It's worse than that, those "algorithmic" interviews are testing IQ, more-so than knowledge. The hard knowledge that is tested is pretty much only the syntax of the programming language candidate decides to solve the interview question with.
I put "algorithmic" in question marks, because those aren't even that hard as problems in programming contests or questions from uni comp sci coursework. For a phone screen with an online doc, it can be even something concrete, maybe implement a simple feature of a text editor (does not have to be as complex as the online doc the interview is in - it's important that the subject matter is easy to understand).
For an onsite interview, it would be something not likely to have been practiced in advance (questions leaked to websites get banned). Still, it would be something that reasonably fits in RAM, so there's no need to use a database or any networking technology (we can discuss those, too, just not necessarily request to use it in code - similarly, it may be taken for granted that the input data will be valid and there's no need to code defensively, the interview duration time is kinda fixed).
But as a policy we don't tell the candidates what their mistakes were, so those who failed may never know what they were up to. Goes without saying, we get to interview software engineers from different backgrounds - physics PhD s, or high-school IMO medalists, too.