AIs need to teach by visualizing their search strategies

One of the common things that I find missing in most implementations of machine learning is a presentation on how The algorithms working, and how it has reached its results.

I think much of research and advancement of technology has come through huge trial and error being reducible into principles and simple techniques — often things that only came after great effort to see the big picture. Weatherby thought that out through years of pencil and paper notebooks, or thousands upon thousands of CPU hours, I think the essential mechanism has been the same.

Now a new wave of computational exploration is reaching a new level of maturity, along with that is coming the complex new learning strategy for us to understand. This is no longer purely a parallelization of basic maths — we are (for most) in a ‘Brave new world’ of deep mathematical associations and statistical bookkeeping.