The Worlds I See — Curiosity, Exploration, and Discovery at the Dawn of AI

The Teenager Who Changed AI

Fei-Fei Li arrived in the United States from China as a teenager, speaking limited English, with parents who worked in a laundry and a restaurant to keep the family afloat in a country that had not opened easily to them. She was also exceptionally gifted — mathematically, scientifically, intellectually — and would eventually end up at Princeton, then graduate school in neuroscience and computer science, then a professorship at Stanford and a stint leading AI at Google Cloud.

The Worlds I See is the account of this trajectory, and it is a memoir that has earned the right to its ambitions. Fei-Fei Li is not simply narrating her career; she is telling the story of how the AI revolution happened — specifically, how the particular technical breakthrough that enabled modern computer vision occurred, who made it happen, and what it cost.

What ImageNet Was

In 2007, Fei-Fei Li conceived an idea that was regarded in the AI research community as a distraction from real science: a massive database of images, carefully labelled by humans, that could be used to train machine learning systems to see. The project — which would take years to assemble, require the labour of thousands of Amazon Mechanical Turk workers, and consume resources most AI labs considered wasteful — was called ImageNet.

The results, when the ImageNet Large Scale Visual Recognition Challenge was launched in 2010, were transformative. The competition became the crucible in which deep neural networks demonstrated their capability, most dramatically in 2012 when AlexNet achieved error rates that were orders of magnitude better than anything previously achieved. That result — enabled by ImageNet’s data, by GPU computing, and by the mathematical insights of Geoffrey Hinton’s group — is now identified as the moment the deep learning revolution began.

The account of ImageNet’s creation is the book’s most remarkable section. Fei-Fei Li narrates it with the excitement of someone describing an adventure, but the adventure has a specific texture: the difficulty of assembling something as unglamorous as a labelled image database, the scepticism of colleagues who thought the project a waste of time, the logistical challenges of running a competition that attracted researchers from around the world. This is science as it actually happens — not the eureka moments of popular mythology but the long, laborious, frequently discouraging work of building the conditions in which breakthroughs become possible.

The Human Labour Question

The Worlds I See is honest about a dimension of AI development that receives insufficient acknowledgement: the human labour on which AI systems depend. ImageNet was assembled with the labour of thousands of Mechanical Turk workers — people who clicked through images for small payments, labelling objects and scenes with a conscientiousness that most had no particular reason to provide but generally did.

This labour created something of enormous value, and the people who provided it received almost nothing from that value. Fei-Fei Li grapples with this directly and with more honesty than the AI industry as a whole tends to display. She does not reach a clean resolution — there isn’t one to reach — but the honesty of the engagement is itself significant.

The same question applies to the broader data ecosystem on which contemporary AI depends: the text, images, and other content generated by millions of people who did not consent to its use as training data and will not benefit from the systems built on it. This is not a problem that any single researcher or book can solve, but naming it clearly and refusing to wave it away is a genuine contribution.

Stanford, Google, and AI’s Power Politics

The middle sections of the memoir — Fei-Fei Li’s Stanford years, her time as Chief AI Scientist at Google Cloud, and her founding of the Stanford Human-Centered AI Institute — are rich with the institutional politics of AI development in an era when the field transformed from academic backwater to the most contested territory in global technology competition.

Her time at Google is particularly interesting. She describes the specific experience of a senior researcher-executive in a commercial AI operation: the tension between scientific values and commercial imperatives, the management of media attention and policy engagement, the specific discomforts of representing an organisation that is simultaneously making genuinely exciting scientific progress and accumulating unprecedented power over human information systems.

Fei-Fei Li’s Ethical Position

The ethical dimension of the memoir is its most demanding element. Fei-Fei Li co-created tools that are now deployed in surveillance systems, weapons targeting systems, and commercial monitoring operations that raise serious civil liberties questions. She is clear that she did not intend these applications and has actively opposed some of them. But the question of what responsibility attaches to building powerful tools that can be misused is not answered by good intentions.

Her engagement with this question is genuinely thoughtful. She does not claim that the technology is neutral; she does not argue that concerns about misuse are exaggerated; she does not retreat to the position that scientists are not responsible for applications. Instead, she maintains that the right response is continued engagement — arguing for particular uses and against others, using whatever platform the work has provided to advocate for responsible development.

Whether this is adequate is a question the book leaves appropriately open.

Our rating: 4.6/5 — A rare memoir that earns its ambitions. The origin story of modern AI, told from the inside.