I am not talking about implementing an AVL tree without any help or resource. That is simple stupid, we have internet we must use it and save time. My point is that you need to know that there are balanced trees, know some of them and they characteristics. That knowledge will save you time when you face problems related to trees. You are not going to know all the trees because researches are working on new ones every day, but you should have some knowledge in the area.
If you interviewer drop you because you hesitate in something, maybe it is the interviewer problem and not yours. We are not perfect, we should know people will make mistakes and hesitate. I was rejected of interviews when I hesitate when they asked something about linked lists. I was happy it happened, because I was in shock when I was explanning how I did something with and A* algorithm and a octree and I couldn't believe the next question was about a linked list. I wouldn't be happy in that company.
You're attacking a strawman which has little to do with the original post or what the other commenter is saying. There's a lot of developers out there who lack the knowledge to ask the question you pose in the first place, so wouldn't end up doing that searching. The type of interviewing you're criticizing is intended to identify developers lacking the kind of base knowledge needed as a foundation for further inquiry.
No, what would happen is you'd say to yourself "Hmm, looks like I need a self-balancing tree. Haven't thought about those in N years..." and do a few seconds of keyword searching. The idea that people will be utterly helpless on their jobs without instantaneous photographic recall† of AVL tress (and A-star, and all the other crap people are bullied into memorizing these days) just doesn't hold water.
† Which, as we know, is the only level of recall acceptable in the modern interview process. Even momentarily hesitating in your recall of certain definitions will easily get you flushed by some interviewers.