Now, guesses should never be presented as absolute facts, but having a guess puts you much closer to the truth than having nothing. If you really, truly have nothing, no relevant knowledge---which should rarely happen if you're well-educated and/or experienced, in which case you're a novice to the subject, where somewhat different rules apply---then say so; otherwise, go with what you know and learn the rest on the way. Will mistakes sometimes happen? Will you sometimes head down dead-ends? Absolutely, but that's part of the learning process and we shouldn't be afraid of mistakes, as long as we're in a process which is tolerant of them. (If you're pushing code to production, for example, then that's a time either to know or not to know, not to guess. But that's the end of the process, not when you're asking questions.)
So if you said to me, "I don't know", I might look at you funny because it's probably not true: you probably do know, at least partially; it just seems like you're not willing to acknowledge that.
Just say "I do not know" and do not waste your and my time with monologues full of guesses and sort-of-related-but-not-relevant info. It does not help. It just wastes time.
If you say, "I do not know" we can work on it. We can file it as an open question until you know. We can ask somebody else or search for information.
And btw, the answer to "Will mistakes sometimes happen" question is: guesses cause them very often and they slow analysis a lot. You just do not see it, because I can not go to you and show you all your wrong guess mistakes. I have to smile and pretend that I trust you while triple checking everything you say which again, slow down analysis.
Working with people able to say "I do not know" is way easier and faster.