As a daughter, a mother, a Chinese immigrant in the US and a leading artificial intelligence researcher, Fei-Fei Li has seen several different worlds. Her life experience has also inspired her academic expertise: training computers to recognise and interpret images.
As Li writes, it is astonishing how the human brain — “a wet, organic lump about five inches in diameter” — can outperform warehouse-sized supercomputers when it comes to understanding the “gist” of everything we see. Flash a photograph of a Thanksgiving dinner in front of any American and most could immediately guess the relationships between the family members and spot they were eating turkey. How does this conceptual magic work? In some respects, this instinctive ability to understand context defines what it is to be human. “Understanding how we see, therefore, is to understand ourselves,” she writes.
In The Worlds I See, Li explores three themes with a deft touch and moving insight. She provides a clear and accessible history of machine learning, describing how this conceptual transformation came about. Li also makes a heartfelt plea to keep the human at the heart of AI, to ensure we maximise the benefits of our new technological powers.
But above all she tells a captivating tale about how a poor, immigrant Chinese girl could land in the US and end up at the forefront of a scientific revolution. The book is a loving portrait of Li’s familial past as well as an exploration of our technological future in an expert interweaving of the personal and professional.
“I was born the only child of a family in a state of quiet upheaval,” Li writes about growing up in the 1980s in Chengdu, capital of Sichuan province. Like many of China’s ancient cities, Chengdu’s narrow streets, open courtyards and green landscapes were being obliterated by soulless Soviet-style architecture. But that bleak urban backdrop contrasted with the warmth and imagination shown by her parents, bound together in ways that no one else could understand. “They’d both lost faith in the institutions of the world, and that made them partners, even accomplices, in a daily practice of defiance,” Li writes.
Her father, whose job in the computing department of a chemical company was more a facade than a career, had a pathological allergy to seriousness, a boundless curiosity in nature and a goofy ingenuity. He came alive when searching for butterflies, watching water buffalo splashing around in rice fields and trapping rodents and stick insects to keep as pets. And although he missed his daughter’s birth because he had been distracted on a birdwatching trip, he whimsically named her Fei-Fei. In Mandarin, fei means to fly. When rendered into a Chinese character, her name resembles a bird. “The highest compliment I can pay my father doubles as the most damning critique: that he’s exactly what would result if a child could design their ideal parent in the total absence of adult supervision,” Li writes.
Li’s mother, a high-school teacher turned office worker, was an equally singular individual, “an intellectual trapped in a life of imposed mediocrity”. Her own grandmother had been among the first generation of women to attend university. But her mother’s own intellectual career was blighted by the family’s associations with the defeated opposition Kuomintang party. Her life was also plagued by ill health, which sapped even her ferocious energy.
Her coping mechanism was to escape into literature, a passion she passed on to her daughter, who devoured the Chinese translations of Simone de Beauvoir, Charles Dickens and Ernest Hemingway. At school, Li was angered to overhear a teacher telling the boys that they were “biologically smarter than girls”. That quiet heat of indignation fuelled her studies as she fell in love with the beauty and grandeur of physics.
Li’s father moved to the US to start a new life for his family. After three years apart, the then 15-year-old Li and her mother joined him in New Jersey. Li spoke barely a word of English when she enrolled at the Parsippany High School, where everything was “brighter, faster, heavier and noisier” than the world she had left behind. But she was lucky to be taught by Bob Sabella, or the “big bearded math teacher”, as her parents called him. Thanks to his inspirational guidance — and financial support — Li won a place at Princeton University, a world apart from the laundry her parents were running.
Li gravitated towards the field of computer vision, a relatively neglected area that was being transformed by the deep learning revolution. Her main claim to academic fame has been her role in creating ImageNet, which at the time of its launch in 2009 was the largest hand-curated data set in AI history, containing 15mn images across 22,000 categories, annotated by more than 48,000 contributors from 167 countries. Combined with the application of smarter algorithms and massive computing power, ImageNet has triggered an explosive growth in computer vision capability. These advances have helped researchers to develop self-driving cars and the text-to-image generation models, such as Stable Diffusion, DALL-E and Midjourney, that so wow users today.
The third theme of the book is a discussion about how we might maximise the benefits of AI while minimising its potential harms. As Li explains, a subject that was once the insular domain of expert scientists should now concern everyone. After decades spent in vitro, AI is today very much in vivo, as she puts it. The tech researchers and world leaders mingling at the AI safety conference in Bletchley Park this week highlight how the topic has shot up the global policy agenda.
As one of the few women leading the field, Li is particularly concerned about the industry’s startling lack of diversity. And having worked as chief AI scientist at Google Cloud, she says she is unnerved how the vast wealth, power and ambition of the big technology companies outmatches anything in the university sector. In 2015, for example, Uber poached almost the entire robotics team from Carnegie Mellon University to develop its self-driving cars.
To redress the balance, Li has helped launch AI4All to encourage more diverse entrants into the industry. She also co-founded the Stanford Institute for Human-Centred Artificial Intelligence, which aims to strengthen AI research, education and policy for the public good.
The Worlds I See ends with a sense of awe at our ability to create a technology akin to a force of nature, “something so big, so powerful, and so capricious that it could destroy as easily as it could inspire”. But it will take a lot more than corporate bromides, she concludes, to make AI worthy of our trust.
The Worlds I See: Curiosity, Exploration, and Discovery at the Dawn of AI by Fei-Fei Li Flatiron Books $29.99 336 pages (publishing in the UK in December £25.99)
John Thornhill is the FT’s innovation editor
Read the full article here