Knowledge wherein causality is, at most, implicit
Ex. write a story given a picture
Can we just use similar mapping of actions to picture w/o abstracting causal patterns
AI agent creates propositional representation out of visuospatial reasoning -> Archydas
Multiple levels of abstractions including some invisible layer that needs to be inferred
SBF, but not the bitcoin one...
Level 1: what does a flashlight do
Level 2: structure of the flashlight
Level 3: behavior - state-transitions-state transitions