As a layman, it seems to me that the biggest advantage that humans have over machine learning is flexibility. While machine learning performance frequently riv

As a layman, it seems to me that the biggest advantage that humans have over machine learning is flexibility. While machine learning performance frequently rivals or exceeds humans at many individual tasks once sufficiently constructed and trained, only humans excel at dynamically choosing which tasks to pursue, switching levels of analysis, and when to break the rules for the win (losing at GO? unplug the computer). To speculate heavily, animal based cognition may be composed of just such a multitude of specialized trained modules, akin to machine learning algos of today: object recognition, emotion recognition, language recognition, typical script/structure of a given scenario, etc. But above that will be classifiers that interpret internal and external environmental signals to choose which of those specialized modules to engage and suppress. In lower animals lacking a heavily recurrent prefrontal cortex, the higher order modules are probably directed by mid brain structures to engage basic fight flight fuck behaviors needed for survival (e.g. pattern recognition module sees snake, freeze or run modules are engaged). In animals with prefrontal cortex, goal and context driven suppression of prepotent responses becomes possible. Anyway, it seems to me that for machine learning to become a general intelligence, there will need to hierarchies of specialized machine learning classifiers, some specialized in sensory classification, but others in that are meta..classifying those classifiers into scripts, scenarios, etc.