Research: Guiding Human-Centered AI

AI systems use considerable amounts of power and can generate alarmingly high volumes of carbon emissions. Accurate reporting is essential for understanding the potential climate impact of research projects. A team from Stanford, Facebook AI Research, and McGill University devised a framework for tracking real-time energy and carbon usage, along with a leaderboard to motivate engineers to think about the environmental impact of their work. Stanford computer science PhD student Peter Henderson served as lead author of the study, working closely with Stanford professors Dan Jurafsky and Emma Brunskill. By making it easy to measure the carbon footprint of AI, this group hopes to drive more commitments to energy-efficient algorithms.

Advances in machine learning and contactless sensors have given rise to ambient intelligence—physical spaces that are sensitive and responsive to the presence of humans. Fei-Fei Li, HAI Co-director, Arnold Milstein, Professor of Medicine and Director of Stanford’s Clinical Excellence Research Center; and Albert Haque, Department of Computer Science graduate student, co-authored a study showing how this technology could improve patient outcomes in healthcare settings. They concluded that thoughtful use of emerging data science and AI technology would enable a better understanding of the complex interplay between physical environments and health-critical human behaviors. Their findings were published in the September 2020 issue of Nature.

Computational neuroscientist and Assistant Professor of Psychology and Computer Science Daniel Yamins and Professor of Human Biology Michael Frank have been collaborating on developing curiosity-driven systems that mimic the way children learn, with the ultimate goal of making more effective diagnoses for a variety of developmental disorders. In July 2020, Yamins published “Active World Model Learning with Progress Curiosity,” which provides a foundation for their research into children’s curiosity.