Evolution is a subtle paring knife. For many evolved traits, it’s not obvious how and even if they contribute to fitness. Some traits are neutral. Others have not evolved because they affect survival, but because of sexual selection. Here is one example of a trait for a complex for which it’s unclear if it contributes to fitness, and here is another.
For human behaviors especially, if you think you can tell which traits increase fitness and which decrease it, you’re probably wrong. That’s why it’s absurd to argue that humans not having as many babies as we possibly can is a sign that evolutionary alignment has failed, and that therefore AI alignment is going to fail as well. I first came across this idea here and here, and I felt the need to comment since I have some familiarity with evolutionary theory.
I have no idea if AI alignment is possible. I also don’t think that human behavior is shackled to what evolution dictates. Instead, my point is that drawing parallels to human evolution is not helpful for answering this question.
Nehaveigur 