We've got a 60 hour gen II (love it!), and never run out of space, usually have 20-30 shows stacked up at higher quality.
ONly purpose I guess is if you wanted to archive stuff, but its not really a good long-term archive, unless they also figure out how to support hot-swapping and RAID for failsafe.
With radio stations having to pay an increasingly large fee for each song on the playlist, it's no wonder that they play a much smaller selection of songs than they used to (say, back in the 80s).
I think you've got it backwards. Labels pay stations to play their tunes.
NASA seems to alternate between press releases of "Water/Life on Mars", "Yet Another Module of a Usless Space Station Launched", "Some 'Kids' Program" and "30 Years Since We Last Did Something (Orbit/Moon etc)".
LOL, true.
The search for life in the universe is important but should it really be the program's primary goal?
Yea, I think so, what more exciting important and lofty goal could there be?
IMHO, we should be trying to commercialize space (for humans not just satellites). NASA should help corporations build space hotels, start charging a $million a flight and fund their science that way. The Mars fossils aren't going anywhere! With a good space infrastructure looking for life becomes much easier.
I disagree, I think new discoveries would get people much more interested than as the other response suggested, watching a bunch of self-indulgent tourists go into space. I think that would tend to cheapen and debase it and turn people off.
We should find more ways to commerically exploit space, but I think that ultimitly basic research is what should motivate NASA. Funding for this, like the arts and humanaties and basic science, should be because it enriches all of us in the long-term. I think the more we can get away from short-term thinking the better off we are as a society, exp. given global warming, etc..
I think you're referring to the Schelling model, not the GAS models that the orginal poster was referring to. In any case, AFAIK the Schelling model is quite robust to changes in grid-size, as are the Sugarscape models, btw.
"Really? Then what possible use are the results?"
Hang on, you're making a big leap here. To say that grid-size has an effect on results is not the same as saying the results are useless! To make an extreme example, of course a grid-size of 1 x 1 in any CA or ABM is going to effect the results. The key is to understand how robust ressults are to these changes and why. You have to say more than that "results are dependent on grid-size" in order to make a useful critique.
The final point is not to confuse these results for absolute truth about human society! We do need to examine them closely and decide wether they fit the situation or not.
"A fatal flaw of this simulation (as a model of real society, that is) is that it includes the "Cincinatus" characters - the incorruptible agents - but does not include the "Dillingers" - agents who are not deterred by punishment, of themselves or of others."
A cool aspect of the civil violence model is that it in fact does include people who are not deterred by anything. Their grievance level is so high that they will attack no matter what. In some versions of the model, the hardship level is turned up slowly so that like a frog in a pot, people are gradually placed in this category and are picked off one by one.
This is a great recipe for a police state, by the way. "First they came for the communists..."
"The complaint about their lack of understanding of computer science issues was a throw-away comment and was irrelevent to the substance of my main beef with the book."
Fair enough, its irrelevancy was what made it ad hominem. I was left wondering why you made it, and that's where the "small-minded" assumption came from, sorry.
"...results that are not reproducable, and conclusions drawn from insufficient data."
Such as...?
(I'm not really sure what you meant by the grid-size comment, since of course results are dependent on this.)
...CAs usually have simultaneous (parallel, synchronous) update rules, whereas agent-based models (asynchronous) typically don't.
To give a concrete example, in Conway's life each agent (Cell) determines each period wether it will be dead or alive in the next period and then those values are updated all at once. In contrast, in a model like sugarscape, an agent who "dies" would typically remove itself immediatly.
These kind of seemingly insignificant implementation details can have huge effects on model outcomes and dynamics. This also means, btw, that one person's "artifact" can be another's "feature" and vis. versa.
I'm making fun of his post by pointing out that he pretends to have a clue about the subject but doesn't seem to. To be an ad hominem attack, my critque of his post has to be irrelevant to his argument. But it is relevant, because it goes to credibility. It is a classic ad hominem attack, however, to claim that Epstein & Axtell cannot be credible because (horrors!) they do not have a CS degree.
Among other things, its an incredibly simple minded way to judge someone's capabilities and many CS grads can't code their way out of a paper bag.
You might be interested to know that for instance, Axtell (who got his doctorate at CMU, btw) knows far more about classic CS issues than almost anyone I know, CS grads included, and Epstein carried around (and studied) a copy of Godel's theorems everywhere he went one year. These are not people who do not have an appreciation for computer science. Nothing could be further from the truth.
To claim that a CS degree is neccessary and that E&A don't have the relevant background is a pure, uninformed ad hominem attack, and kind of small minded, I think. If the poster had instead offerred an actual critique of the science, I would have responded differently.
Apologies for being so pedantic but this kind of thing bugs the hell out of me...
I think this example does somewhat show the limitations of these traditional models though; for example the model might not uncover the fact that pollution per capita is very hetergenous, or that it could be effected by cultural factors (got to have that SUV!), or technogically changes as I think someone else pointed out, etc..
