Who do you think are these researchers? More ML or more neuroscience?
///These disciplines are very relevant to understanding how memories are stored. More relevant than neuroscience in fact. Say, if you'd study modern computer vision, you will have some understanding about mechanisms in the works of human visual subsystem; about how humans memorize, recognize and associate images and image sequences. You will not have that understanding, if you'd study modern (basic) neuroscience.
And you actually might learn some real stuff from people like these guys from UC Berkeley, who can recover video sequences by identifying an ML model with the visual cortex an then proceeding further and recovering unseen video sequences from fMRI of V4 area of neocortex.
... Now I only wish I could find a forum where a link to a research publication like this:
* https://news.ycombinator.com/item?id=3031698 [Reconstructing visual experiences from brain activity evoked by natural movies]
Would not have an order of magnitude less interest than a link to a slate article like this:
* https://news.ycombinator.com/item?id=6349590 [ How Can You Tell a False Memory From a True One?]
The mechanisms of memory are not, as far as I know, well understood. However, it's well established that people can and routinely do remember things that did not occur. The why and how may be speculative, but the fact that it happens is not.
Here is the link again: https://news.ycombinator.com/item?id=3031698 [Reconstructing visual experiences from brain activity evoked by natural movies]
Hint: how can you explain that sometimes, the recovered image contains an object completely different from that seen by a subject of the experiment? [say, tree instead of a man]
Hint: can similar experiment be done with the part of the pre-frontal cortex responsible for associating visual memories with the condition in which these memories were formed? [imagined, dreamed, seen]
What I find to be idiotic is how trivial you make your contradiction, without any facts to back it up, adding computerish language to pander to the HN crowd ("metadata", really?), when it's extremely well established that people, with disturbing frequency, remember things that simply did not happen.
It's fine to disagree, but this sort of straightforward contradiction adds little to the conversation. It's the internet equivalent of Monty Python's argument clinic.