I wonder if Keith's posting indicates that at least some arguments have been
(perhaps not surprisingly) at cross purposes.
Keith Sawyer wrote:
> 2. RE the question of common themes. Our discussion has not been very
> closely related to Bruce's original article. I agree with the first two of
> Scott's proposed themes, but the discussion has been more directly relevant
> to MABS practice than Scott's phrasing implies. Here is how I would
> restate Scott's first two themes (which I believe are the major themes):
>
> (1) what is the proper relationship between sociology and MABS? Alan
> started the debate with an offhand philosophical question about whether
> groups were "real" or not. I elaborated this idea, and made my initial
> claim that MABS cannot be successful without a close relationship with
> sociology, partly in order to address such questions and their relevance
> for MABS. This was followed by Scott's response that MABS doesn't
> necessarily need sociology. The ensuing debate about the role of
> sociology has been the single most central theme of this thread. (Note
> that our subject line has been "The relevance of sociology" for some time
> now.)
Just to be clear: my concern is that representational social simulation should be
validated with respect to real, target social systems and their components. When
there are well validated social or other theories that support mechanism and/or
agent designs implemented as a social simulation model, then the confidence we can
have in that model will be strengthened by verifying the model or its components
with respect to such a theory. I habitually employ agent designs taken from
cognitive and social psychological relations experimentally and observationally
validated by cognitive scientists and social psychologists. So the relevance of
sociology for me (and, if I read them correctly, Bruce and Rosaria) depends on the
extent to which there are well validated relationships that are captured by
sociological theory. This is, of course, a general point and you can substitute
the name of any other discipline for sociology in the previous sentence.
> (2) is it ever appropriate for a MABS to explicitly model any entities
> other than individual agents? If so, when is it, and how do we determine
> when; e.g., by reference to sociological data or theory, or through our own
> engineering concerns? Phrased this way, this theme is in part a subtheme
> of (1).
This indicates to me that Keith (by no means alone) has been concerned primarily
with sociology while the practicing MAS modellers in this discussion have been
concerned primarily with agent and mechanism design. That is probably why I think
that question 4 (proposed by Alexis Drogoul) is the most fundamental. For
convenience, Alexis' question was
4. What is the role assigned by each of us to simulation and how does it
influence our methodological or theoretical choices ? (by "role", I
essentially mean "goal", in fact : prediction, explanation, exploratory
work, pedagogical work, model shaping, etc. )
My question 3 is a special case of that
3. Should theory drive application or should application drive theory?
Questions 1 and 2 then follow as (increasingly) special cases of 4 and 3.
1. Issues of application -- whether simulation as description should be
bottom-up or top-down and the conditions in which each is valid.
2. Issues of sociological theory -- nominalism and methodological
individualism versus social realism and Durkheimism
I have argued elsewhere that economic theory is in principle not possible to
validate and therefore is irrelevant to social simulation. Keith and others have
not only argued that sociology is relevant to social simulation but the ways in
which it is relevant. Presumably, this means that social simulation models should
be verified with respect to specific sociological theories and philosophical
principles. I argue that, for social simulation, validation is more important
than verification. When the verification of representations with respect to any
formalism systematically results in the validation of the models, then the
formalism is useful in the class of conditions in which such validations are
realised.
I conclude that those sociological theories that are well validated are natural
candidates for verification by social simulation modellers. If unvalidated
sociological (or any other) theories imply mechanism and/or agent designs that can
be implemented in agent based social simulation models, then the validation of
those models stengthens confidence in the sociological theories with respect to
which they are sound and consistent (i.e. verified). So one role that might be
assigned to simulation is the validation (or demonstration of the invalidity) of
sociological theories.
Do these alternative specifications of the questions help to identify the
differences between us?
--
Professor Scott Moss
Director
Centre for Policy Modelling
Manchester Metropolitan University
Aytoun Building
Manchester M1 3GH
UNITED KINGDOM
telephone: +44 (0)161 247 3886
fax: +44 (0)161 247 6802
http://www.cpm.mmu.ac.uk/~scott
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|