At 11:01 AM 7/10/2009, Nuno David wrote:
>>I can think of one specific example, primitive wars, I can write a fairly
>>complete description of what should be in the model. It does have the
>>agents observing and the results of their observations affecting behavior.
>>We even have observations to check the model behaves like the real world.
>>http://cniss.wustl.edu/workshoppapers/gatpres1.pdf
>
>But then again, you know a priori what agents are going to observe and
>conceptualize, the vocabulary of the agents is pre-defined and the concept
>underlying it was already defined and programmed.
Not being intimately familiar with the jargon of social modeling, I
am not exactly sure what all the ramifications of the word "agent"
are or why it is a problem to define the vocabulary with which agents
communicate.
However in an attempt to model stone age wars, I can see at least
four classes of interacting agents, ones corresponding to genes, to
individuals, to memes and to tribes. You have to keep genes in mind
because they are what builds the psychological mechanisms in humans,
shapes human behavior toward relatives and activates behavioral
switches. But for the time constant of a few generations genes can
be considered practically static.
The model I am thinking about is a 2 D world where the substrate
feeds so many people per square km. Humans, being what they are
organize into tribes. The population of the tribes grows from births
and shrinks from deaths. The tendency of a tribe to go to war
depends on human agent observation of the state of the ecology and is
coupled though some time constant with the spread of xenophobic memes
that spread better when the observation for the future is bleak.
Perhaps this is too close to system dynamics. It is tied directly to
physical reality and we have several historical or prehistorical
human data sets to calibrate the model's performance against. We
could even turn off considerations of xenophobic memes and see if the
model reverted to chimpanzee wars/genocides. Another test, what
parameters have to be adjusted to get Bonobo behavior (no wars)?
Such a model would be of considerable interest, perhaps even as a
commercial sim game.
Keith
>Agents can indeed observe and react but that is not sufficient to say that
>they observe and understand emergent effects. Of course, that depends upon
>the meaning one ascribes to "emergent effects" or "emergent macro
>structures" etc. If micro-macro and macro-micro effects is just a way of
>referring to and describing, the dynamics of the simulation (with whatever
>language one uses to do it), then I would say that most (if not all)
>simulations have "downward
>causation", "upward" etc, feedback mechanisms etc. I am not convinced that
>the
>intention of (2) and (3) below is just that one:
>
>>>>2. The simulation then automatically develops explicit data structures
>>>>that capture the emergent macro level structures. (These are not
>>>>internal
>>>>to agents, but are autonomous from them.)
>
>>>>3. These new data structures then have causal effects on the local agents
>>>>and their interactions.
>
>best regards
>Nuno David
|