I had a similar idea a few years back for robotics that I jotted down in a note, I'll copy it verbatim here. (It's not rigorous in any way, just sort of a possi

I had a similar idea a few years back for robotics that I jotted down in a note, I'll copy it verbatim here. (It's not rigorous in any way, just sort of a possible vision for the future.) "For effective AI (for robotics specifically), we need to let separate, hardware driven, purpose built OSes run sensing equipment, and then only use a central OS fusing signals as a moderator/clearing house for those signals. Each separate OS should have autonomy over a certain set of functions to varying degrees, and should be overrideable only by strong consensus of the other systems. The important point is that the central OS not act like a traditional scheduler telling the camera and feelers and whatnot what to do all the time. The nature of the machine will instead be defined by complicated subsystems being moderated, but not fully controlled by a central authority, much like the brain/nervous system."