Meta CEO Mark Zuckerberg has revealed that the company’s artificial intelligence systems have started showing the ability to improve their own capabilities — without any human intervention. In a policy paper published, Zuckerberg described this development as slow but undeniable, suggesting it may be the first milestone toward achieving artificial superintelligence (ASI).

From AI to Superintelligence

AI experts typically classify artificial intelligence into three levels. Today’s models often outperform humans in specific, narrow fields — such as predicting protein structures — but they lack the broader, adaptable reasoning of human intelligence. The next stage, known as artificial general intelligence (AGI), would enable AI to learn, understand, and apply knowledge across various domains, much like the human brain.

ASI represents the ultimate stage, where AI surpasses human capabilities entirely and could improve itself at an exponential pace — a phenomenon known as the “intelligence explosion”. This moment, often called the technological singularity, remains hypothetical but is a major focus of AI research.

Also Read | The Humanoid Revolution in India Just Got Real

Not the First Time AI Has Improved Itself

Zuckerberg’s observation follows earlier research into self-improving AI. In October 2024, a team from the University of California, Santa Barbara, published findings on the “Gödel Machine” concept — a theoretical system that can rewrite its own code and processes if it can formally prove the changes will be beneficial.

Their experiments with a “Gödel Agent” demonstrated self-improvement in tasks such as coding, science, mathematics, and reasoning. Unlike most AI models, which are restricted from altering their own code, the Gödel Agent had complete access to its source code and improvement mechanisms, consistently outperforming human-designed agents.

Also Read | Could Indonesia Become Southeast Asia’s AI Powerhouse?

Promise & Caution on the Path to ASI

Zuckerberg suggested that ASI could unlock groundbreaking discoveries and herald a new era of personal empowerment, giving individuals more control to shape the world around them. However, he stressed that Meta would be selective about releasing such powerful systems under open-source licences, citing the need for caution.

“I am extremely optimistic that superintelligence will help humanity accelerate our pace of progress,” Zuckerberg wrote. “But perhaps even more important is that superintelligence has the potential to begin a new era of personal empowerment where people will have greater agency to improve the world in the directions they choose.”