Re: Combining antropology & complexity science, anyone?

[glen e. p. ropella]

At 12:18 AM 11/28/2001 +0200, you wrote:
> constituents, and have beer shipped around. But how do you 'model' e.g. 
> the wholesaler's 'anxiety' of running out of stock and thus a.o. cause the 
> bullwhip effect? I'm at a complete loss on how to do that, save for a 
> random number generation on an 'anxiety' gene in an agent but that's just 
> too simple to my taste.

"We" (a.k.a. the Swarm stylists [grin]...if I may be so bold as
to speak for the rest of you) have a modeling pattern for this
type of thing.  I wouldn't claim that it is effective or accurate
because I don't know of anyone who's done any science
this way.  But, it *is* testable if an experiment is designed
correctly for it...

Anyway, the pattern is "running simulations in the head".

Basically, it consists of a model where the constituents of
the model *model* their environment and other constituents
of the model.

So, applying this pattern to your question, one would allow the
inventory manager to *simulate*, rationally, what the likely
situations are to obtain with respect to supply and demand.  That
simulation could consist of a simple spreadsheet or a full-blown
internal ABM, or any of the other usual decision support techniques.

"Anxiety" would then be installed as an uncertainty or error in
the manager's model of his system.  Further, that manager could
be parametrized with a "risk tolerance" or "tolerance of ambiguity"
factor that could regulate his hoarding behavior.

> Other example; I can have a supply chain modelled to what we know of ant 
> behaviour, but how do I capture the fact that warehouse manager A hates 
> the guts of warehouse manager B and thus goes out to have his coffee first 
> rather than responding to the urgent urgent fax from Mr. B, thus causing 
> the whole supply chain to become unbalanced?

This would be modeled with a latency on Mr. A's inputs-to-output
process.  That latency could be controlled by any number of
physiological, psychological, or pure behavioral factors.

As an example, I built a model where a category of agents were
parametrized to prefer using the internet to buy things versus
to prefer interacting with other humans when they buy things.
The collective behavior of that category was "emergent" in the
sense that it arose from the dynamics of the space in which they live
and the inscribed parametrization of the individuals.

> All this ran me into considering 'les particularites humaine' as a 
> modelling layer -on top of- a simple thing (sic) like a supply chain. Can 
> I construct a bunch of agents that 'just interact' and display human 
> behaviour in all it's idiocraties? Thus, I (think I) need to somehow 
> bridge a gap between what we know on human behaviour through antropology 
> (yes, human behaviour -is- predictable to some extent) and complexity 
> science/ABM etcetera. So I need to know whether something similar has 
> alrady been considered, tried, written up, modelled, referenced, ...

The question, as I think others have pointed out, is what you want
to inscribe and what you want as result.  I doubt that one can
(at this stage of the game) achieve "human behavior in all its
idiosyncrasies" solely through a bottom-up model.  (If this is
truly what you're after, then you should do your literature searches
starting with robotics, I'd think.)  However, you can design explicit
experiments to achieve certain particular aspects of human behavior
(either in the collective or in the individuals) by designing the
individuals or the collective from the bottom-up.

If you're looking for emergent behavior in the collective, then
I'd use one or more of the standard psychological models for the
individuals and allow for self-configuration towards some
anthropologically sound collective behavior.

If, however, you're looking for emergent individual behavior
(like a generative model of "anxiety"), then you should construct
the individual out of a more detailed model (like neural
darwinism or connectionism) and allow for self-configuration
towards the resultant behavior (expression of anxiety or
belligerent avoidance).

In addition to searching anthro- and psycho-logical literature,
I'd suggest looking through the combat modeling that the
military does.  They have made great progress modeling (or
at least executing clinical psych experiments therein)
"human behavior under stress".

Also, I'd suggest you pick one or two particular observational
science experiments to base your modeling on.  Don't go too
far into interpretation or speculation on what those particular
experiments are supposed to be telling us about humans.  Just
stick to the barest facts and see if you can get a model to
reflect those facts.

glen

=><==><==><==><==><==><==><==><==><==><==><==><==><==><==><==><=
The competent hawk hides its claws.       =><=        Hail Eris!
glen e. p. ropella   =><=   H:831.335.4950  =><=  C:831.247.7901



                 ==================================
  Swarm-Modelling is for discussion of Simulation and Modelling techniques
  esp. using Swarm.  For list administration needs (esp. [un]subscribing),
  please send a message to <address@hidden> with "help" in the
  body of the message.
                 ==================================