Alan
I found your comments most interesting.
May I offer a clarification in regard to your first paragraph?
As President Clinton might say, we have to define "useful".
"Useful" does not imply
"the immediate provision of precise predictions
in a specific situation".
The insights contained in generic representations or models at a high
level of generality are useful.
But such conceptualisations may not be capable of giving precise or
applicable results in specific cases.
So the issue is the way that a model can be used.
I do not see a theoretical model to be the opposite of a useful model.
To the contrary, I see a theoretical model as a helpful step towards a
useful model.
The attraction of multi-agent models is that they embody entities or
agents with a 1:1 representation of the real-world structure.
In that sense the models are valid.
I see that as an encouraging foundation.
Maybe the urban gravitational models that you mention do not do that.
Their equations may provide valid predictions
but the models may not be good representations of the real world.
So I interpret Christof to be concerned with the matter of
computability.
But even if he accepts that to be the case, that does not solve the
matter. Because data derived from subjective estimates or which uses
data that is not generally agreed can also be used in computable models.
Consequently I understand him to be concerned not only with models that
are computable, but which use hard or replicable data.
Ben
PS I should explain that my perspective comes from mathematics of long
ago and from my current involvement in a multi-agent project. Ben
In message <029f01c02930$e33f0120$de3b2880@alans>, Alan Penn
<[log in to unmask]> writes
>Ben's response to Christof raises an issue I have been trying to sort out in
>my mind for a while. What is the difference between 'modelling' and 'real
>science'? It all seems to turn around the issue of what a theory is, and I
>find it illuminating the Christof takes a THEORETICAL model to be the
>opposite of one that is useful (and that Ben seems to accept this). My view
>is that 'real science' is based entirely on theory - theories that explain
>the observed data or phenomena. A side effect of 'explanation' is that these
>theories turn out to be predictive and so 'useful' in that they can help
>answer 'what if' questions.
>
>I suspect that there are two kinds of models. Type 1. tends to borrow
>theories from other areas of science and then apply them out of context.
>Consider, for example, the development of 'gravity models' of urban systems;
>the 'theory' is borrowed from Newton where it explains and predicts very
>satisfactorily what we observe about the motion of planets or freefalling
>objects. However, when it is applied to human systems one is left trying to
>quantify parameters, and guess what, these have to be established on a case
>by case basis through a process of model 'calibration'. The variation in
>paramaters from case to case is vast and even varys qualitatively -
>parameters on occasion change sign.
>
>Type 2. are what I think of as 'engineering models'. These models are
>pragmatic attempts to make the best of a field that as yet has no
>substantive explanatory theories of its own by constructing a fairly
>mechanistic set of components with inputs and outputs that look sensible.
>Traffic engineers models are of this sort. They are useful when calibrated
>to represent a specific system to look at likely effects of minor changes,
>but have great trouble with making 'what if' predictions where changes are
>radical. Note that when engineers have explanatory theories they do not use
>models of this sort - they do the calculations. Engineering models are
>reserved for places where the theory is missing and we have still to make
>decisions.
>
>A lack of explanatory theory is the reason why modellers tend to be frowned
>on by 'real scientists' when real explanatory theories exist in some domain.
>Look for instance at the fate of the 'biological modelling' fraternity
>following the theoretical and empirical advances of developmental biology.
>This is not to say that modelling is not useful in the process of science,
>just that it is useful early on before real theory and empirical methods are
>developed for a particular domain.
>
>If I am right in this, then what of the 'simulating societies' field? Well,
>here I think that the key thing is that we are _not_ modelling in ether the
>'theory borrowing' or the 'engineering' sense, we are developing simulations
>whose role is to create phenomena for us to theorise about. For us, these
>simulations are the equivalent of the lab bench and the petri dish, given
>that it is hard to be experimental with whole societies. The multi-agent
>simulation acts as a simplified 'model' of a society in the same sense that
>the fruit fly or nematode act as 'animal models' for the developmental
>biologist interested in how genes are switched on and off in humans. The MAS
>is simple, has a few key characteristics that we can control, and gives rise
>to interesting phenomena about which we go on to develop and test
>explanatory theory. So far, admittedly, we havent seen much of the latter,
>but then again we havn't seen much of it in the social sciences in
>general....OK maybe that's a bit sweeping :-)
>
>Bruce, Sorry.... I havn't managed to get to your paper yet.
>
>Alan Penn
>
>>
>> Your point is a good one.
>> May I offer an approach?
>>
>>
>> 1. Admit the shortage (or absence of, or lack of agreement on) objective
>data.
>> But simultaneously, point to the current inadequacies in current
>forecasting
>> methods,
>> and set out to allow the user to replace his/her subjective estimates at
>the
>> macro level
>> by subjective estimates at the micro(agent) level.
>>
>>
>> 2. Define the project as one of providing a useful tool for the task.
>> That is, the designer may start, if he wishes, by creating a generic
>agent-based
>> model, but that is an intermediate stage of the project.
>> You use "Theoretical" as the way to describe this.
>> In my eyes it is not theoretical.
>> It is a true model that has significant limitations of use
>> because of the unavailability of objective data.
>> (That happens with other types of model, I think.)
>> The designer then has to shape that first model into a form useful to the
>user.
>>
>>
>> 3. Specify that it will be a user task
>> to define his/her own assumptions and forecasts.
>> The model will allow and assist this. It may provide default assumptions.
>> But the assumptions remain a user responsibility.
>>
>>
>> 4. In which case, project specification made by the client
>> cannot be for the production of the "hard new information", you
>mention.
>> The specification would be for a
>> tool that would compute the consequences of the user assumptions in a
>valid way.
>>
>>
>> What do you think?
>>
>>
>> Ben Aston
>>
>
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|