AI can develop a language that will be beyond the understanding of humans, scientist warns

A warning has been issued once again regarding Artificial Intelligence (AI), and this time it has been given by Geoffrey Hinton, often referred to as the "godfather of AI." During a recent conversation on a podcast, he stated that in the future, machines may develop their own language and thinking, which humans will neither be able to understand nor track.

Hinton said, "Right now AI systems think in English through a process called 'Chain of Thought', through which we can understand what they are doing. But the danger will increase when they create their own language to communicate with each other." He further said that AI systems can already process dangerous ideas.

Hinton, who won the Nobel Prize for Physics in 2024, has brought revolutionary changes in the field of AI. But now he himself is worried about this technology. He said, "I should have understood long ago what the dangers could be in it. I assumed that the future is still far away, but now I think I should have been alert then itself."

Hinton believes that the biggest difference between AI systems and human brains is that whatever AI learns, that knowledge can be instantly shared in thousands of systems. Giving an example, he said, "Imagine what will happen if 10,000 people learn something together? This is what happens with AI systems."

He says that models like GPT-4 have already surpassed humans in general knowledge, and are also progressing rapidly towards complex thinking.

Hinton believes that many people inside big tech companies understand these dangers, but avoid speaking publicly. He told that Google DeepMind CEO Demis Hassabis is working seriously in this direction.

On the reason for leaving Google, he said, "I did not leave Google in protest. I was 75 years old and was not able to code properly. But after coming out, I can now talk openly about these dangers."

While AI technology is progressing rapidly, governments around the world are trying to make policies and controls for it. America has recently introduced an "AI Action Plan", which aims to create a balance.

Hinton says that unless AI is made "guaranteed benevolent" i.e. completely human friendly, its development can be dangerous.