The Future of AI is anybody’s guess, but I can safely say that Fei-Fei Li will be remembered as “the godmother of AI”. In her recently published memoirs — The Worlds I See: Curiosity, Exploration, and Discovery at the Dawn of AI — Dr. Li shares the inspiring journey from her humble childhood in China, to her successful accomplishments in academia and the AI industry.
The book opens with Dr. Li narrating the hours before delivering her testimony before the US House Committee on Science, Space and Technology in 2018. In what she describes as a career-first event, Dr. Li was one of the expert witnesses invited to address the Congress on the topic of artificial intelligence. As she reflects on the role of AI in our society and its future, the narrative sets the stage for Fei-Fei Li’s remarkable story.
It matters what motivates the development of AI, in both science and industry, and I believe that motivation must explicitly center on human benefit.
| Li, Fei-Fei. The Worlds I See (p. 7). Flatiron Books. Kindle Edition.
In the following chapters, Dr. Li goes on to describe her childhood and early teenage years in China where she found her love for Physics. At the age of 15, Fei-Fei and her mother moved to the U.S., joining her father who had moved three years prior. Settling in New Jersey, the family endured many hardships that became even harsher owing to a heart condition affecting Fei-Fei’s mother’s health.
With all these struggles going on, Fei-Fei met Mr. Sabella, her high-school math teacher, who would become a mentor and a life-long friend. Overcoming the cultural shock and with renewed confidence, Fei-Fei managed to get accepted into Princeton University, the same place that was once home to her hero Albert Einstein, also an immigrant.
After earning her physics degree from Princeton, Fei-Fei moved to California where she would complete her master’s and PhD at the California Institute of Technology (Caltech). Focusing her research on the intersection between neuroscience and computer vision, was a rather unconventional choice at the time. In 2003, Dr Li. published her first research paper (A Bayesian Approach to Unsupervised One-Shot Learning of Object Categories) earning her a spot in the International Conference on Computer Vision (ICCV 2003), considered one of the top conferences in computer vision. Dr. Li recollects how researchers at the time seemed to be chiefly focused on the development of new learning algorithms but without proper consideration of the training data.
It was this work that led young Fei-Fei and her advisor, Pietro Perona, to the realization that the dataset used to train their algorithm played a crucial part in their success. They embarked on the arduous task of expanding their dataset to 100 image classes. For reference, the largest dataset ever assembled at Caltech at that time contained 7 image classes. So, the task of gathering, labeling, and organizing such a dataset on their own was a big deal. Finished in 2004, the dataset, officially dubbed “Caltech 101” contained 9000 labeled images spread across 101 categories. It was the humble beginnings of what would later be the ImageNet project.
In parallel to her story, Dr. Li takes a dive into the origins of AI in the 50’s and 60’s, revisiting the ups and downs of the field. Throughout the book we learn the historical background connected to Dr. Li’s work, putting in perspective the significance of her achievements.
Dr. Li recounts the dramatic story of how ImageNet almost didn’t happen. In 2007 heading her research lab at Princeton, Dr. Fei-Fei Li embarked on a daunting task: creating the largest ever data set in machine learning. A work that would take more than 2 years to complete. The first version of ImageNet was completed in mid-2009, with Dr. Li and her team now at Stanford University.
The ImageNet database and the “ImageNet Large Scale Visual Recognition Challenge” helped propel the deep learning revolution, making it one of the most significant contributions to modern AI.
In the following years, Dr. Li served as a doctoral advisor for notable students at Stanford. A Canadian student by the name of Andrej Karpathy under Dr. Li’s wing, did pioneering work on image captioning systems, bridging computer vision and natural language processing through deep learning. (Karpathy later became one of OpenAI’s founding members, as well as director of AI at Tesla).
Influenced by her ailing mother’s frequent visits to hospitals, Fei-Fei set to work on AI systems for healthcare. Teaming with colleagues from Stanford Medical School, they produced “ambient intelligence”. An AI system meant to help clinicians and caregivers in their daily tasks. Dr. Li describes this work as “the most scientifically demanding project” her lab had ever attempted.
In 2016 Dr. Fei-Fei Li took a sabbatical from Stanford and joined Google Cloud as chief scientist of AI. This was Dr. Li’s first job outside of academia and she used this opportunity to help democratize the use of AI technologies. AI was no longer a niche discipline within computer science departments, it was now adopted by tech giants such as Google and Microsoft investing millions in research and product development. It was clear that AI was rapidly becoming a “privilege”.
Back at Stanford since 2018, Fei-Fei Li’s work has focused on the idea of “Human-centered AI”. Aiming to become a “hub for cross-disciplinary collaboration”, Dr. Li established the Stanford Institute for Human-Centered Artificial Intelligence (Stanford HAI).
In the final chapters, Dr. Li recounts the advancements in the field since the ImageNet days to LLMs and Generative AI. She concludes with a reflection on her future endeavors and aspirations moving forward, what she calls her “North Stars”.
The book’s engaging storytelling provides a unique perspective on the struggles and personal sacrifices that have made Fei-Fei Li a successful scientist. A highly recommended read for AI practitioners and enthusiasts, but also those interested in science in general.
Thanks for reading! I hope you find this review informative, feel free to share it and leave me your comments.
For the interested reader, here is a list of some of the research papers mentioned in the book:
[1] FF. Li, R. Fergus, P. Perona. A Bayesian Approach to Unsupervised One-Shot Learning of Object Categories. In ICCV, 2003. https://lear.inrialpes.fr/people/triggs/events/iccv03/cdrom/iccv03/1134_fei-fei.pdf
[2] A. Karpathy, A. Joulin, FF. Li. Deep Fragment Embeddings for Bidirectional Image Sentence Mapping. In NeurIPS, 2014. https://proceedings.neurips.cc/paper_files/paper/2014/file/84d2004bf28a2095230e8e14993d398d-Paper.pdf
[3] T. Gebru, J. Krause, Y. Wang, D. Chen, J. Deng, E. L. Aiden, FF. Li. Using deep learning and Google Street View to estimate the demographic makeup of neighborhoods across the United States. In Proc. Natl. Acad. Sci., 2017. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5740675/pdf/pnas.201700035.pdf
[4] A. Haque, A. Milstein, FF. Li. Illuminating the dark spaces of healthcare with ambient intelligence. In Nature, vol. 585, pp. 193–202, 2020. https://www.nature.com/articles/s41586-020-2669-y.pdf
