Current AI isn't truly intelligent, say experts, as focus shifts to next-generation systems

As artificial intelligence continues to transform industries, some of the field's leading researchers argue that today's AI systems are far from achieving true intelligence, with the next wave of innovation expected to focus on machines that better understand the physical world.

Yann LeCun, one of the world's leading AI researchers and former chief AI scientist at Meta, believes current large language models (LLMs) such as ChatGPT, Claude and Gemini have significant limitations despite their impressive capabilities.

LeCun, who left Meta in 2025 to establish Advanced Machine Intelligence Labs (AMI Labs), said existing AI systems are not designed to deal with the complexity of the real world.

"We don't have robots that are nearly as good at understanding the physical world as a rat," LeCun said during France's VivaTech conference.

According to him, current AI models can generate text, solve coding problems and handle mathematical tasks, but they lack genuine understanding.

"They basically just accumulate knowledge. They can regurgitate something; you train them to regurgitate, but they're not particularly smart. They don't have an underlying understanding," he said.

LeCun argues that scaling up existing language models alone will not lead to human-level or even animal-like intelligence.

"They're not a path towards human-level or human-like intelligence, or even animal-like intelligence, because they cannot deal with real-world data; they just are not built for that," he said.

AMI Labs is developing a new type of AI called Joint Embedding Predictive Architecture (JEPA), which aims to understand how the physical world works instead of simply predicting the next word or sentence.

Unlike traditional language models that rely on statistical patterns, JEPA is designed to create abstract representations of the real world, helping AI evaluate possible outcomes without attempting impossible predictions.

LeCun illustrated the concept with the example of balancing a pen on its tip. While people instinctively know the pen will fall, no one can accurately predict the exact direction. He said current language models may attempt such predictions, whereas JEPA would recognise that the direction cannot be determined in advance.

The approach has attracted strong investor interest. Earlier this year, AMI Labs announced it had secured more than $1 billion in seed funding from investors including Nvidia and the investment fund managing Amazon founder Jeff Bezos' private wealth, making it one of Europe's largest early-stage funding rounds.

Developing AI that can safely perform everyday physical tasks remains one of the biggest hurdles for the robotics industry.

Although billions of dollars have been invested in humanoid robots, teaching them routine household tasks such as ironing clothes or loading a dishwasher remains difficult and expensive.

LeCun believes current language models are not suitable for such work.

"LLMs are largely hopeless for robotics," he said.

"The claims that somehow by just scaling up LLMs, we're going to reach superhuman intelligence, that is simply not going to happen."

Many AI researchers share LeCun's view that future progress will depend on systems capable of understanding cause and effect rather than simply recognising patterns.

Ingmar Posner, Professor of Applied Artificial Intelligence at Oxford University and director of its Applied AI Lab, said the next decade is likely to focus on AI systems that can explain their reasoning.

"My view is that the next decade will really be about systems that can explain... You need models that can answer questions like. What matters? What causes what? What would happen if I did something else - like if I took a different action?" Posner said.

He and his team have spent the past four years developing an alternative approach known as mechanistic world models, which organise knowledge so AI can recall, combine and adapt information more effectively.

The broader concept of World Models has gained momentum since a landmark 2018 research paper by David Ha and Jürgen Schmidhuber suggested AI could learn by building internal simulations of the world.

Since then, companies including Google's DeepMind, autonomous driving firm Wayve and World Labs, founded by AI pioneer Fei-Fei Li, have been pursuing similar research.

LeCun said AMI Labs plans to continue refining its AI model throughout this year before introducing it in industrial applications next year.

If successful, he believes the technology could eventually support more general-purpose AI systems capable of performing a wide range of tasks with minimal additional training.

"Eventually down the line we'll have sort of general generic intelligence systems that can be applied to just about anything in the world with minimal training or fine-tuning," he said.

Despite advances in AI, LeCun believes humans will continue to play the central role in creativity and decision-making.

"We're still going to need humans to figure out what questions to ask, what to build, what to create, which is really the properly human aspect," he said.

He added that future AI systems, even if they surpass human intelligence in some areas, would function more like highly capable assistants.

"Our interaction with future AI systems - even if they are smarter than us - is going to be like the interaction between a captain of industry or a political leader with their staff of assistants - many of whom are smarter than they are."