Rescorla & Wagner (1972):
How does this model tie with the presupposedly existing domain-specific, innate modules? Would domain-specific expectations (e.g. principles of object solidity, spatio-temporal continuity... etc.) be included somehow in the weight parameter of association? In other words, where does the expectation that a toy train cannot levitate affect learning about this magical train in the Rescorla-Wagner model?
Lake et. al (2017):
Would I be oversimplying it (a lot) by thinking that intuitive theories can be inputted in AI through simple "rules" (e.g. Objects exert force given gravitational pull relative to mass....)? Would a dictionary/collection of "basic rules" suffice to build an intuitive causal model of the world? And if so, what kind of counterfactual/causal model should be used to govern these rules?
P.S.: As Dennett once said, to which I agree, AI should remain "smart machines", not artificial collegues.