US and China race to build self-improving artificial intelligence systems

The race to develop artificial intelligence systems capable of building and improving themselves has emerged as the defining frontier in the US-China technology competition, with both nations viewing the capability as potentially transformative.

Autonomous AI development becomes new battleground

Leading AI companies in both the United States and China are increasingly deploying advanced models to write code, design experiments, and develop training techniques aimed at creating more powerful successor systems. The industry has termed this ambition recursive self-improvement, or RSI, the creation of AI systems that can train themselves without human intervention.

The technology holds the potential to fundamentally reshape the AI landscape. As AI systems contribute to building more capable successors, which in turn become better at improving the next generation, the process could create an accelerating feedback loop that propels the evolution of AI onto an entirely new trajectory.

OpenAI outlines research goals

In a blog post last week, OpenAI stated it aimed to build an automated AI researcher capable of advancing deep learning research, though the company acknowledged it remained unsure "how to safely get all the way to aligned, full RSI". The company described its GPT-6 Astra model, released in early September, as "the world's most intelligent and aligned model".

Both the United States and China view recursive self-improvement as a game-changer in the AI race. The competition raises questions about which side will achieve the capability first, and whether the technology will pose broader risks.

Source: South China Morning Post