Fei-Fei Li is a symbolic figure in artificial intelligence and the fight for more diversity in the IT sector. Here is the portrait of the woman who opened the doors of artificial intelligence as we know it today and raised the first questions about the potential risks associated with an unframed and exclusive use.
“If you read all the AI articles and extract all the quoted names, you can guarantee that women rarely appear. For every woman named on artificial intelligence technologies, there are 100 times as many men named.” Fei-Fei Li for Bloomberg in 2016 [1].
Fei-Fei Li was born in 1975 in Beijing, China, and arrived in the United States in 1991 with her parents. From 1995 to 1999 she studied physics, computer science, and engineering sciences at Princeton University and obtained a master’s degree in physics, applied mathematics, and physics for the engineer. She continued her studies at the California Institute of Technology (Caltech) from 2000 to 2005 where she defended a thesis entitled «Visual Recognition: Computational Models and Human Psychophysics». Following her thesis, Fei-Fei Li entered the University of Illinois as an assistant professor in the Department of Computer Science and Electronics, a position she held until 2006 [2].
ImageNet Creation
During her academic career, she realized that it is just as important, if not more, to work on the quality and quantity of learning data as on the artificial intelligence algorithm used. This achievement marks the beginning of his reflection on the development of the ImageNet dataset. As she explains in her TED conference in Vancouver in 2015 [3], if we want to allow computers to learn to recognize objects, places, and situations as a child would, it would take a phenomenal amount of annotated images to train with.
Accepting a position as assistant professor in the Department of Computer Science at Princeton University in 2007, Li met Christiane Fellbaum, a professor and one of the creators of WordNet, a lexical database, from which she built ImageNet [4]. This database is filled with some 1 billion images on the Internet and the cleaning, sorting, and labeling work is carried out by self-employed workers (Amazon Mechanical Turk). In 2009, ImageNet had more than 15 million images in 2200 different categories, was presented at the Conference on Computer Vision and Pattern Recognition (CVPR) in Florida, and made available in open data to the scientific community.
In 2010 the ILSVRC competition (ImageNet Large Scale Visual Recognition Challenge) was launched, it is an annual competition where ImageNet is used to evaluate image processing algorithms on their accuracy for several computer vision tasks [5]. The aim of this competition was both to make a state of the art of computer vision performance and to promote the development of more efficient techniques [6].
In 2009 she was recruited by Stanford University, where she still works today. From 2013 to 2018 she was the director of the Stanford Artificial Intelligence Laboratory (SAIL).
Co-founder of AI4ALL
In 2015, Olga Russakovsky, one of her Ph.D. students, told her about her idea for a program to open the doors of the fields of computer vision, machine learning, deep learning, and cognitive and computational neuroscience to underrepresented people. Fei-Fei Li and Olga Russakovsky, accompanied by Rick Sommer (Executive Director of Pre-University Studies), founded SAILORS, a summer program for high school girls to learn about human-centered AI [7]. In 2017, with the help of several grants, SAILORS became AI4ALL, a national non-profit organization aiming to make AI more diverse and inclusive. From this date, partnerships with different universities are multiplying.
In December 2022 AI4ALL reached more than 10,000 people in each of the 50 states of the United States and around the world. The existence of this type of program is crucial when we know that only 14% of people working in the AI world are women and only 11% are Hispanic or Black. However, in addition to the fact that the economic impact of AI continues to increase, these technologies have a growing impact on populations and a lack of diversity in the scientists behind these technologies can only lead to an increase in the biases associated with them. Fei-Fei Li challenged the US government on this issue, during his intervention in 2018 as a witness for the committees on science, space, and technology: “There’s nothing artificial about artificial intelligence: it’s inspired by people, created by people, and most importantly, it impacts people. […] With the right guide, AI will make life better. But without it, it is bound to widen the wealth gap even further, make technology even more exclusive, and reinforce biases that we have spent generations overcoming. It will be an ethical, philosophical, and humanistic challenge.”
Co-founder of the HAI
This awareness of the need to frame the development of artificial intelligence led her to co-found, in 2019, the Human-centered AI Institute (HAI) at Stanford, where she returns after spending the 2017-2018 period as a vice President and CEO of Artificial Intelligence for Google Cloud. The mission of this institute is to continue AI research, regulations, and practices to improve the human condition [9].
In 2020 she was elected a member of the US National Academy of Engineering.
“I believe [AI] is a powerful technology that can make a difference. But all technology is a double-edged sword,” she said at the 2019 Grace Hopper Celebration (a ceremony to celebrate women in computing) [10].
This desire to improve AI research while preventing its dangers still follows her today.
Through the many conferences in which she participates and the books she publishes, informs the scientific community and the general public about the positive and negative effects that this technology can have and the need to think about its development by and for humans. She met President Biden in June 2023, as a member of the working group dedicated to the establishment of the NAIRR (National Artificial Intelligence Research Resource), a federal resource to make available data, algorithms, and computing resources. In July, a bill was introduced to establish the NAIRR to provide researchers with the resources needed to develop AI safely. In 2023, she was named one of the 100 most influential people in AI [11].
Sources:
[1] Bloomberg Professional Services. (2017, 28 mars). Artificial intelligence has a ‘sea of dudes’ problem | Insights | Bloomberg Professional Services.
https://www.bloomberg.com/professional/blog/artificial-intelligence-sea-dudes-problem/
[2] Fei-Fei Li. https://fr.wikipedia.org/wiki/Fei-Fei_Li
[3] Li, F. (s. d.). Comment apprendre aux ordinateurs à comprendre des images [Vidéo]. TED Talks. https://www.ted.com/talks/fei_fei_li_how_we_re_teaching_computers_to_understand_pictures?language=fr. 2015.
[4] ImageNet. https://fr.wikipedia.org/wiki/ImageNet
[5] ImageNet. https://image-net.org/challenges/LSVRC/
[6] Brownlee, J. (2019, 5 juillet). A gentle introduction to the ImageNet Challenge (ILSVRC).
MachineLearningMastery.com. https://machinelearningmastery.com/introduction-to-the-imagenet-large-scale-visual-recognition-challenge-ilsvrc/
[7] Results - AI4ALL. (2022, 20 décembre). AI4ALL. https://ai-4-all.org/about/results/
[8] House Science, Space, and Technology Committee. (2018, 26 juin). Hearing - AI - With great power comes great responsibility (Event EventID=108474) [Vidéo]. YouTube. https://www.youtube.com/watch?v=_ObbBp5Vo9U
[9] Stanford Institute for Human-Centered Artificial Intelligence. https://hai.stanford.edu/about
[10] GHC19 - Fei Fei Li, talking about who would change AI [Vidéo]. YouTube.
https://www.youtube.com/watch?v=MOxRohTGXt8 (2019, 13 octobre).
[11] Time. (2023, 7 septembre). TIME100 AI. https://time.com/collection/time100-ai/
Written by Loane D.
Comments