Associative memory models

I was reading various translated articles in Russian, and found this one (linked to translation), talking about unusual types of computer memories:

http://bit-player.org/2018/the-mind-wanders

two parts caught my eye:

From a computer-engineering point of view, this version of sparse distributed memory looks like an elaborately contrived joke. To remember 10,000 items we need a million hard locations, in which we store a thousand redundant copies of every pattern. Then, in order to retrieve just one item from memory, we harvest data on 11,000 stored patterns and apply a subtle majority-rule mechanism to unscramble them. And all we accomplish through these acrobatic maneuvers is to retrieve a vector we already had. Conventional memory works with much less fuss: Both writing and reading access a single location.

With a conventional computer memory, you can do a core dump: Step through all the addresses and print out the value found at each location. There’s no such procedure for a distributed memory. I learned this troubling fact the hard way. While building a computational model of the SDM, I got the pieces working well enough that I could store a few thousand randomly generated patterns in the memory. But I could not retrieve them, because I didn’t know what to ask for. The solution was to maintain a separate list, outside the SDM itself, keeping a record of everything I stored. But it seems farfetched to suppose that the brain would maintain both a memory and an index to that memory. Why not just use the index, which is so much simpler?

.

I personally don't understand mathematical concept behind all this, yet it was interesting to read, because it sort of resonates with my own skepticism about building exact digital 'copy' of someone .... I remember I was reading about associative memories in some 80x era russian book, yet details of how it was supposed to work escaped me .... They still escaping me, but at least I had some interesting reading for night!