Yet more notes to self
The main AI idea that will enter the dissertation:
—Mood in most general sense as Deep Learning super high order feature (which, like always in Deep Learning high order features, might express something causally primary — like in the lighting case in image analysis example in Deep Learning textbook).
—Modernist collage as training set for concept-learning (‘what interesting class do all these items belong to?’)
— Symbolist ‘symbols’ as prototypes of the mood structure, for prototype-based clustering of things that embody the mood structure
— Exformation surprise-that-it’s-not-a-surprise as proof of extensiveness of information contained in a mood
— Uncanny textual objects (Pinter, Langpo etc.) as a different technique for same, by breaking triangle inequality — same as Symbolist ‘symbols’ but involving big bold concept-creation rather than refinement