AI threatens to orchestrate sustainable social reform would belong in the 2025 review, but may I suggest adding a new kind of agenda to your taxonomy for 2025?
Most of your current categories focus on technology, but this article focusses on safety, on the nature of our self-destruction/warfare, and explores what is needed technically from AI to solve it. It sees caste systems that predate AI, notes that they are dangerous (perhaps increasingly dangerous) and how to adjust AI designs and evaluation processes accordingly.
Perhaps the title of the agenda could be "Understand safety" or "Understand ourselves", and the increase in social impact research could reflect here.
Having documented epistemic consequences of realist metaphysics (https://doi.org/10.6084/m9.figshare.19204896.v1), it has been pointed-out to me that even non-realists might endorse them. That should be no surprise since non-realists are in community with realists, and sharing epistemic practices (like sharing language) makes community easier. No doubt, realists would likewise endorse non-realist practices, in return, if non-realism made any such practice necessary.
What makes this observation relevant to this article is the potential for non-realist positions to become irrelevant in practice because realist epistemic practices so deeply shape the practice of rationality. Community is an asset, so we can expect the best ASI to foster community, and for such community (including the ASI) to engage in realist epistemic practices. One might temporarily isolate into a “truth” selected to bring comfort, and one might even win political power by facilitating such comfort, but community division is unlikely to be sustainable. In other words, realism might inevitably end-up treated as true, even if false.
To get the “orthogonality” part, I think the definition of the thesis also needs to include that increasing the intelligence of the agents does not cause interpretation of (some) goals to converge.
In particular, the dismissal of the concern that policy must include an absolutely perfect specification of the perfect goal does not deny that an agent could have a goal to maximize paperclip production, but rather asserts that the paperclip maximization goal may seed ASI adequately because a perfect intelligence pursuing it would behave the same as a perfect intelligence pursuing the perfect goal (although we imperfect intelligences do not realize this because we do not appreciate all the overlapping instrumental goals that both entail—for example, truly intelligent paperclip maximization may start with generating a maximally intelligent planner, and that may take so long that no actual paperclip get made).
I like this extension of The Alpha Omega Theorem away from the most simplistic God threat. If we extend far enough, then maybe reality counts as an Alpha Omega and the Alpha Omega hypothesis combines with Moral realism to entail that smart-enough superintelligences would follow moral laws for the same reason why they would open doors before trying to pass through: because moral laws are real and reality always defeats those who fight it.
On the flip-side, if it so happens that the morally-better behavior would be to let humanity go extinct--if it appears that the Alpha Omega has set up a progressing system in which we are just the most recent version of the dinosaurs and it is only through our death that our worst norms (racism?) will ever fully fade away--then should we ourselves dare to defy the Alpha Omega by trying to preserve our existence?
It cuts both ways, but @Darklight offers a nice piece of logic. It shifts my priority from trying to figure-out how to survive to trying to figure-out what the real moral laws happen to be.