Evaluating Sustainable AI

Overview

This research focuses on two key themes: (1) Ensuring that new AI models are scalable and efficient, reducing the environmental impact, and (2) Ensuring that we have the tools available to benchmark the quality of output of these new models. AI models are built for people, and the way we measure quality should therefore be based on the user's expectations. Our goal is to create solutions that enable humans to make better decisions in the enterprise, without harming the environment. Outcomes include new generative AI algorithms that are efficient and scalable, and new evaluation techniques that can be used to benchmark system performance. We are creating evaluation methodologies that accurately measure user satisfaction so that our partners can confidently deploy new solutions that their customers can trust and want to use.