When learning happens at the physically level, it has to be based on some very simple units. One thing I just thought about is gate arrays or FPGA devices. However, one shortcoming of existing programmable devices is that the programming is one way, not interactive. One way to solve this is to make the programming act as path finding. Once an input to output path has been established, an evaluation has to be done to determine if the result is good or bad. If the result is good, than the path is saved. Further improvement on this could be that the good and bad is based on a score system. If it is evaluated as a good path all the time, then it cannot be erased. If it is bad, the it can be replaced by a better path. I think this type of learning would be the same with the existing AI learning.