What or how do you do a not example category

This is a tough one…
the best i can do is draw an analogy…

how do you train a net to know whether something is new vs known?
you have a database of known… but you have no such set of samples of new or unknown to the database
the closest is the idea of orignal knowlege vs known or unique vs known…
like when someone thinks of all the answers in a search for finding answers not thought of yet

given its so hard to state, it doesn’t help trying to thing of a way to do it

i guess what i would want is you put a sequence in and want to get out whether its known or close to something known, or if not, some indication that its unknown to the system…

the set of known is the easy part, but there can be no real set of the unknown or more rightly, not known (to the net).

this is not an attempt to train originality!
but more the idea of trying to induce in the net thinking of things it lacks experience of

its a real world problem that borders on the philosophical…
as humans we can target the negative space by way of the positive space
but a net?

right now i have no idea of how to approach this given that i have the known set which numbers in the millions and would like it to say whether something is in the known set or close to a known… or not at all known (in relation to the set)