[FRIAM] sometimes an onion is just an onion...

[glen ☣]

On 06/14/2017 03:46 PM, Marcus Daniels wrote:
> Others behaviors come to mind, like the agent that requires or expects fully contextualized unambiguous linear arguments.   This could be due to long term memory limitations, due to a desire to teach (supervisory learning), or an agent that defaults to priors in absence of systematic communication. Premature registration could be a sign of a simple hardware failure or inequitable / entitled expectations of the correspondent.   In organizations, some roles might well be encouraged even if they aren't adaptive in general, e.g. the boss vs. the nuanced facilitator.


You're suggestion of memory limitations threw me at first. But if we put it in terms of something like a "field of view" or a set of pattern bins, then we could have agents who select behaviors differently based on how many of their bins have something in it, and maybe some higher order properties like uniqueness of the bin contents, whether bin contents fit together in some way, etc.  E.g. An agent has 2 bins and registers object "flower" but does not register object "leaf" or "vase".  Without "leaf" or "vase", it has to fault, assume "gift bouquet", or any other conclusion but "still in ground".  (Or more usefully, an agent registers a sigmoid in one variable and expects/assumes a saturation in another variable, else the sigmoid it did recognize has too little context.)

I'm not sure how that might work for a teacher agent.  Maybe an outcome like "I see you have a flower but no leaves or vase.  Hence you have killed the flower and must now put it in a vase."  ... I guess that's the [Pedant agent].


-- 
☣ glen