Our memory act as a very powerful database, being able to store a huge load of data. Thing is, that "instinctive data" you learned someday is still there. It might get erased eventually, but as it is "fetched" and used, it gets stronger.
Memory retrieval act akin to a computational weighted-graph navigation, where once you need to remember something, you start navigating in your memories. Memories with greater "weight" or "importance" may be fetched earlier and given a stronger "stamp", i.e., they are more relevant in a given context.
CONSTANTINI, WIKIPEDIA *sigh*:
Learning and memory are direct consequences of attentional processing: the more attention is directed or devoted toward something, the more likely it is that it will be later learned and remembered.
SURESH, PAVAN KUMAR, AISHWARYA:
[page 2, §2] Another example which everyone can associate or relate to is remembering an almost forgotten event after a very long time. It so happens that, when we come across a bite of a cookie after decades or smell of a perfume after a long time, we take time to recall what it is about and our association with it.
[page 3, §3] [In a face querying context] In the Memory Repeat, the brain repeats the same querying activity until it deduces a relationship between the second image and the fifth image. Once a relationship is establishes, corresponding activities that are chained/outlined to form a bountiful array of events.
Based upon the data present at the back of the brain, relevant output can be expected. The output might be an association of that image with certain events. It can either be the first time they met or the assignment that any
had done together or a project/ paper that they had worked upon or classes they had attended, etc.
It should, however be noted that not all images can yield a positive result. Some associativity (read: data) may have gotten erased as the passage of time.
A positive result is named as Hit Rate and False Alarm Rate. Hit rate is the positive outcome of a query while False Alarm Rate is the irrelevant or negative outcome of the querying. False alarm rate is the output when an image (or a face) is identical to another image.
If the data is fetched, but never used, it will eventually wear off from the results, but it will only disappear from your brain upon physical memory removal, which takes time.