AMI Labs CEO Alexandre LeBrun: Why we avoid the terms AGI and superintelligence

While most representatives of the AI industry are rushing to label their developments as "AGI" (Artificial General Intelligence) or "superintelligence," Alexandre LeBrun, CEO of the startup AMI Labs, prefers to abandon such lofty terms altogether. In an interview with TechCrunch, the head of this company, founded by Yann LeCun, emphasized that these concepts lack a precise scientific definition and have no practical value. This is reported by Techcrunch.com .

According to LeBrun, industry players are constantly busy inventing new marketing terms. "We have never used the word AGI. Now everyone has started talking about superintelligence, and next time they will come up with another name. What is superintelligence anyway? I don't know, it's not a very useful word," he says. This approach differs sharply from the strategy of other giants at the center of the AI race.

World models and the robotics revolution

AMI Labs is currently working on "world models." Unlike text-based models (LLM) like ChatGPT, this technology focuses on understanding the laws of the physical world and predicting the sequence of events. For example, predicting that a glass falling off a table will break and the liquid inside will spill is the primary task of a world model. According to LeBrun, this is exactly the "brain" that current robots are missing.

Although the hardware in the field of robotics has developed surprisingly well today, their intelligence remains at a low level. Robots mainly operate based on strictly defined algorithms and cannot adapt to a changing environment. World models allow robots to understand context, which is crucial for ensuring their safety.

Safety and real-world problems

LeBrun noted that a solution has not yet been found for robots to move safely in open environments, such as on the street or inside a home. He cited an incident where a robot dancing at a public event kicked a young child. If the robot had a world model, it would have understood the surrounding context and would not have made such a mistake. Today, robots can only work effectively in closed and controlled areas in factories.

At the same time, the AMI Labs head noted that world models will not replace LLM systems, but rather complement them. Just as the language center and logical reasoning center are separate in the human brain, future AI systems will develop text processing and physical world understanding functions in parallel. Currently, the company is working on establishing connections with major partners in the manufacturing and electronics sectors in countries like South Korea.