Poll: which agent-based model is most fundamental?

[Jochen Fromm-4]

In Physics there are many equations, but only a few
are really fundamental (for example the Maxwell
equations, or Newton's laws, etc.).

There are a number of basic agent-based models, too,
for example Arthur's El Farol Bar Model, Craig Reynolds'
Boids, Schelling's Segregation Model, and
Axelrod's Tribute or Dissemination Models.

What agent-based model do you know and which are
the most fundamental? Do we have a basic model for
every basic agent interaction pattern, see
http://www.cas-group.net/wiki/Agent_interaction_pattern   ?

-J.

============================================================
FRIAM Applied Complexity Group listserv
Meets Fridays 9a-11:30 at cafe at St. John's College
lectures, archives, unsubscribe, maps at http://www.friam.org

[Marcus G. Daniels]

Jochen Fromm wrote:
> In Physics there are many equations, but only a few
> are really fundamental (for example the Maxwell
> equations, or Newton's laws, etc.).
> What agent-based model do you know and which are
> the most fundamental?
Hmm, in systems biology, one technique that is used heavily is the
http://en.wikipedia.org/wiki/Gillespie_algorithm

It's a stochastic ABM-like technique where interesting reactions are
rare, and uninteresting ones common, and mechanically amounts to a
dynamic discrete event simulation such is often found in ABM models.  
Epidemic models like Kermack-McKendrick SIRS fit into this sort of
approach as well.

============================================================
FRIAM Applied Complexity Group listserv
Meets Fridays 9a-11:30 at cafe at St. John's College
lectures, archives, unsubscribe, maps at http://www.friam.org

[Owen Densmore]

One approach might be to break it down a bit finer, into the  
techniques used.  Two spring to mind:
1 - Optimization Techniques (GAs, Ant Algorithms, ..)
2 - Game Theory types (ultimatum game, prisoner's dilemma, ...)

It's be nifty to see the "spanning set" underlying the popular models.

     -- Owen

On Dec 31, 2008, at 9:35 AM, Jochen Fromm wrote:

> In Physics there are many equations, but only a few
> are really fundamental (for example the Maxwell
> equations, or Newton's laws, etc.).
>
> There are a number of basic agent-based models, too,
> for example Arthur's El Farol Bar Model, Craig Reynolds' Boids,  
> Schelling's Segregation Model, and Axelrod's Tribute or  
> Dissemination Models.
>
> What agent-based model do you know and which are
> the most fundamental? Do we have a basic model for every basic agent  
> interaction pattern, see http://www.cas-group.net/wiki/Agent_interaction_pattern 
>    ?
>
> -J.

============================================================
FRIAM Applied Complexity Group listserv
Meets Fridays 9a-11:30 at cafe at St. John's College
lectures, archives, unsubscribe, maps at http://www.friam.org

[glen e. p. ropella-2]

Thus spake Jochen Fromm circa 31/12/08 08:35 AM:
> What agent-based model do you know and which are
> the most fundamental? Do we have a basic model for every basic agent
> interaction pattern, see
> http://www.cas-group.net/wiki/Agent_interaction_pattern   ?

I think the question is ill-formed.  Agent-based _models_ are just
models. The phrase agent-based model is context free, unlike physics or
biology.  And without context, there isn't any one model that's more
fundamental than any other model.

A better question would be "what agent-based model is most fundamental
in physics" or "... in biology" or "... in politics".

Regarding "patterns", a pattern is just a particular inference made by
an observer.  Granted, there may be some dominant patterns we settle on
by consensus as prominent or important; but, such consensus will always
assume some context.  And the prominence of that (class of) pattern(s)
will go away if that assumption changes.

--
glen e. p. ropella, 971-222-9095, http://agent-based-modeling.com


============================================================
FRIAM Applied Complexity Group listserv
Meets Fridays 9a-11:30 at cafe at St. John's College
lectures, archives, unsubscribe, maps at http://www.friam.org

[Jochen Fromm-4]

Thus spake Glen e. p. Ropella:
> I think the question is ill-formed.  Agent-based _models_ are just
> models. The phrase agent-based model is context free, unlike physics or
> biology.  And without context, there isn't any one model that's more
> fundamental than any other model.
>

Good point. In principle, the same can be said for equations,
there is no equation that's more fundamental than any other
equation. But certain equations define whole areas and subfields
of physics, for example
* Newton's laws define (classical) mechanics,
   including forces, masses, accelerations, ..
* Maxwell's equations define electrodynamics,
   including fields, charges, currents,..

