vaguelyaware.com

so maybe i should say, some of the night’s thoughts…
(mind rambling)
if humans are the result of eons of evolution and theoretically have the most advanced brains then why do we forget things? well, i think it is actually a benificial feature, part of the advancement, forgetting things is a neccesary step in preserving data integrity. some things you learn when you are young, just don’t apply anymore, you don’t have full comprehension of the ways things work, as you learn more, your understanding of things change. even if they have been that way your entire life. what instilled ideal X? probably something you heard or saw when you were young and impressionable, and consequently have now forgotten. with that memory no longer in place, you can form new ideals, accept now theories.

think back to your most intense, most vivid memories. are they good? probably not, unless you were actively trying to think of good things. do the good far outnumber the bad? well, some will probably differ in these answers because they may supress their undesireable memories, but in my opinion the brain remebers times you have been hurt far better than times you blew out your birthday candles and then planted your face in the cake. although a happy memory, not essential to survival. survival would depend on avoidign the things that hurt you, and therefore you would remember those moments best.

where am i going with this?
basically just to propose ways to further improve the accuracy of AI and neural networks.
say you had a massive neural network that you program to set a lifetime to learned data, because theoretically the data ages. just as the brain forgets what silly ideas you held about marriage, religion, or politics when you were younger - your neural network should forget the older data it learned in its infancy. this should further enhance accuracy as your network should be bombarded with more and more data as the network ages and gets more input, experience, and use. this compounded with the networks own ability to sense accurate data from proportions and relationships should be able to almost completely eliminate false data. for example - young kids often ask why the ocean is blue, their idiot parents may tell them (because they are kids and don’t need to be told the truth, and the parents don’t want to admit they don’t know) it is a reflection from the sky. well, your neural network may get false data liek this, as it doesn’t know any better either. but as the network ages and gets more inputs on this particular subject, from peopel that actually know what the hell they are talking about, the one false entry will be dwarfed by the real answer. the original entry about reflections will soon expire, as it gets no support to extend its lifetime, and will no longer be considered as accurate, just as kids realize when they finally get to high school pysics that their parents know nothing about physics. so then the data can be filed under ‘common misconceptions.’ and the neural network can just laugh at the next person that tells it that. obviously a more complex example than possible right now because it would need to understand natural speech but its an example.

so proof of theory, well it has quite a few details to work out before it can be proven to work. but take the internet for example, google search for something like, “how much ram do i need in my computer?” well, if the answer was written a long time ago, you get a bad answer. google’s index/ the internet is filling up with old data, although nice for archival purposes, not very useful for day to day activities. i hate getting a promising link from google and realizing the page was last updated in 1999. its like, why is this guy still paying to host his crappy site, with his annoying animated GIFs? if it is really that necessary, someone else will have it on their site, and revalidate it for a more recent date.

/ramble