Konrad you might be interested in reading Hubert Dreyfus' What Computers Can't Do as l well as the short book my Winograd/Flores. About abstracting knowledge into rules, etc. I read them decades ago (gasp) but they're likely relevant still. Maybe you've read them.