Meta chief AI scientist and NYU professor Yann LeCun has been setting the direction of the artificial intelligence field for decades. His latest initiative, however, has raised some eyebrows. LeCun co-founded a startup called Advanced Machine Intelligence (AMI) which will focus on developing AI that can comprehend the real world. The initiative has already attracted $1.03 billion in funding.
For those curious, there is a list of investors who participated in the Series A round online.
The interesting part is that AMI will not be working on chatbots. In fact, the company is not working on conversational AI or competing with the next generative AI model. This sounds a bit odd in a time when the AI community is infatuated with chatbots. But LeCun is not a fan of large language models (LLMs), saying that they are not intelligent. At best, LLMs can predict the next word in a sentence, LeCun says. But that is not intelligence.
And in a way, he is right.
LLMs can generate entire articles, summarize documents, and tell jokes. They can converse with humans in a way that many times humans forget that they are speaking with a machine. But LLMs do not understand the world around them. They have no concept of what the world is or how it works. They can describe the world because they have seen examples of this world in their training data, but they do not understand this world.
The concept of understanding is pretty basic. If a human child pushes a glass of the edge of a table, the glass will fall to the floor due to gravity and will probably break. Humans understand that. But AI models do not understand gravity. They do not understand physics. They do not understand cause-and-effect.
But LeCun wants to change that.
The Meta scientist and his colleagues have been advocating for “world models,” AI models that have a concept of how the world operates. World models are not text prediction models. Instead, they learn the relationship between action and outcome. They learn how the world works. They learn cause-and-effect.
This is not what the rest of the AI community is working on. Companies like OpenAI, Google, and Anthropic are focused on making larger language models, training them on more data and bigger hardware, and scaling LLMs to new heights. The strategy has delivered impressive results in recent years.
LeCun’s idea is a bit different. Maybe more theoretical. Maybe more ambitious. Depending on how you see it.
Rather than scaling language models, LeCun is asking how to make AI models that understand the world around them. That is a much more challenging question.
To achieve this, AMI will have to develop models that reason about the physical world. These models will have to be able to predict what will happen when something is done. These models will have to learn by observing the world around them. These models will have to learn cause-and-effect. Researchers have already begun exploring these types of models, describing AI that can make hypotheses, check predictions, and learn to comprehend the world around them.
Yet, it is hard not to be a little skeptical.
The AI community has a tendency to reward anything that appears shiny and bright. The biggest models. The flashiest demos. The fastest benchmarks. The AI market sometimes feels like a circus in which companies are trying to outdo each other for attention as much as for innovation.
That is part of why LeCun’s effort is so interesting. It is not about building the biggest and baddest AI model. It is about taking a step back and trying to answer a more profound question: What is intelligence, really?
If AMI is successful, the world could see AI models that do not just describe the world. AI models will be able to reason about the world. Interact with the world. Learn about the world like humans do. That would unlock an entirely new generation of AI technologies, especially in areas like robotics and scientific research.
And if it fails? So be it. That is what happens when you dream big.
But it is pretty clear that $1.03 billion in funding for a project with an unorthodox mission says something. In the midst of the generative AI frenzy, some of the most important people in the AI community think that the next revolution could come from an entirely different direction.