Its an interesting point that these big early systems dynamics models were almost taken too seriously in the late 60s and 70s and then discredited. Important to keep this in mind and limit our claims as we develop our brave new models.
btw, hey Paul, looks like old home week for all of us.:-)
..the distinction between explanation and prediction is especially important here.
That said, the possiblity of prediction on some level is certainly possible. For example, if you were to run the Artificial Anasazi model forward after calibrating it properly, the results could be just as predictive if not more so than a traditional model -- say a PDE model.
But the basic point stands out; precisely because complex adaptive systems are so path-dependent and involve so many loosly coupled pieces, actual prediction of future events ala Seldon is very questionable, unless you think about it in terms of this classic joke:
"I have a complete model of the world economic system. It's life-size and runs in real-time."
That is, you can begin to throw more and more things in to your model, but at some point you lose the explanatory power that the model gave you in the first place. I've seen this quite a bit; people throw the kitchen sink in and the model becomes a muddled mess. The real value of these models, especially at this point, is in their explanatory power; ie.e they help you to generate (Epstein's usage) phenomenon you observe in the real world with quite simple rules.
This in turn helps you to understand how relativly simple processes may be at play in complex phenomenon, and may even give you some ideas about how to work with them. In this way, these models could have a real policy impact in a much more engaged and robust way than the Club of Rome example offerred in another thread.
"I found some things in Epstein and Axtell that were not, or should not have been, replicable from the information given in their book."
More detail would help, I couldn't find any more information on your website. You may want to distinguish between a) not replicable and b) not replicable by you, potentially very different things.
I reimplemented all of the chapter two and three models in Java and while I did have the advantage of having the authors next door, I found that I was able to achieve the same results quite easily without asking many questions. This situation seems to me to be very akin to any kind of research endeavor, where a paper describes just the critical issues. The implementations details (think lab aparatus, etc..) really cannot be usefully and completely covered in a general treatment. But the results are very reproducable, and I must say, quite robust.
It is interesting that you say that you are using a discrete event simulation environment in your work. Generally, these models have much more of a time-step quality, though even that doesn't really capture things completetly. Discrete event approaches have artifacts and issues of their own of course; I suspect that a lot of your issues are occuring along this boundary, but that's just a guess.
Software for Exploring Artificial Socieities...
on
Simulating Societies
·
· Score: 1
I wanted to point out to/.s out there that there is quite a bit of software available to explore artificial societies on your own. I also wanted to say that I have had the rare privledge of working w/ these folks for many years and all of the positive comments (and none of the negative ones:-)) in this thread are right on.
Anyway, Ascape, a software framework for agent-based modelling that I developed is available for download at the Brookings Website. Many other interesting models are also described there..
Its all in Java, and the source code for versions of many models is available. (For the person who was complaining that the results aren't reproducable, this will prove you completely wrong. In fact, Epstein and Axtell and others in the field have spent a lot of time thinking about how models can be independently understood and verified.)
Someone has allready mentiond JASSS; there is an article in that journal on Ascape.
The Ascape build on the Brookings website is now quite old. I joined the Bios Group some time ago and we've been improving and enhancing Ascape as part of our work in using complexity science in "real world" applications. So there should be a new public release RSN, but the version on the website now is relativly robust and has a lot of features. Note that the mailing list at the Brookings website appears to be down at the moment.
Slashdot Mirror
Oh, and don't forget - this isn't a real junkyard. It's a set. Whaaa...? %-(
We've got a 60 hour gen II (love it!), and never run out of space, usually have 20-30 shows stacked up at higher quality. ONly purpose I guess is if you wanted to archive stuff, but its not really a good long-term archive, unless they also figure out how to support hot-swapping and RAID for failsafe.
I think you're referring to the Schelling model, not the GAS models that the orginal poster was referring to. In any case, AFAIK the Schelling model is quite robust to changes in grid-size, as are the Sugarscape models, btw. Hang on, you're making a big leap here. To say that grid-size has an effect on results is not the same as saying the results are useless! To make an extreme example, of course a grid-size of 1 x 1 in any CA or ABM is going to effect the results. The key is to understand how robust ressults are to these changes and why. You have to say more than that "results are dependent on grid-size" in order to make a useful critique.
The final point is not to confuse these results for absolute truth about human society! We do need to examine them closely and decide wether they fit the situation or not.
OK, "we".
Fair enough, its irrelevancy was what made it ad hominem. I was left wondering why you made it, and that's where the "small-minded" assumption came from, sorry. Such as...?
(I'm not really sure what you meant by the grid-size comment, since of course results are dependent on this.)
...CAs usually have simultaneous (parallel, synchronous) update rules, whereas agent-based models (asynchronous) typically don't.
To give a concrete example, in Conway's life each agent (Cell) determines each period wether it will be dead or alive in the next period and then those values are updated all at once. In contrast, in a model like sugarscape, an agent who "dies" would typically remove itself immediatly.
