AI doom usually imagines AI evolutionary fitness going to infinity, with human fitness staying fixed. From this comes "we need to merge", etc. But there are many ways we can increase our effective relative fitness without drastic changes to neural architecture. This is what culture does, for example. This is what engineering does, for example. And even if your framing is this evolutionary one, and you're thinking about longer timescales, we have a huge head start. Greater intelligence doesn't computationally necessitate greater fitness. We have to figure out how to disentangle these more than we currently know how, yes, but they aren't totally and perfectly entangled in all worlds. Insofar as it's something we have to deal with at scale, which there is mounting evidence we do, strong instrumental convergence is ultimately an engineering skill issue. We will be able to simply program these systems much better than we can today.