Exploring AI’s Evolution: Fei-Fei Li’s Quest for Active Vision in Virtual Worlds

AI makes virtual worlds more like our own

Fei-Fei, then a Princeton University computer scientist, invented in 2009 a data collection that would revolutionize artificial intelligence. The data set, called ImageNet, included millions of images with labels that trained sophisticated machine-learning algorithms to recognize objects in pictures. In 2015, the machines were able to recognize more than humans. Li started looking for another \”North Star\” which would push AI in a new direction.

She was inspired by the Cambrian Explosion, 530 million-years ago, when a number of land-dwelling species first appeared. According to a popular theory, the explosion of new species is partly due to the development of eyes which could see their environment for the first. Li realized vision is not a separate phenomenon in animals, but rather \”deeply integrated into a holistic body\” that must move, navigate, adapt, and survive in an ever-changing environment. \”That’s what made it very natural for my pivot to a more active AI vision.\”

Li’s current work is centered on AI agents who can not only accept static images in a data set, but also move and interact with their environment through simulations of virtual three-dimensional worlds.