AI labs are automating their research cycles, with systems that can debug and optimize themselves as humans step back from the loop.

AI has been used in AI design for years, and new coding models can now debug themselves and optimize their architectures. Humans remain in the loop for now, but design cycles are getting shorter.

A researcher told IEEE Spectrum that true self-improving systems are "right around the corner." In 1966, a computing pioneer predicted an "intelligence explosion" when an ultraintelligent machine could design even better machines—a concept called recursive self-improvement.

In a survey last year, all but two of 25 AI experts said automating AI research could lead to an intelligence explosion, and 20 of them rated it a "severe and urgent" risk.