The head of artificial intelligence, Meta, says that we are a long way from achieving a human-level AI model

Do today’s AI models really remember, think, plan and reason like the human brain? According to Ian Likan, Meta’s Chief AI Scientist, the answer is no for now. He thinks we can get there in a decade or so by following a new method called the “Global Model.”

According to TechCrunch, some AI analysts believe that we are very close to AGI (Artificial General Intelligence). However, during a recent talk at the Hudson Forum, Likan, chief AI expert at Meta, believes that this AI optimism is still far from reality. He says about this:

“We need machines that understand the world. (Machines) that remember things, have intuition, have common sense, and can reason and plan at the level of humans. “Despite what you may have heard, current AI systems are incapable of doing any of these things.”

Earlier this year, OpenAI released a new feature called “memory” that allows ChatGPT to “remember” your conversations. The startup’s latest AI model, o1, displays the word “thinking” as it produces output, and OpenAI says the models are capable of “sophisticated reasoning.”

Today’s big language models, such as those powering ChatGPT and Meta AI, are a long way from “human-level artificial intelligence,” LeCun says. Humanity may be “years to decades” away from achieving such a thing, he said.

We are far from human-level artificial intelligence models

The reason for this is simple: large language models work by predicting the next token (usually a few letters or a short word), and today’s image/video models predict the next pixel. In other words, language models are 1D predictors and AI image/video models are 2D predictors. These models have become very good at predicting the relevant dimensions, but they do not yet understand the 3D world.

For this reason, modern artificial intelligence systems cannot perform simple human tasks. LeCun points out how humans learn to clean the dinner table at 10 and drive a car at 17, both within hours. But even the most advanced AI systems in the world today, built on thousands or millions of hours of data, cannot yet work reliably in the physical world.