On 7/7/23, Undescribed Horrific Abuse, One Victim & Survivor of Many <gmkarl@gmail.com> wrote:
On 7/7/23, Undescribed Horrific Abuse, One Victim & Survivor of Many <gmkarl@gmail.com> wrote:
regarding focus on a tree rather than a prompt, we might imagine instead some form kf access or memory present in the system. a robot for example might travel near important information, or pursue a step toward a goal, or an agent might be in or not in a state where a goal is already partly met
the tree is roughly the decision tree i think which i have somewhat-mysterious-inhibition around. it can go different spots.
for example if we are playing tictactoe and want to win, we might consider all the different moves we might make. this list of moves forms the second row of a tree of parts of the world being considered, kinda maybe rooted in what is certainly known about the present moment which has only one option. then for each of those move states we consider other things such as what the other player might do, or how judgements apply to the move based on things known … … at some point in this considering we form judgement or find states where the game is won or lost.
regions or possibilities in the tree where the game is lost fail to meet the goal of winning the game and consideration is no longer proceeded there and the areas need no longer be held in memory, it’s maybe important to remove them to ensure they are properly nonincluded during furhter c9nsideration (like an assertion check), assuming they can always be reconsidered to rebuild the same data
meanwhile regions or possibilities where the game is won meet the goal of winning the game. the tree can be considered a planning structure for meeting this goal, finding a path from the present moment to the goal, start8ng from the present moment for efficient information access if there is time pressure
i imagine it can also be quite useful to start in the middle especially in large spaces, this might assume something is known about usual ways of meeting the goal in the environment, or at the end if more is known at the end, and it’s nice to imagine including all three in some way, unsure. basically new information might crop up anywhere i suppose.
it’s interesting to think of probabilistic information on various areas being summaries or guesses of hidden expansions in detail
something i eventually found nice success around here was writing a tree exploration algorithm capable of changing where it was in the tree as priorities of different branches changed during exploration. this was pretty inhibited for me! i wrote a couple and i had chatgpt write some and i feel much more comfortable with it now! took some years. i still don’t use other parts of tree structures unfortunately and kinda strangrly. i think i probably emgaged that part because i associated it both with data structures and with AI, so it had maybe double the inhibition for me. i thought it would open up trees more, but maybe implementing it more would also help … it feels more inhibited after writing this