A recent study by nature.com discusses the carbon emissions of AI systems compared to humans in the tasks of writing and illustrating. The study found the use of AI can have a significant impact on reducing the carbon footprint compared to human activities in certain tasks. AI systems, specifically in the tasks of writing and illustration, emit significantly lower amounts of carbon dioxide equivalent (CO2e) compared to their human counterparts.
Optimizing AI Architecture at Adobe
Artificial intelligence (AI) / machine learning (ML) – and especially generative AI – are transformative technologies evolving at incredible speed. As any new technology, AI raises environmental, ethical, and social concerns. Adobe is committed to addressing concerns with accountability, responsibility, and transparency to ensure that the power of AI can be harnessed for the good of technology and society.
Over just a few months, Adobe Firefly, the family of creative generative AI models, became the most successful beta in Adobe's 40-year history. The company recognizes that training and using generative AI products requires a significant amount of energy and is actively seeking new ways to deliver this technology faster and even more responsibly. For example, the training model optimization focuses on minimizing two primary variables: energy consumption and computing time.
Adobe is investing in code optimization, minimizing redundant steps, avoiding unnecessary content generation, and implementing efficient scheduling and batching strategies, to significantly reduce the required training and increase training velocity. These developments are foundational for the journey to decrease the footprint of generative AI. Consistently focussing on energy efficiency and emissions reduction as an integral part of our AI innovations is a high priority.