It’s a well known fact that people are unreliable when it comes to knowing what they want, and what they will do.
A human can decide to do something, and then fail to follow up on that decision, often without realising it until much later. Furthermore, they can decide that they won’t do something, only to go ahead and do so. As a species we are pretty bad at predicting our own behaviour, because we often do not realise that our predictions have been so off-base.
If UX researchers could routinely only have access to fMRI scans of users, I wonder what we would learn.