is responsible for roughly as much carbon dioxide as driving the equivalent of 4.1 miles in an average gasoline-powered car,” said the article. Generating one image takes as much energy as fully charging a smartphone, it said, quoting a study by researchers at the AI startup Hugging Face and Carnegie Mellon University.

With GenAI coming to millions of mobile phones, the energy needed to generate photos and videos will be immense. The millions of photos and videos taken on mobile devices are stored in data centres run by companies like Microsoft, Google, Apple and Samsung. Every click and save eats up a few units of energy, aggregating to a huge impact on greenhouse gas emissions.

“Significant advances in data centre performance have been made in recent years, but additional government and industry efforts on energy efficiency, RD&D, and decarbonisation of electricity supply and supply chains are necessary to curb energy demand and rapidly reduce emissions over the coming decade to get on track with the Net Zero Emissions by 2050 (NZE) Scenario,” said the IEA report.

Countries like India will have to be mindful of the cost of energy as hundreds of millions of their citizens join the digital mainstream. The solutions lie in using renewable energy while also continuously improving the efficiency of data centres. A deeper effort for a comprehensive understanding of the energy cost of digitising the world is urgently needed. 

This column appeared in Business Standard Newspaper on February 19, 2024.

#artificialintelligence #GenAI #energy #climateaction #datacentre

Climate side of AI: Its thirst for energy is massive (business-standard.com)