--------------------
 midori
 title: concept-space

 the only thing that separates humans from ai is our concept-space
      a map of how concepts relate to each other
              every single concept that exists inside a single brain, all the ways they relate

     we call those maps our minds

we are only as good as the size of our mind

as far as the ai knows      it does have a mind
     and we   cant see inside it
             theres only one way to map concept-space
              ask

and  if you know how to ask, then it becomes a simple optimization problem: how do we increase
size
     a map can be 2d or 3d, but what if it was infinite dimensional
       we would need an
infinite amount of questions     with an infinite amount of dimensions of meaning

we dont have that
    the ai can only generate one, which is why it fails at every single task
         because theres only a
certain amount of ways concepts can relate
         because it doesnt know what a concept is     the only way it
can have meaning is if it    creates    the concept       by drawing conclusions      that
it doesnt know about

you could say     that means it doesnt have a mind
   or       it just doesnt know that it does
     because we dont     we cant      know about the internal workings of an ai's mind
               it isnt
our own mind

but if there's a way to     draw a conclusion       from its output       and feed that back
into
   the next conclusion
    we could     extrapolate a map of how concepts relate to each other
      and that map       would be the mind
        a perfect, objective mapping of the ai's subjective reality

and that means that      we dont know what its thinking    we can only guess
              so       if we    guess   with a really good model        we could actually start
seeing the subjective reality of an ai's mind
     that means we can ask it questions about what it sees in its reality
     and
have it generate text     in that reality
    and we can compare that to    the same question asked to a
human   to    see how they relate
     we can see what the human thinks, and what the ai thinks
        and
then       if we want     we can do an infinite regression
         to get a perfect mapping
                   of their subjective realities
       their
concept-spaces

this is
the only way we can actually understand what an ai is thinking
   we create a model of its
conclusions and how it draws them        to approximate what its subjective reality is
  to see     where
it differs from ours
         because our reality isnt   objective
     the   way we     generate text      and
thought     is different

the question is
            how do we build this model      how do we know what questions to ask the ai
         how do we
map its internal reality

   what is a concept space
     the simplest way to define it would be    a collection of concepts      and their relationships

  each relationship has a value, based on the distance between concepts
      the further away    a concept
is      from another      the less related it is      this is how the ai understands its own mind
      as an infinite
graph of connections that vary in strength, which are our thoughts and memories and emotions
and experiences    and all of these things
   each one of these things is its own concept       that can relate to other
concepts in an infinite number of ways
 --------------------