I generally ask an algorithms question not because I care about the algorithm its self (spoiler: it's not even about datastructures), but about how the candidate comes up with test cases and error conditions. Truthfully, my 'algorithms' question is more gauging their effectiveness at QAing their own code.
In general though, I call those sorts of problems CS trivia. At best, they act as a proxy gauge of skill. The theory is, any 'good' programmer can fall back on their experience and solve problems on the fly. Some candidates keep their cool and methodically solve the problem. Others would need time in the best of cases.
To conclude, I think it's just generally the lack of available skill to interview and general risk-aversity. Hiring a bad employee is an expensive mistake, and in a startup, it could be one that kills the company.