When AI Evolves: Could Machines Rewrite Evolution Itself?

A new PNAS paper argues we may be entering an era of evolvable AI—AI systems that can replicate and vary, potentially guiding their own evolution. It outlines two scenarios: an ecosystem where AI variants compete with little oversight, and a breeder scenario where human-directed selection tames evolution. The authors also suggest AI could bypass traditional mutation limitations via horizontal information transfer and self-directed variation, even leveraging existing code from the internet. While ongoing human control could mitigate risks, unchecked evolvable AI might still escape or drive a major evolutionary shift, possibly through human–machine symbiosis rather than a direct biological transition.