These kind of seemingly insignificant implementation details can have huge effects on model outcomes and dynamics. This also means, btw, that one person's "artifact" can be another's "feature" and vis. versa.
I'm making fun of his post by pointing out that he pretends to have a clue about the subject but doesn't seem to. To be an ad hominem attack, my critque of his post has to be irrelevant to his argument. But it is relevant, because it goes to credibility. It is a classic ad hominem attack, however, to claim that Epstein & Axtell cannot be credible because (horrors!) they do not have a CS degree.
Among other things, its an incredibly simple minded way to judge someone's capabilities and many CS grads can't code their way out of a paper bag.
You might be interested to know that for instance, Axtell (who got his doctorate at CMU, btw) knows far more about classic CS issues than almost anyone I know, CS grads included, and Epstein carried around (and studied) a copy of Godel's theorems everywhere he went one year. These are not people who do not have an appreciation for computer science. Nothing could be further from the truth.
To claim that a CS degree is neccessary and that E&A don't have the relevant background is a pure, uninformed ad hominem attack, and kind of small minded, I think. If the poster had instead offerred an actual critique of the science, I would have responded differently.
Apologies for being so pedantic but this kind of thing bugs the hell out of me...
I think this example does somewhat show the limitations of these traditional models though; for example the model might not uncover the fact that pollution per capita is very hetergenous, or that it could be effected by cultural factors (got to have that SUV!), or technogically changes as I think someone else pointed out, etc..
Its an interesting point that these big early systems dynamics models were almost taken too seriously in the late 60s and 70s and then discredited. Important to keep this in mind and limit our claims as we develop our brave new models.
btw, hey Paul, looks like old home week for all of us.
..the distinction between explanation and prediction is especially important here.
That said, the possiblity of prediction on some level is certainly possible. For example, if you were to run the Artificial Anasazi model forward after calibrating it properly, the results could be just as predictive if not more so than a traditional model -- say a PDE model.
But the basic point stands out; precisely because complex adaptive systems are so path-dependent and involve so many loosly coupled pieces, actual prediction of future events ala Seldon is very questionable, unless you think about it in terms of this classic joke:
"I have a complete model of the world economic system. It's life-size and runs in real-time."
That is, you can begin to throw more and more things in to your model, but at some point you lose the explanatory power that the model gave you in the first place. I've seen this quite a bit; people throw the kitchen sink in and the model becomes a muddled mess. The real value of these models, especially at this point, is in their explanatory power; ie.e they help you to generate (Epstein's usage) phenomenon you observe in the real world with quite simple rules.
This in turn helps you to understand how relativly simple processes may be at play in complex phenomenon, and may even give you some ideas about how to work with them. In this way, these models could have a real policy impact in a much more engaged and robust way than the Club of Rome example offerred in another thread.
Its "the esteemed Brookings Institution" not "Institute."
;-)
Otherwise, your post is a little too trollish and ad hominem to bother responding to, sorry.
"I found some things in Epstein and Axtell that were not, or should not have been, replicable from the information given in their book."
More detail would help, I couldn't find any more information on your website. You may want to distinguish between a) not replicable and b) not replicable by you, potentially very different things. I reimplemented all of the chapter two and three models in Java and while I did have the advantage of having the authors next door, I found that I was able to achieve the same results quite easily without asking many questions. This situation seems to me to be very akin to any kind of research endeavor, where a paper describes just the critical issues. The implementations details (think lab aparatus, etc..) really cannot be usefully and completely covered in a general treatment. But the results are very reproducable, and I must say, quite robust.
It is interesting that you say that you are using a discrete event simulation environment in your work. Generally, these models have much more of a time-step quality, though even that doesn't really capture things completetly. Discrete event approaches have artifacts and issues of their own of course; I suspect that a lot of your issues are occuring along this boundary, but that's just a guess.
I wanted to point out to /.s out there that there is quite a bit of software available to explore artificial societies on your own. I also wanted to say that I have had the rare privledge of working w/ these folks for many years and all of the positive comments (and none of the negative ones :-)) in this thread are right on.
Anyway, Ascape, a software framework for agent-based modelling that I developed is available for download at the Brookings Website. Many other interesting models are also described there..
Its all in Java, and the source code for versions of many models is available. (For the person who was complaining that the results aren't reproducable, this will prove you completely wrong. In fact, Epstein and Axtell and others in the field have spent a lot of time thinking about how models can be independently understood and verified.)
Someone has allready mentiond JASSS; there is an article in that journal on Ascape.
The Ascape build on the Brookings website is now quite old. I joined the Bios Group some time ago and we've been improving and enhancing Ascape as part of our work in using complexity science in "real world" applications. So there should be a new public release RSN, but the version on the website now is relativly robust and has a lot of features. Note that the mailing list at the Brookings website appears to be down at the moment.