Tying the threads on this AI thing

We consider a simple world, all information is in the form of a directed JSON graph.

Consider the simple engine that performs a binary, commutative convolution of two JSON graphs, with a rudimentary conditional traverse based on simple key values. A machine that just got you around, the top layer map.  Then we add functionality to this machine by inserting a different grammars using attachments. Let us add weighting and inventory across the links and nodes, an attachment.  

Now the adapted general machine solves a whole variety of AI matching tasks.  One can envision graph pruning functions, matching functions, neural net functions.  Each of the functionality created by presenting the problem as an traversable JSON graph, inserting the enhanced rules of grammar for the context.  The machine treats all the attachments with graph function calls.

The trading bots can easily get around with this capability, building their own JSON map as they match their portfolio to the local terrain.  If we examine traffic on the links it is likely almost gaussian arrivals as bots compete, like De Broglie waves the random arrivals represent our imprecision.