Current AI technology is limited, just good at language: AMI Labs executive chairman Yann LeCun

The next major phase of artificial intelligence (AI) will centre on systems that can truly understand and model the physical world, rather than just process language, said Yann LeCun, executive chairman of AMI Labs.

“The type of AI systems we have access to at the moment manipulate language very well, and this fools many people into thinking that we have already achieved systems that are as intelligent as humans, because we tend to view language ability as inherently human. It's just false. Our current technology is limited,” LeCun said at the AI Impact Summit in New Delhi.

According to Yann LeCun, understanding the real world is far more complex than processing language.


“Language is a sequence of discrete symbols, which makes it easier for computers to handle. However, the real world is messy, high-dimensional, continuous, noisy and far more complicated,” he said.

LeCun, formerly chief AI scientist at Meta, argued that human-level intelligence cannot be achieved by training AI systems solely on human-produced text. Instead, he said, future systems will need to learn from sensory inputs and develop a deeper understanding of how the real world works.

“The industry, particularly Silicon Valley, which comprises the dominant players, is entirely focused on LLMs. Everyone is working on the same thing, stealing each other’s engineers and pursuing similar approaches, because no one can afford to try something different and risk falling behind. This creates a kind of monoculture that leaves the industry somewhat short-sighted,” he added.

LeCun noted that heavy investments need to go into academic research, as this line of work needs stronger support, and that the dominance of LLM research should not crowd out alternative approaches. He added that governments and industry also have a role to play in backing and scaling these models once they prove effective.