There is something very useful about idea of innate or ‘atomic’ meanings in linguistics while grappling with AIs. (This is, of course, model-relative, existing in learned geometry of representations )
In linguistics a word whose meaning is innate or atomic, is one that is not reducible to other words. Kind of the prime numbers of word galaxy.
And that is essentially how an entity can be thought of in interpreting AI actions. It is what is, can’t confused or reduced to something other.
To an AI model, an entity is irreducible.
If you think your target is atomic, you’re on track.
#hiddenstatedriftmastermind