May 2024
AI Part 4: Unsung Mind Behind AI Imaging
Fei-Fei Li made computers ‘see’

While watching 3 Body Problem, a Netflix series adapted from a novel by a Chinese author and produced by a Chinese national, the portrayal of women scientists shows them working harder and longer than their male counterparts to solve problems, as women have always had to do to be taken seriously.

It immediately made me think of Fei-Fei Li, whose book The Worlds I See relates to her journey to becoming one of the most influential but least recognized AI researchers.

While investors like Elon Musk get most of the media attention for buying up technology and exploiting it, the current state of AI would not be nearly as advanced if it weren’t for the work of Li, a woman who refused to compromise when it came to her work.

Li created the foundation for AI to recognize images, which led to the incredible trajectory that vaulted AI forward when she developed ImageNet, the largest database of images when it was released in 2010. Even more critical, she established a contest to find algorithms that would better use the database for recognition tasks, the ImageNet Large-Scale Visual Recognition Challenge (ILSVRC). The future of computing was forever changed when a group of programmers used a neural net to compete — and got spectacular results.

That was the beginning of the realization that machine learning could be bested by the “old” way of thinking about AI, as a distributed “brain” of weights and biases that could compute answers. It was the connection that led to the Large Language Models that are in use today in ChatGPT, for instance.

Born in China, which actively discouraged women from pursuing math, science or other fields considered traditionally male-oriented, Li encountered discrimination from an early age. I can only imagine how devalued females were after the one-child policy led to the mass murder and abandonment of girl babies. One of Li’s teachers even called females “stupid,” though young girls bested boys in all her classes.

Li realized she would never reach her potential in her studies because of inherent bias against girls. Her mother, who’d shown great promise as a scientist, had to give up her dreams because of the harsh treatment of women. Li’s father was sympathetic. Her parents felt so strongly about encouraging her that they found a way to emigrate to the US, even though none of them spoke English. Li translated her high-school books with a dictionary, and learned through television and many painful experiences to adapt to American ways. A high-school mentor helped her adapt, and she ultimately was offered a full scholarship to Princeton University.

But Li's work didn’t get the respect it deserved till she pursued the creation of ImageNet, unshaken in her belief that it would be a breakthrough for digital visual recognition.

Li worked for Google for a short stint and was attacked for discouraging talk about the use of AI in the Maven Project, developed for the Defense Department. The weapon caused a walkout at the company before Google abandoned it. Mostly she’s worked in academia and research, most recently to improve medical care because of her own experiences in hospitals with her mother’s precarious health and heart issues.

Li also developed AI4All with a “mission to educate the next generation of AI technologists, thinkers and leaders by promoting diversity and inclusion through human-centered AI principles.” It offers free summer programs for high-school students at universities including Stanford, Princeton, Carnegie Mellon, Boston, the University of California Berkeley and Canada’s Simon Fraser University.

Currently Li continues her research as a professor at Stanford and works to develop AI tools to prevent hospital mistakes that harm or lead to the deaths of patients. She’s published more than 300 peer-reviewed research papers and has a raft of awards in her field and in science in general. She’s now the Director of the Human-Centered AI Institute at Stanford and a member of the National AI Resource Task Force, which advises the White House.

Apple Vision Pro and VR: A review

An alien-made headset in 3 Body Problem created a very real video-game world, engaging all the users’ senses during its use — touch, smell, hearing and vision. Its design evokes the Apple Vision Pro, though it looks more comfortable.

My husband obtained an Apple Vision Pro for development purposes and I’ve had the opportunity to use it a few times. While it can conjure up a virtual reality that’s better than anything else out there, it’s primarily for entertainment, in my opinion. The graphics can be very immersive, especially the 3D graphics created for it. You can play with dinosaurs that interact with you in the “Explore Dinosaurs” app —they move around when you move and sniff at you — and you can look around on the surface of the moon, for instance. Video games, which project life-sized, are far more interactive than on a TV or computer screen. I played Synth Riders, a musical game that involves catching Tron-like balls and rods. The practicality of using the Vision Pro for work seems dubious, although a cookbook writer found it useful for prepping ingredients without having to look up a recipe. You can see through the interior screen according to the opacity level you set. Still, till the headset is light as a feather or more like a partially open helmet, as in 3 Body Problem, most people are not going to want to wear it for more than an hour. When it gets uncomfortable, real reality sets in.

Toni Denis is a frequent contributor.

Thank you! Your submission has been received!
Oops! Something went wrong while submitting the form.