Yann LeCun, Meta's Director of Artificial Intelligence, won the Turing Prize, believes that general intelligence does not actually exist. According to him, the concept of AGI is often used to refer to intelligence at the human level, but human intelligence is more super-specialized than general.
LeCun argues that although humans are very good at adapting to the real world, communicating sociably and handling vague situations, they are not superior in tightly organized tasks such as playing chess or optimal problems.
We often think we have an overview, but it is an illusion. We only solve problems that the brain allows us to imagine well, Mr. Yann LeCun shared in a recent podcast.
In response, Demis Hassabis, CEO of Google DeepMind, said that LeCun had confused between general smart space and general smart space.
In a post on social network X, he affirmed that the human brain is the most complex and sophisticated phenomenon ever known, and possesses an incredibly superior level of popularity.
Hassabis also cited the lyse of no free lunch in machine learning, which suggests that there is no single algorithm that can optimize for all problems.
That means, in finite systems, there always needs to be a certain level of specialization. However, he emphasized that overall thinking does not lie in being the best at every task, but in the ability to learn anything that can be calculated if there is enough data, time and resources.
According to Hassabis, both the human brain and modern platform AI models can be considered Approxative Turing machines, meaning that in theory they can learn any calculable problems.
He also rejected LeCun's argument when comparing humans to game machines, saying that it is humans who can invent chess, science or Jets, which is a strong demonstration of the superiority of intelligence.
This debate reflects two different approaches in the AGI race. Hassabis believes that expanding the scale of major language models (LLM) is not enough, and the AI field still needs more fundamental breakthroughs.
Meanwhile, LeCun believes that LLM is a "dingle" because of the lack of continuous learning and understanding the real world.
Instead, LeCun believes that it is necessary to focus on developing world models, which are AI systems capable of understanding and simulating physical laws, causation and time movement. According to him, the concept of advanced machine intelligence reflects more accurately the research goals than the controversial term AGI.