I wonder if we can find fundamental agent-based models which
define a certain area or subfield and explain abstract terms of
fundamental importance (like 'power' or 'culture')

-J.


============================================================
FRIAM Applied Complexity Group listserv
Meets Fridays 9a-11:30 at cafe at St. John's College
lectures, archives, unsubscribe, maps at http://www.friam.org

[glen e. p. ropella-2]

Thus spake Jochen Fromm circa 01/01/09 07:05 AM:
> I wonder if we can find fundamental agent-based models which define a
> certain area or subfield and explain abstract terms of fundamental
> importance (like 'power' or 'culture')

I think such an attempt will necessarily devolve into a discussion of
how "agent-based modeling" is different from (if at all) other methods
for exploring the world (e.g. continuum math).  FRIAM's beaten around
that bush a couple of times since I've been subscribed and it hasn't
really gone anywhere.  In my opinion, combinatoric methods like ABM are
no different from any other form of math.  But others on this list
insist that ABMs are not mathematics.  Still others see a large
distinction between math (method) and science (knowledge), where I see
them as so deeply intertwined that the distinction is false (but useful).

But we can be good computationalists/deists and set up the
discussion/simulation, watch it play out, and draw our conclusions
post-execution. [grin]  Marcus and Owen have already gotten us started with:

1) transformation with variation (e.g. gillespie algorithm)
2) estimation (i.e. optimization with variation), and
3) teleological ("anticipatory") deduction (e.g. game theory)

My guess is that variation and teleology are absolutely necessary for
describing complex systems, especially like we find in biology.  Biology
is the _middle_ of science.  As we go "down" (finer grain) to physics,
variation and teleology begin to disappear (until we get to entanglement
anyway).  I also suspect that as we go "up", variation and teleology
disappear. (Personally, I believe the cause of the disappearance is the
abstraction required and available when you go "down" or "up".
Abstraction mitigates against variation and teleology.)

The domain of biology is huge and I think social systems, wherein we use
words like "power" and "culture", are well within it and, hence, my
guess would be that variation and teleology are necessary (though
perhaps not _true_).

So, the task is to further slice up the classification so that each
method can be evaluated in the context of a domain as, say, "very useful
in that domain", "useful", "not very useful".  (If you don't like the
categories (1-3) above, then come up with some others.  You can also
replace "useful" with "common".)  I suspect when/if we got to a
classification granularity of 5-9 (possibly falsely) distinct methods,
we can begin to assert where each method is _most_ useful.

And where a method is most useful and other methods are least useful,
then we can say that that most useful method is "fundamental" to that
domain.

--
glen e. p. ropella, 971-222-9095, http://agent-based-modeling.com


============================================================
FRIAM Applied Complexity Group listserv
Meets Fridays 9a-11:30 at cafe at St. John's College
lectures, archives, unsubscribe, maps at http://www.friam.org

[Marcus G. Daniels]

glen e. p. ropella wrote:
> So, the task is to further slice up the classification so that each
> method can be evaluated in the context of a domain as, say, "very useful
> in that domain", "useful", "not very useful".  (If you don't like the
> categories (1-3) above, then come up with some others.  You can also
> replace "useful" with "common".)  I suspect when/if we got to a
> classification granularity of 5-9 (possibly falsely) distinct methods,
> we can begin to assert where each method is _most_ useful.
>  
Then make a pretty area-proportional Euler diagram out of the result..  :-)

http://www.informatik.uni-ulm.de/ni/staff/HKestler/vennm/doc.html

============================================================
FRIAM Applied Complexity Group listserv
Meets Fridays 9a-11:30 at cafe at St. John's College
lectures, archives, unsubscribe, maps at http://www.friam.org