Not quite sure why I posted this this way to be honest... What I was thinking of, was a story that I can no longer find anywhere, that claimed recent research made a breakthrough showing that the efficiency of neural networks could be improved by teaching the system to look for relationships between numbers, for example, recognizing "3" as half of "8", not in a mathematocal sense, but in physical sense. The point is, using arbitrary symbols that have no relationship to each other to represent mathematical concepts such as integers taxes our brains and artificial intelligence systems in a way that is unnecessary and inhibitiory. Using symbols that can be recognized as a series of patterns should improve the capacity of deep learning systems to process information. Rather than being snarky as in my original post, I should be clear that my knowledge is limited but I do believe I am onto something. I am glad you find the concept interesting, let me know if you have any insights.