Why experts are split on how close artificial general intelligence really is?

Top tech leaders say AGI is almost here

Silicon Valley’s most prominent voices — OpenAI’s Sam Altman, Anthropic’s Dario Amodei, and Elon Musk — are increasingly confident that artificial general intelligence, or AGI, is on the brink of arrival. Altman reportedly told US President Donald Trump it would appear before the end of his administration. Musk suggests it could come by year’s end.

AGI refers to a type of machine intelligence that matches or surpasses human cognitive abilities — an idea that has fascinated and haunted scientists and storytellers for decades. While there is no agreed-upon definition, the concept has become shorthand for AI systems capable of handling virtually any intellectual task a human can.

The recent acceleration in AI performance, especially with large language models like OpenAI’s ChatGPT and Anthropic’s Claude, has intensified these claims. These systems are already transforming how people work, create, and interact with machines. But for every confident prediction, there are equally compelling voices urging caution, the New York Times reported.

Technologists warn: the science isn’t there yet
Despite the bold forecasts, many experts remain sceptical. Nick Frosst, a former Google researcher and co-founder of AI startup Cohere, argues that today’s models are still primitive compared to human intelligence. “They predict the next word or pixel — not understand meaning,” he said.

A recent survey by the Association for the Advancement of Artificial Intelligence showed that more than 75% of researchers believe current approaches are unlikely to result in true AGI The gap lies in how AI models function versus how the human brain works: machines rely on patterns and probabilities, whereas human cognition is rooted in flexibility, creativity, and context.

Steven Pinker, a Harvard cognitive scientist, notes that machines may excel at narrow tasks, but this doesn’t generalise to broader intelligence. “There’s a temptation to engage in magical thinking,” he said. “But these systems are very impressive gadgets, not omniscient problem-solvers.”

Scaling up — or hitting limits?
Supporters of imminent AGI often cite scaling laws, which suggest that the more data and computational power AI systems are given, the better they become. Jared Kaplan, chief science officer at Anthropic, helped formalise this concept, arguing that reinforcement learning and greater practice could allow AI to match human learning eventually.

But others point out that the internet’s English-language data is almost exhausted, pushing companies to train systems with trial-and-error methods like reinforcement learning. While this has led to major wins in fields like math and programming, it has yet to yield comparable results in subjective fields like creative writing or ethics — where humans still outperform.

From chatbots to the physical world
Even as chatbots generate essays and solve equations, human intelligence is not limited to language or logic. “Part of intelligence is knowing when to flip a pancake,” said MIT’s Josh Tenenbaum, pointing to our physical interaction with the world.

Training AI in robotics — which would allow machines to learn from real-world experiences — is far more difficult than training them on internet data. It requires time, expensive hardware, and environments like homes or warehouses. As a result, robotic development is far behind chatbot innovation.

And even in digital realms, AI struggles with ambiguity and novelty — the very things that define human problem-solving.

The dream persists — but timelines are uncertain
For centuries, humanity has dreamed of building intelligent machines, from the Golem of Jewish folklore to HAL in 2001: A Space Odyssey. That fantasy now feels closer than ever — but the technological reality may still be far away.

Pioneers like Yann LeCun, chief AI scientist at Meta and a Turing Award winner, remain cautious. Despite helping build the foundations of today’s AI systems, LeCun believes a new paradigm is needed to achieve AGI. “A lot is riding on figuring out whether the next generation architecture will deliver human-level AI within the next 10 years,” he said. “It may not. At this point, we can’t tell.”