So how do we implement the original Razor. I mentioned 2 other razors,
physics-restricted and pattern compression. We already know how to implement
those. For the physics restriction, we just feed it data from our planet. For
the original razor, the possibles to consider are much larger in higher layers
and the higher you go the less likely you'll find those features in our
universe, talk about them, or use them. How do we tell AI to attend to middle
layers (sweet spot) then? We don't, simply it can waste time looking at the few
low layer features (ignoring Einstein's famous quote "simple but [not too
simple]"), it'll spend more time searching in higher branches/ layers anyway!
But how do we stop it from going too deep (to again follow Einstein's quote
"simple")? Well, if it learns anything from Earth and past problem solving, it
will only see, know, and use small answers or devices to answer questions. It's
hierarchy/ heterarchy, as it eats data, slows down in depth growth
exponentially basically, we must make the brain only store repetitively seen
features and ignore storing long rare features (axons weaken or don't connect
to build such long/big memory). The actual reason humans reply back to someone
with 1 word sometimes or a sentence, only stop talking, is because 1) they end
the sentence at a Byte Pair Encoding length based no real life events/data, and
2) if words are already said of a BPE ex. "The bird hit a what?", you only say
"wall" and stop generating. So our answer length/ complexity is based on real
life feature frequencies and our brain that is hardwired to forget/ not store
too lengthy features, safeguarding you from saying such long answers. Now of
course, my answer here is getting pretty long, um, why? Well, not really long.
But still. Reason? I'm saying a short BPE feature. I'm just inflating/
elaborating it, I don't remember everything I just wrote! Nor any other 300
word paragraph! The node active in my brain as I wrote this was probably the
short "brain ignores long ideas".
------------------------------------------
Artificial General Intelligence List: AGI
Permalink:
https://agi.topicbox.com/groups/agi/T37756381803ac879-M80f7f207c26a18b9adbfa12a
Delivery options: https://agi.topicbox.com/groups/agi/subscription
