Analysis

AMI Labs and the new AI challenge: teaching machines to understand the world

With over a billion dollars raised, AMI Labs challenges the limits of language models to create AI capable of understanding causality and real-world physics

by Francesco Branda*

10 March 2026

Yann LeCun, presidente esecutivo di AMI Labs, all'AI Impact Summit di Nuova Delhi, India, giovedì 19 febbraio 2026.

4' min read

Translated by AI

Versione italiana

4' min read

Translated by AI

Versione italiana

The recent $1.03 billion funding raised by AMI Labs, the start-up founded by Yann LeCun, is much more than market news. It is a statement of intent, a strong signal that artificial intelligence is entering a new and potentially revolutionary phase. It is no longer just about generating text, code or images: it is about trying to teach machines 'common sense', the ability to understand the world as humans and animals do.

To understand the scope of this bet, one must start with a radical critique of the dominant paradigm: Large Language Models (LLM). Models like GPT-4 are extraordinary tools, capable of manipulating natural language with impressive precision, but LeCun calls them "stochastic parrots", that is, they repeat what they have seen in the data, they predict sequences of symbols, but they do not understand reality. They have no notion of causality, physics or contest, they don't know what happens when a cup falls, when water gets wet or glass breaks. Their 'hallucinations' are not accidental errors, but the inevitable result of systems that have no model of the world.

This is where AMI Labs comes in. Its ambition is to build World Models, i.e. machines capable of having internal representations of reality. It is not about writing better or faster, but about simulating the world, anticipating consequences, making informed decisions in complex contexts. It is a situated intelligence, not purely linguistic. It is the step towards machines that can understand the physics, causality and logic of the real world.

Imagine the practical implications: a domestic robot that does not break glasses, understands the dynamics of a crowded room and knows how to manipulate fragile objects; an autonomous driving system that not only responds to road signs but also interprets complex and unpredictable scenarios; a medical assistant that can understand drug interactions and long-term consequences of clinical interventions. This is not science fiction: it is the direction in which AMI Labs wants to push AI.

The funding and prestige of investors, from Jeff Bezos to NVIDIA, reflect not only confidence in LeCun's leadership, but in a radical idea: to invest in basic research, without obsessing over the immediate product. It is a return to a classical scientific model, where failure and error are not only tolerated, but necessary to discover new laws and principles. At a time when the race to market dominates the technological narrative, this choice is almost revolutionary.

Leaving Meta to found AMI Labs is not only personal: it is symbolic. Within large companies, even those who lead the most advanced research encounter inevitable constraints: quarterlies, engagement metrics, pressure to quickly integrate new technologies into concrete products. AMI Labs represents the freedom to explore bold ideas, risky hypotheses and unconventional approaches. It is a bet on the long term, on the fact that true innovation comes from scientific curiosity rather than commercial efficiency.

Then there is a broader issue of an epistemic and cultural nature: artificial intelligence cannot become a monolith. If the whole world focused only on perfecting LLMs, we would risk building a technological and intellectual bubble. Parallel initiatives such as World Labs and AMI Labs are crucial because they explore alternative avenues, fertilise the field with new ideas and remind us that intelligence is a multifaceted phenomenon.

Here is the most fascinating point: if AMI Labs succeeds in equipping machines with common sense, causality and understanding of the world, the cultural leap will be enormous. We will no longer speak of AI as a mere tool, but as a cognitive partner, a collaborator capable of reasoning, anticipating scenarios and interacting with reality in ways similar to humans. We will be looking at an AI that not only knows, but understands.

This challenge also has profound economic and social implications. In robotics, healthcare, autonomous driving, safety, a system that thinks with common sense can reduce errors, increase efficiency and open up markets that do not exist today, but above all, redefine the relationship between man and machine: no longer passive tools, but intelligent interlocutors, capable of suggesting creative and adaptive solutions.

The path is long and full of unknowns. The transition from theoretical JEPA (Joint Embedding Predictive Architecture) models to functioning systems is complex, risky and expensive, but that is the nature of any scientific revolution: it requires patience, vision and courage. Readers cannot help but think that we are witnessing something epochal: an attempt to teach machines what until now seemed impossible, 'common sense'.

AMI Labs is not just a start-up. It is a manifesto of intellectual and scientific audacity. It is proof that the future of AI will not only be determined by the speed with which we produce products, but by the depth with which we understand the world and translate that understanding into artificial intelligence. We are talking about a change that can redefine industry, science and society. A bet on the common sense of machines that reminds us that sometimes the most extraordinary progress comes from careful observation of the real world and the willingness to teach it to those who do not know it: the machines.

And so, for those reading, the biggest question remains: if we can teach machines common sense, can we really teach them to think with responsibility, creativity and judgement? If the boundaries between language, action and understanding are becoming more permeable today, then the future of artificial intelligence is no longer just about technology. It is a journey towards a new form of reasoning, a new alliance between man and machine, where curiosity, vision and imagination become the real fuel of evolution. In other words, we are only just beginning to discover how far we can go.

* Unit of Medical Statistics and Molecular Epidemiology, University Bio-Medical Campus of Rome