Article on Epstein

classic Classic list List threaded Threaded
6 messages Options
Reply | Threaded
Open this post in threaded view
|

Article on Epstein

Robert Holmes
Epstein has a new book and MIT Tech Review are running an article on
artificial societies on the back of it

http://www.technologyreview.com/Infotech/18880/page1/

And again, there's that old chestnut: these models explain, not predict. Do
we still believe this? I agree - they do not predict, but do they even
explain? I'm getting increasingly troubled about this whole notion that the
rules the researcher puts in the agents actually have some sort of analog in
actual people. Even when conclusions are presented as "this is AN
explanation" not "this is THE explanation", I suspect that the ABM
researcher is being somewhat optimistic.

So what is the relationship between the rules in the artificial agents and
the rules in real people?

Robert
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://redfish.com/pipermail/friam_redfish.com/attachments/20070626/ce39aa92/attachment.html 

Reply | Threaded
Open this post in threaded view
|

Article on Epstein

Pamela McCorduck
What kind of explanation of social behavior would satisfy you?


On Jun 26, 2007, at 8:31 AM, Robert Holmes wrote:

> Epstein has a new book and MIT Tech Review are running an article  
> on artificial societies on the back of it
>
> http://www.technologyreview.com/Infotech/18880/page1/
>
> And again, there's that old chestnut: these models explain, not  
> predict. Do we still believe this? I agree - they do not predict,  
> but do they even explain? I'm getting increasingly troubled about  
> this whole notion that the rules the researcher puts in the agents  
> actually have some sort of analog in actual people. Even when  
> conclusions are presented as "this is AN explanation" not "this is  
> THE explanation", I suspect that the ABM researcher is being  
> somewhat optimistic.
>
> So what is the relationship between the rules in the artificial  
> agents and the rules in real people?
>
> Robert
>
> ============================================================
> 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

?Good judgment comes from experience, and experience ? well, that  
comes from poor judgment.?       A.A. Milne



-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://redfish.com/pipermail/friam_redfish.com/attachments/20070626/4d94f1af/attachment.html 

Reply | Threaded
Open this post in threaded view
|

Article on Epstein

Carl Tollander
 From the article:
"Artificial society modeling allows us to 'grow' social structures /in
silico/ demonstrating that certain sets of microspecifications are
/sufficient to generate/ the macro?phenomena of interest."

The issue hinges on what "sufficient to generate" means for a  
particular model in terms of  that model's explanatory power.    I have
come to suspect that it does not mean as much as is sometimes thought.  
There are questions about whether a model's description of the domain is
unique and/or salient, whether the local dynamics are stationary, how we
characterize the influence of the experimental design, what does it mean
to validate the model, and so on.  Assumptions about the answers to
these questions can be as influential (and hidden) as assumptions about
rational economic actors.

To me, growing the model is a fine methodology (since it explicitly
recognizes that we create models relative to a specific epistemological
context), but we recognize that for a big class of these models that one
can generate a lot of models from the same specification (a kind of
pleiotropy), and a lot of different specifications may generate very
similar models under certain conditions.

A given ABM is less for explanation or prediction than for exploration
and understanding; it helps (or not) clarify the issues and concepts
under consideration relative to some space of such ABMs.  Whether we can
build a particular model that generates some expected social behavior
does not necessarily mean that the particular model constitutes a
complete explanation.  As we are coming to understand in developmental
biology, whether a gene microspecification is associated with some some
macrophenomena trait has minimal explanatory power.  It's important to
understand how the RNA works.   In the same way, it would not be at all
surprising to find that "rules" were not sufficient microspecifications
for spaces of models of social behavior.  Modeling complexity is itself
a complex activity.

Carl

Pamela McCorduck wrote:

> What kind of explanation of social behavior would satisfy you?
>
>
> On Jun 26, 2007, at 8:31 AM, Robert Holmes wrote:
>
>> Epstein has a new book and MIT Tech Review are running an article on
>> artificial societies on the back of it
>>
>> http://www.technologyreview.com/Infotech/18880/page1/ 
>> <http://www.technologyreview.com/Infotech/18880/page1/>
>>
>> And again, there's that old chestnut: these models explain, not
>> predict. Do we still believe this? I agree - they do not predict, but
>> do they even explain? I'm getting increasingly troubled about this
>> whole notion that the rules the researcher puts in the agents
>> actually have some sort of analog in actual people. Even when
>> conclusions are presented as "this is AN explanation" not "this is
>> THE explanation", I suspect that the ABM researcher is being somewhat
>> optimistic.
>>
>> So what is the relationship between the rules in the artificial
>> agents and the rules in real people?
>>
>> Robert
>>
>> ============================================================
>> 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
>
> ?Good judgment comes from experience, and experience ? well, that
> comes from poor judgment.?       A.A. Milne
>
>
>
> ------------------------------------------------------------------------
>
> ============================================================
> 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


Reply | Threaded
Open this post in threaded view
|

Article on Epstein

Robert Holmes
In reply to this post by Pamela McCorduck
Good question - an explanation that's grounded in actual field research I
guess.

IMHO, an ABM can never offer an explanation for a social behaviour. All it
can ever do (and I'm not being dismissive, I think this is important) is
offer a suggestion for an explanation that can subsequently be confirmed or
denied by real social research/anthropology/enthnological field research
program.

I don't think this is a particularly strong claim. The logic behind the a
sugarscape or Netlogo style ABM seems to be (i) apply some micro rules to
checkers running round a checker board, (ii) generate an unexpected macro
behaviour, (iii) offer the micro rules as an explanation of the macro rules
then (iv) claim that this checker-board behaviour is analagous to behaviour
of real people/animals/companies/other real world entities.

