How to Make Artificial Intelligence More Environmentally Sustainable (www.infoq.com)

At the recent GOTO Copenhagen conference, Charles Humble highlighted the environmental impact of artificial intelligence (AI), revealing that leading tech giants like Microsoft and Google have significantly increased their carbon emissions due to rising electricity consumption from AI. Humble emphasized that the IT industry is responsible for 2-4% of global emissions, comparable to the aviation sector, but unlike aviation, tech companies have yet to implement substantial measures to reduce their environmental footprint. He urged a shift in focus towards "Green AI," recommending the use of smaller, pre-trained models, and deploying compute resources in regions with abundant low-carbon electricity. Humble outlined actionable strategies to mitigate the carbon footprint of AI, including demand shaping, where workloads are adjusted based on the carbon intensity of electricity sources, and optimizing model training processes. He stressed that sustainability should be prioritized alongside traditional concerns like cost and performance when designing AI systems. By adopting techniques such as using renewable energy for training, leveraging cloud services from green data centers, and employing demand-driven computing, AI practitioners can potentially cut carbon emissions by up to 99%. His insights challenge the AI/ML community to rethink how systems are built and operated to foster a more sustainable future.