> The explanation is something like "we extrapolate from the way that rationalists think and realize that their philosophy leads to dangerous conclusions."
I really like the depth of analysis in your comment, but I think there's one important element missing, which is that this is not an individual decision but a group heuristic to which individuals are then sensitized. Individuals don't typically go so far as to (consciously or unconsciously) extrapolate others' logic forward to decide that it's dangerous. Instead, people get creeped out when other people don't adhere to social patterns and principles that are normalized as safe in their culture, because the consequences are unknown and therefore potentially dangerous; or when they do adhere to patterns that are culturally believed to be dangerous. This can be used successfully to identify things that are really dangerous, but also has a high false positive rate (people with disabilities, gender identities, or physical characteristics that are not common or accepted within the beholder's culture can all trigger this, despite not posing any immediate/inherent threat) as well as a high false negative rate (many serial killers are noted to have been very charismatic, because they put effort into studying how to behave to not trigger this instinct). When we speak of something being normalized, we're talking about it becoming accepted by the mainstream so that it no longer triggers the ‘creepy’ response in the majority of individuals. As far as I can tell, the social conservative basically believes that the set of normalized things has been carefully evolved over many generations, and therefore should be maintained (or at least modified only very cautiously) even if we don't understand why they are as they are, while the social liberal believes that we the current generation are capable of making informed judgements about which things are and aren't harmful to a degree that we can (and therefore should) continuously iterate on that set to approach an ideal goal state in which it contains only things that are factually known to be harmful.
As an interesting aside, the ‘creepy’ emotion, (at least IIRC in women) is triggered not by obviously dangerous situations but by ambiguously dangerous situations, i.e. ones that don't obviously match the pattern of known safe or unsafe situations.
> Sometimes people don't or can't practice this protection for various reasons, and that's fine; it's a local problem that can be solved locally. But it's very insidious to turn around and justify not practicing it: "actually it's better not to behave morally; it's better to allocate resources to people far away; it's better to dedicate ourselves to fighting nebulous threats like AI safety or other X-risks instead of our neighbors".
The problem with the ‘us before them’ approach is that if two neighbourhoods each prioritize their local neighbourhood over the remote neighbourhood and compete (or go to war) to better their own neighbourhood at the cost of the other, generally both neighbourhoods are left worse off than they started, at least in the short term: both groups trying to make locally optimal choices leads (without further constraints) to globally highly suboptimal outcomes. In recognition of this a lot of people, not just capital-R Rationalists, now believe that at least in the abstract we should really be trying to optimize for global outcomes.
Whether anybody realistically has the computational ability to do so effectively is a different question, of course. Certainly I personally think the future-discounting ‘bias’ is a heuristic used to acknowledge the inherent uncertainty of any future outcome we might be trying to assign moral weight to, and should be accorded some respect. Perhaps you can make the same argument for the locality bias, but I guess that Rationalists (generally) either believe that you can't, or at least have a moral duty to optimize for the largest scope your computational power allows.