Step (i) through (iii) are OK (though most ABM papers I see aren't as
upfront about the many-to-one nature of the explanation as Carl is in his
email) but (iv) strikes me as a bit of a stretch; certainly I'd like more
than vague assurances from the researcher that yes it's valid, honest. It
doesn't strike me as unreasonable to ask for some evidence that the leap in
(iv) is reasonable. But how often do we see that in the literature? As I
suggest above, there's plenty of social research techniques that could
generate that evidence. But I get the impression that the detailed
comparison of model with reality that you get in (say) the Ancestral Pueblo
study is the exception rather than the rule.

And this is why we need more Mike Agars in this world.

Robert

On 6/26/07, Pamela McCorduck <pamela at well.com> wrote:

>
> What kind of explanation of social behavior would satisfy you?
>
> On Jun 26, 2007, at 8:31 AM, Robert Holmes wrote:
>
> Epstein has a new book and MIT Tech Review are running an article on
> artificial societies on the back of it
>
> http://www.technologyreview.com/Infotech/18880/page1/
>
> And again, there's that old chestnut: these models explain, not predict.
> Do we still believe this? I agree - they do not predict, but do they even
> explain? I'm getting increasingly troubled about this whole notion that the
> rules the researcher puts in the agents actually have some sort of analog in
> actual people. Even when conclusions are presented as "this is AN
> explanation" not "this is THE explanation", I suspect that the ABM
> researcher is being somewhat optimistic.
>
> So what is the relationship between the rules in the artificial agents and
> the rules in real people?
>
> Robert
>
> ============================================================
> 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
>
>
> "Good judgment comes from experience, and experience ? well, that comes
> from poor judgment."       A.A. Milne
>
>
>
> ============================================================
> 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
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://redfish.com/pipermail/friam_redfish.com/attachments/20070626/2e1818fd/attachment.html 

Reply | Threaded
Open this post in threaded view
|

Article on Epstein

Owen Densmore
Administrator
In reply to this post by Robert Holmes
On Jun 26, 2007, at 8:31 AM, Robert Holmes wrote:

> Epstein has a new book and MIT Tech Review are running an article on
> artificial societies on the back of it
>
> http://www.technologyreview.com/Infotech/18880/page1/

Well, I bought the book, because of our teaching the SFI modeling  
section, and wanted to have it as a reference.  It seems reasonable  
enough .. covering an interesting spectrum of what we all are  
prodding at and not a "popular" book as much as a survey of  
techniques and history for practitioners.  I like the structure of  
the book as well -- basic classics preceded by a unifying preface.  
Prefer NOT to have the CD, a web site would be better.  I suspect  
it'll be a classic.

> And again, there's that old chestnut: these models explain, not  
> predict. Do
> we still believe this? I agree - they do not predict, but do they even
> explain? I'm getting increasingly troubled about this whole notion  
> that the
> rules the researcher puts in the agents actually have some sort of  
> analog in
> actual people. Even when conclusions are presented as "this is AN
> explanation" not "this is THE explanation", I suspect that the ABM
> researcher is being somewhat optimistic.

You're being a bit unfair: the entire paragraph is:
.....
     Altogether, in fact, Epstein stressed that his models were  
mostly aimed at achieving explanatory power. "To explain something  
doesn't mean that you can predict it," he said. He pointed out that  
though we can explain lightning and earthquakes, we can't forecast  
either. If we're hoping, like Asimov, to predict the future,  
Epstein's models will disappoint. In fact, because his models give  
widely divergent results even when their agents are programmed with  
very simple rules, they indicate that predicting the future will  
never be possible. Still, Epstein's artificial societies do more to  
make plain the hidden mechanisms underlying social shifts--and their  
unexpected consequences--than any tool that social scientists have  
hitherto possessed. In the future, they and others like them could  
suggest how policymakers can engineer the sorts of small, cheap  
interventions that have large, beneficial results.
.....
Physics, in other words, has many of the same difficulties when  
considering ensembles.  Its quite reasonable for us as well.  Yet we  
don't dis physics when it can't predict details, only explain  
behavior, in say statistical mechanics.

> So what is the relationship between the rules in the artificial  
> agents and
> the rules in real people?
>
> Robert

At the summer school this year, Tom Carter made a great point:  
Consider the narrative of your model .. the story it tells and the  
story you will tell as you explain it to others as the model evolves  
before their eyes.  So in the Schelling model of segregation, you  
can't predict which neighborhood will become a ghetto, but you can  
predict with near-statistical-mechanics-certainty that segregation  
will occur.  And all with a simple preference behavior.

     -- Owen


Reply | Threaded
Open this post in threaded view
|

Article on Epstein

Robert Holmes
On 6/26/07, Owen Densmore <owen at backspaces.net> wrote:
>
> <snip>

So in the Schelling model of segregation, you
> can't predict which neighborhood will become a ghetto, but you can
> predict with near-statistical-mechanics-certainty that segregation
> will occur.  And all with a simple preference behavior.
>
>      -- Owen


In other words, this is an explanation that works in some cases but not in
others? And there's no a priori way of discriminating between the occasions
the explanation will work and the times it won't. So how can we consider it
an explanation?

Also, when I did a quick check into the research on segregation and
residential preferences (I found papers from W.A.V. Clarke at UCLA,  Farley
at al in Housing Policy Debate, Fossett at Texas A & M), none of them gave
single factor explanations for segregation. There's all these other things
like cost of housing, availability of employment, overt discrimination in
housing and lending markets, all sorts of demographic stuff.

I think the Schelling example illustrates my point: ABMs can suggest an
explanation but it requires real social science research to provide an
explanation.

Robert
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://redfish.com/pipermail/friam_redfish.com/attachments/20070626/442ddb56/attachment.html