What Could Solum Mean

Next in the backlog of articles was Larry Solum's "Artificial Meaning" (Washington LR 89:69), which concerns how an AI system generates meaning.The main point of the article is that meaning does not require something in a person's head.  Meaning is communicative content.  Communicative content can be generated by AI, and there is no one whose head it could have come from.  I think the overall line of thought is about right.  Meaning is not some special thing or special way of thinking.  My cavil is that I am having trouble connecting it to Solum's Fixation Thesis, or to his Complex Traffic Problem. 

"This sense of meaning refers to communicative content--the content conveyed by the legal text given the context in which it was authored."  That is the key sentence.  The Complex Traffic Problem is, more or less, traffic in City gets too intense and complicated to be managed by human beings, so AI is created to manage the system and succeeds.  In the course of doing so, it creates new traffic signs and terms, etc.  It creates meaning, the system's conduct conveys meaning.  

Here are the two problems for me.  The AI meanings are not "in the context in which authored" or do not look like they are.  The authoring seems to be in the writing of the code which operates the system.  That code does not convey the communicative content it leads to.  The meaning comes about later.  Or perhaps it is the later events which amount to the authoring?  That seems to me wrong because the later events are not really authored by anything but code interacting with machines.  And it is not clear that circumstance involves communicative content because, from the perspective of the for them moment authoring there is only instructions.  So then move the context to the whole system in action?  Maybe that will do it, if you are okay with the notion that meaning does not have to depend on an intentional creature. The second is this: why is context temporally limited?  What does fixing a time have to do with the AI generating communicative content that is set in place once for all.   In other words, why doesn't the context include the purposes of the system's actions?