Series: Laboratorio del Futuro – Voices Building Tomorrow

Five voices. Five views on the future of work. From Musk's utopia to Hinton's warning, by way of Amodei's alert, Altman's reorganization and Li's humanization.

The laboratory of the future is not a place with robots assembling boxes. It is a space for debate, for collective decision-making, for building institutions. Because in the end, the question about work is not a technical question. It is a question about what kind of society we want to be.

And as Hinton said, the answer is not in the laboratories. It is in all of us.

The Museum Scene, February 2026

The Museum of Modern Art in New York is not, at first glance, the place where you would expect to find one of the most influential minds in artificial intelligence. Its galleries are made for canvas, sculpture, installation. For the tactile, the visual, whatever can hang on a wall or occupy a volume in space. And yet, that February afternoon, Fei-Fei Li was there, at MoMA, surrounded by artists and curators, talking about machines.

On her left, Holly Herndon and Mat Dryhurst, Berlin-based artists who have spent years exploring the boundaries between the human voice and the generative one. On her right, Michelle Kuo and Paola Antonelli, the museum's curators. The topic was as unusual as the setting: "How to See the Human in the Machine."

Fei-Fei Li, with that mix of academic precision and warmth that characterizes her, let drop a sentence that floated over the room like a question with no easy answer:

"I think human intelligence and machine intelligence manifest in very different ways. We don't just survive; we live, we work; we also express emotions, we feel, we have compassion, we have creativity that we can't explain. Machines don't have independent values. Machines' values are human values."

Her entire philosophy was condensed in that sentence. While other tech leaders talk about what machines can do, Fei-Fei Li talks about what machines should be. And that difference, subtle but abysmal, is what makes her an indispensable voice in the chorus of the future.

Fei-Fei Li’s thesis: spatial intelligence

An Intelligence That Is Not Just Language

To understand Fei-Fei Li's vision, we have to go back a few weeks, to January 6, 2026. On the stage at CES in Las Vegas, the world's most important technology show, she shared a conversation with Lisa Su, CEO of AMD. The contrast was interesting: Su, a hardware engineer, talked about chips and computing power; Li, a Stanford researcher, talked about something more elusive.

"The world has a potential for AI that is still underestimated," Li said. "Language models have made enormous progress in recent years, but intelligence is not limited to language. For humans, spatial intelligence is what allows us to act, judge and create in the real world."

And then she delivered the sentence that sums up her thesis: "Language is a tool for describing the world, but it is not the world itself."

This distinction is crucial. Over the past decade, progress in artificial intelligence has been measured almost exclusively by how fluently machines handle language. Models capable of reading, writing, summarizing, translating and generating text have redefined entire industries. But for Fei-Fei Li, that approach has a structural limit: the world in which people and machines operate is not textual or flat, but three-dimensional, dynamic and full of ambiguity.

A fly, Li likes to remind us, has no trillions of parameters and has not been trained on the internet. And yet it can enter a messy room, dodge obstacles at high speed and land with millimeter precision. That is spatial intelligence in its purest form: the ability to understand depth, distance, occlusion, gravity and causal relationships in real time.

World Labs and the New Frontier

That understanding of space is not just an academic curiosity. It is the foundation of World Labs, the startup Fei-Fei Li cofounded in 2024 and which in barely two years has become one of the most closely watched projects in Silicon Valley. With a valuation of more than a billion dollars and funding of more than $230 million, World Labs has set out to do something that sounds like science fiction: teach machines to generate entire three-dimensional worlds that are consistent, navigable and governed by the laws of physics.

In January 2026, the company launched its first commercial product: a model they called Marble.

Marble is not a language model. It does not compete with GPT-5 or Claude 4. It takes multimodal inputs (text, images, video or simple 3D prompts) and generates environments that can be navigated and interacted with. What is revolutionary is not just what it does but how it does it: unlike video models, which generate sequences of pixels with no geometric coherence, Marble maintains spatial consistency. What you see from one angle matches what you see from another. Distances are real. Relationships between objects are stable.

Fei-Fei Li describes Marble as early and limited, but functionally different from anything that has existed before. And in that difference lies the key to everything.

Marble, trust, and the data challenge

The Applications That Are Already Here

When talking about spatial intelligence, it is easy to get lost in abstractions. That is why Li makes an effort to give concrete examples, applications that are already happening.

Game developers use Marble to generate entire environments without having to model every tree, every building, every texture by hand. What used to take teams of dozens of artists months can now be generated in hours.

Virtual production teams (the same ones who create digital sets for film and television) use these models to build scenes that actors can inhabit and cameras can move through, without the astronomical costs of location shoots.

Architects are beginning to use spatial intelligence tools to "walk" virtually through their designs before the first brick is laid. They can feel the proportions, the light, the circulation, and adjust on the fly.

Roboticists, perhaps the most interested, use these generated worlds as training environments. A robot can practice millions of times in simulation how to grasp a fragile object or navigate a messy kitchen, without breaking anything or hurting anyone. And when it goes out into the real world, it has already seen almost every possible variation.

But there is one application that surprised even Li herself. Clinical researchers have approached World Labs with an unexpected proposal: to use generated environments for mental health research. In these controlled but realistic worlds, they can study how people react to spatial stimuli, how they behave in contexts that would be difficult or impossible to reproduce in the real world.

Trust as a Technical Problem

At the Cisco AI Summit held in San Francisco in February 2026, Fei-Fei Li took the stage to talk with Jeetu Patel, the company's chief product officer. The room was full of technology executives, infrastructure leaders and analysts. And Li, instead of talking about algorithms or funding, talked about trust.

