Why We Remember So Many Things Wrong
newyorker.com
newyorker.com
Ever had someone with "photographic memory" pay attention to your daily life? It can get creepy very quickly, because she notices and vividly remembers all sorts of things that you never imagined anyone would ever notice. But nowadays, with Facebook and Instagram and sophisticated tracking everywhere, all sorts of strangers have literally photographic memory of what you said and did last summer. I've seen lots of people obsessively delete stuff from their timelines after only a short while, not because they're paranoid, but because feeling uneasy is only a normal response to any violation of our intuitions about how memory is supposed to work. (Unfortunately, our normal response doesn't have normal consequences anymore; the data still exists somewhere.)
We're meant to forget. We're made to rewrite our memories as time goes on, just as any living thing constantly rebuilds itself over time. Homo sapiens never evolved to be reliable preservers of pixel-perfect information. No matter how much our technology and legal framework tries to promote 99.999999999% durability as the golden standard of memory (Amazon S3 actually promises eleven 9's), we need to remember that this concept of memory is a very modern invention that most people still have trouble coping with. Maybe we never will. Maybe we never should. Maybe computers will learn to forget like us, rather than us learning to remember like them.
And it seems that creeping deep through someone's social timeline is fast becoming a faux pas. You already avoid doing it, and if you do end up doing it, leaving a trace is mortifying.
To the point that people will never mention to you in person that they saw something from your profile that happened a long time ago. As someone with a strong-ish internet presence, it's funny to see how flustered people get when they accidentally reveal themselves to know more than they feel it's polite for them to know.
Same way as we've always had the norm that you let people tell their stories even though you've heard the same story ten times already. It's just polite.
What that doesn't fix, however, is the fact that old information is still out there, on somebody's server, every bit unchanged and ready to resurface at a moment's notice. Sometimes the old information is just a statistic in the grand scheme of Big Data. Other times, it can be very specific, such as the one-liner that allowed the Feds to attribute to Ross Ulbricht the very first mention of Silk Road on the open Internet.
Computers aren't learning to forget, they still remember everything. We're just training ourselves to pretend that computers are just as forgetful as we are, because it's a comfortable thought. The problem, of course, that collective games of make-believe have a tendency to break down at some point.
And this seems like an annoying bug, not a feature. Internet is becoming more and more of an ephemeral place. That interesting news article you read few years ago? It's URL has probably changed twice, then it was deleted, and even if not, unless you remember the exact wording of the headline, you probably won't find it on Google as it will be buried under three pages of recent tweets that match your search query - tweets no one will care about or remember next week.
People do want to remember some things perfectly - that's why we invented writing, textbooks, diaries and libraries.
Meanwhile anything that might actually be worth reading is somewhere around page 10...
...much like the way things were in 1997 with Google's predecessors. Funny how despite exponentially more layers of code running (with eons of iterative advances), it all still somehow comes full circle.
For instance, imagine an embarrassing picture that shows you doing something stupid in an obviously intoxicated state. OK, so you were very drunk one night in 2005. So what? Millions of others were too. But what if everything publicly known about you could be mined for clues as to whether or not you have an alcohol problem?
I think that is what we should think about every time some company tells us "we don't ever look at your data, only our algorithms do" while the same company is busy working on AI stuff that draws conclusions from data.
And as for optimality, be careful with statements like that (and what you wish for). I think it ceases to be an optimality problem when it's followed by "if everything were completely different."
Where this breaks down is when features get updated between revisions. If your game patched the "jump" function to increase upward momentum from 1.1 m/s to 1.13 m/s, the Replay would be incorrect. You would be jumping onto platforms you couldn't get up to before, moving faster, maybe even dodging enemy attacks that hit you when you played that match.
The human neuroprocessor is always changing and growing, always revising itself. Thus memories replay incorrectly. You apply old feelings to new mental patterns, and sometimes they lead to weird places. Or sometimes you mistake something easy for being difficult, because your memory data is out-of-date for your current processes.
(The usage isn't quite the same as in the jargon sense of "f is idempotent in case f(f(x)) = f(x)", and I thus found it jarring as well, but it's possible to see the connection in that above gloss.)
Because it takes so long to think, and probably because the amygdala is so small and simple compared to the neocortex, the visual senses we receive hit the amygdala well before they hit the visual cortex. We experience pain and pleasure memory response from senses before we even know what those senses are.
You may have two memories of a horse, one just a plain-old horse, and one a horse while you were getting shocked. Since the amygdala senses the horse first, the emotional memory is returned first, which is why any attempt to bring up "horse" will dig up one of the emotional memory horse records. You then have to wait for the neocortex to catch up and say, hold on a second, this is just a regular horse. When your neocortex is completely swamped by the amygdala it is called "amygdala hijack".
The quoted paper is basically saying "the hippocampus and amygdala can affect each other", which is basically true: usually one does not rule the other, and they work in concert to give you a full picture of things. And just like the amygdala can provide an emotional veneer over logical memories, the hippocampus can temper the emotional response, for example with mindfulness practice.
Since human beings are still evolving, I don't think there's any reason to believe we're meant to do things one way rather than another.
Edit: I should note that hyperthymesiacs are not immune to false memories, despite their superior memory: http://www.pnas.org/content/110/52/20947.abstract
Another interesting read on it: https://www2.bc.edu/elizabeth-kensinger/Kensinger_CD07.pdf