Balancing Machine Learning Performance with Energy Consumption for Sustainable Community

Project Description

Artificial intelligence (including machine learning) has a significant role in enhancing digital sustainability by improving efficiency and reducing waste. However, it is crucial to ensure that AI applications are developed in a sustainable manner. The rapid growth of the AI market may result in substantial emissions. This is mainly due to complex AI models that prioritise accuracy over energy efficiency. Surprisingly, training GPT-3, a large language model, consumed more electricity than 100 US homes consume in a year. It is worth questioning whether such high accuracy is necessary for every application. Research is needed to understand the time and effort required to create, train, and operate ML models. By gaining insight into the causes of increased computational complexity while maintaining acceptable accuracy, performance, and security, we can develop efficient and sustainable digital solutions.