I'm hardly the king of databases, but always using a surrogate key (either an auto-incremented integet or a random uuid) has done me pretty well over the years. I had to engineer a combination of sequential timestamp with a hash extension as a key for one legacy system (keys had to be unique but mostly sequential), and an append-only log store would have been a better choice than an RDBMS, but sometimes you make it work with what you have.

Natural keys are almost always a bad idea though. SSNs aren't natural, which is one pitfall: implicitly relying on someone else's data practices by assuming their keys are natural. But also, nature is usually both more unique than you want (every snowflake is technically unique) and less than you'd hoped (all living things share quite a lot of DNA). Which means you end up relying on how good your taxonomy is for uniqueness. As opposed to surrogate keys, which you can assure the uniqueness of, by definition, for your needs.