AI Systems Have No Hunger: A Thought Experiment on Darwinian Alignment

If I understand correctly, you’re asking whether such a system, with AIs behaving according to classical evolutionary rules of natural living beings, could lead to problems like “cognitive cancer.” If so, it’s not a stupid question at all, because AIs are notoriously verbose and prolix when writing code, by their very nature. So, in a system where they continually self-reprogram for survival, abnormal growth of redundant code, detrimental to the model’s efficiency, could become a serious problem.