The parent comment of yours identified the right need for them (a person who can come up with a solution when the time comes in the real world), but the method of toy algorithms means you could end up with the following scenario:
Person A: Spends months drilling solutions on whiteboards to memorize the answer. They pass as they can regurgitate every toy algorithm possible without thinking. They pass the interview.
Person B: doesn't deal well with interview situations and doesn't get the chance to talk in general about how they'd approach it in real life. They freeze up as they're stood up in front of a panel of people holding a whiteboard marker and being stared down. They don't progress in the interview.
Perhaps person B was the person who would sit at their desk for 20 mins in silence and come up with:
"ok if we're not deduping these on the inserts for our system, I guess the least we can be doing is maintaining one of those in-memory bloom filter things, I read about those one time as a good probabilistic data structure, might end up being a part of solution. I'll ask the tech lead if we have any memory constraints as I see that our AWS instances are compute optimised instead of memory optimised, anyway, we might get an off the shelf solution"
Person A: "which question in my 'interviewing for algorithms' text book does this belong to?"
You want Person B, you optimised your interview process for A.