Yup, new discovery. So cat=dog by shared paths, many, of varying strengths. It
beats the candidate cat=man by a bit. These channels leak reward to the new
node dog more than to man. Similarly, once we have many triggered nodes from an
input ex. cat, ex. dog, man, horse, cloud, these now act as connectors like
that connected cat2dog except they send channels forward syntactically, cat
ran, so ran may have many and of varying strength links, jumped may have less,
and so on. So 'ran' is very triggered by multiple shared paths. Not only is it
like Semantics, it also allows reward/attention checkpoint updating; the
sentence extending itself forward can too infect reward to the future sentence.
This word extending is important because single words are not supposed to be
the only hierarchy nodes that have attention reward on them, phrase nodes
should too. Why? Because like in Blender, we control attention by always/
permanently pushing some favorite words like Horror or Cute or Cats, but if we
Force "kiss cats" then we have order and so it always focuses on kissing cats,
whereas "kiss, cats" would make it talk about possibly cats doing the kissing,
not us kissing them! So, extending a node and building a larger phrase, can
often invite over reward and not just energy to activate it/build it.
------------------------------------------
Artificial General Intelligence List: AGI
Permalink:
https://agi.topicbox.com/groups/agi/Ta79d68474896311f-M5a41c63016ee63ff20f7b092
Delivery options: https://agi.topicbox.com/groups/agi/subscription
