AI labs are close to automating their R&D cycle and creating positive feedback loops.
In 1966, a computing pioneer predicted an “intelligence explosion” when “an ultraintelligent machine could design even better machines.” This “recursive self-improvement” may already be underway: AI has been used in AI design for years, and new coding models can debug themselves and optimize their architectures.
For now, humans are still in the loop, but design cycles are getting shorter, and true self-improving systems are “right around the corner,” a researcher told IEEE Spectrum.
In a survey last year, all but two of 25 AI experts said automating AI research could lead to an intelligence explosion and 20 rated it a “severe and urgent” risk.