Red queen hypothesis – A new way forward for self-improving AI