Today’s computer vision models are incredibly adept, but because models are often trained on images scraped from the internet, they can reflect biases and stereotypes about places or people.
A decade ago, as a Ph.D. student at Stanford, Olga Russakovsky was part of a team that developed ImageNet, a groundbreaking visual database. Now an associate professor of computer science at Princeton, Russakovsky works to expand computer vision capabilities, while also making these AI models more fair, accurate, and transparent.
Her team’s recent work includes an effort to achieve more complete representation of everyday objects from around the globe. Because these images will be critical for applications like robotics, agriculture, and health care, Russakovsky’s team commissioned photos from around the world. By adding images of objects like houses, toothbrushes, and wheelbarrows, the project seeks to counterbalance image sets that heavily rely on images from North America and Western Europe. Vikram Ramaswamy, who earned his Ph.D. from Princeton last year and is now a lecturer in the computer science department, is the lead author of the project, called Geode.
“We’re part of this giant ecosystem that’s rapidly evolving,” Russakovsky said. “I think we have to really engage with these social questions as we’re building our models and advancing the state of AI. We have to step up.”
For Russakovsky, increasing the diversity of AI creators by lowering barriers goes hand in hand with thinking carefully about the technology’s design and applications. Along with advancing ImageNet, Russakovsky and Stanford professor Fei-Fei Li ’99 (with educator and mathematician Rick Sommer) cofounded AI4ALL, a nonprofit that started in 2015 as a high school summer program at Stanford. AI4ALL has introduced thousands of students to AI technology and associated ethical and policy issues, and offers support for entrepreneurship and career development.
Russakovsky codirects an AI4ALL summer program at Princeton, where high school students from low-income backgrounds learn programming and AI basics and collaborate on research projects. She played a key role in the early 2024 launch of AI4ALL Ignite, a virtual accelerator program that helps undergraduates delve into AI in ways that go beyond their school curricula. With guidance from industry experts, students develop independent projects, prepare for interviews, and receive support to land paid internships in the field.