"The ability to understand, reason about, interact with and navigate the real physical world in 3D and 4D is the foundation," she said. "And that foundation is necessary for systems to operate beyond screens."

But what was interesting about her remarks was how she defined trust. Not as an abstract ethical principle, nor as a layer of control added afterward, but as an operational condition. A system that acts in the physical world must be predictable, interpretable and consistent. It is not enough for it to work "most of the time." In fields such as robotics, advanced logistics, manufacturing or assisted surgery, margins for error are narrow. A failure does not mean a wrong answer; it means physical harm, a supply chain disruption or a risk to people.

Li was explicit: spatial intelligence will not, in the short term, solve the productivity challenges that preoccupy most organizations today. Nor will it replace language models as the main interface. Its relevance lies elsewhere: in the medium and long term, it will be decisive for moving artificial intelligence off the screen and turning it into a technology capable of operating autonomously in the physical world.

The Data Challenge

One of the questions Fei-Fei Li faces most often is why spatial intelligence is not advancing at the same pace as language models. Her answer is always the same: data.

Text is abundant, cheap and easy to process. The internet is full of books, articles, conversations and documents. Physical data, by contrast, is scarce, expensive and hard to capture. For a model to understand three-dimensional space, it needs millions of examples of real scenes, with depth, texture, lighting and movement. It needs sensors, simulations and representations that are still under development.

Li compares the process to the development of autonomous vehicles. Companies working on self-driving spent years collecting real and simulated data before achieving limited commercial deployments. Spatial intelligence is at a similar stage. Today's models are smaller and are trained with fewer computational resources than the large language models. That gap reflects both the scarcity of data and the youth of the field.

"Just because the North Star is clear doesn't mean the journey is short," she likes to say.

Humanism, work, and the “North Star”

Humanism as Method

If there is one thread running through Fei-Fei Li's entire trajectory, from her days as an immigrant student working in a laundromat to her current position as one of the most respected voices in artificial intelligence, it is the conviction that technology and humanity cannot be separated.

Her own story is already legendary in Silicon Valley. She arrived in the United States at 16, with her parents and barely $20 in her pocket. The family lived in a small one-bedroom apartment; Fei-Fei's bedroom was a hallway in the kitchen. Her father worked repairing cameras; her mother was a supermarket cashier. After high school, Fei-Fei worked at a Chinese restaurant scrubbing dishes, scrubbing floors, cleaning windows, taking orders and waiting tables. Two dollars an hour.

When it came time to go to college, many advised her to choose a practical career that would give her financial security: medicine, finance, engineering. But she chose physics. And later, the study of computer vision, when almost no one believed machines could ever "see."

That life experience shaped her way of understanding artificial intelligence. In her memoir, The Worlds I See, published in 2023, she writes about the "paradox of precariousness and wonder": how the most basic need can coexist with the loftiest curiosity, how science is not a luxury for the privileged but a necessity of the human spirit.

The Intelligence That Gives Us Back Our Humanity

In the MoMA conversation, Fei-Fei Li was asked directly about the future of work. Her answer was, like almost everything she says, nuanced and deep.

"The question is not which jobs artificial intelligence is going to take away," she said. "The question is which jobs we are going to want machines to do and which ones we are going to want to keep doing ourselves. And that is not a technical question. It is a human question."

Li believes there are dimensions of work that machines cannot replicate, not because they are too complex, but because their value lies precisely in the fact that humans do them. Caring for an elderly person, teaching a child, the conversation between colleagues who share a difficulty, the creativity that arises from lived experience. None of that is automatable, not because it is technically difficult, but because it would lose its meaning if it were.

At the Cisco AI Summit, someone asked her what advice she would give young people starting their careers in a world dominated by artificial intelligence. Her answer was surprisingly simple:

"Learn to use these tools, of course. But don't forget what makes them valuable. Don't forget that technology exists to serve humans, not the other way around. And above all, don't be afraid to ask hard questions about what kind of world we want to build."

The North Star

Fei-Fei Li has an expression she uses often: "the North Star." For her, research is not a blind accumulation of knowledge but a search guided by principles, by fundamental questions, by a vision of what should be.

Her North Star, in this new stage of her career, is spatial intelligence. But not as an end in itself, rather as a means to something bigger: machines that do not just process information but understand the world we inhabit; systems that do not just answer questions but can act safely and reliably in real environments; technology that does not replace the human but amplifies and cares for it.

"Without spatial intelligence," she wrote at the end of a long essay published in late 2025, "our dream of a truly intelligent machine will never be complete. This exploration is my North Star."

The Question That Remains

When Fei-Fei Li stepped down from the stage at MoMA, after that conversation about the human and the mechanical, a young woman from the audience came up to ask her a question. She had no microphone, so her words were not captured in the institutional video. But those standing nearby heard her.

"Do you think machines will ever come to love us?"

Li smiled, like someone receiving a question she has already asked herself many times.

"Machines don't need to love us," she answered. "They need to help us love each other better. That is already quite a lot."

And she disappeared among the museum's galleries, among paintings and sculptures, among works made by human hands to be seen by human eyes. Outside, on the street, the city went on. People were heading to their jobs, to their homes, to their lives. They carried machines in their pockets, machines on their desks, machines on their walls. But they were still human. And that, perhaps, was the only intelligence that really mattered.

In the next article in the Laboratorio del Futuro series: we close the cycle with the most critical and profound voice of all. Geoffrey Hinton, the "godfather of AI," offers us his warning about the price of abundance and the risks of a world where intelligence is no longer exclusively human. We call it "Geoffrey Hinton and the Price of Abundance: The Godfather's Warning".