"Validation of the entire model is not feasible in the near future. The phenomena modeled are too poorly understood and too strongly interconnected. Even with the modest goal of merely “anticipating likely consequences”, the best that can be hoped for is establishment of credibility." -- from pg 16 of https://ntrs.nasa.gov/archive/nasa/casi.ntrs.nasa.gov/201400...
The single open issue (a tortuously-phrased "write some docs") should make clear the amount of effort the authors have made & willingness to make this approachable to newcomers (zero).
The code is pretty readable. And there's a freaking 250-page User’s Guide linked from the documentation. I really think they deserve a little more than some snarky dismissal. There aren't very many open source projects with that level of documentation.
It's not Sim City, but then again maybe they didn't set out to create something that provides insights to people who first came across their field of study five minutes earlier.
If you don't do user-friendly documentation, your project is that much more likely to die for lack of interest.
then again maybe they didn't set out to create something that provides insights to people who first came across their field of study five minutes earlier
The whole point of simulators is to learn things by experiment. If a tool isn't accessible and doesn't come with examples, it's unlikely to see much further use or development.
Are you referring to the 250 pages of reference docs? I didn't see a 'Getting Started' or 'Examples' or 'Guide' or 'Tutorial' section, and I stand by my assessment of their attitude to newcomers.
> maybe they didn't set out to create something that provides insights to people who first came across their field of study five minutes earlier
I wish people would stop submitting those kinds of links to HN. Also: I have an axe to grind (about terrible unapproachable documentation, academic or corporate or hobbyist).
I wish people would stop submitting those kinds of links to HN.
Why would you want that?
Surely the point of HN is to read about things on the bleeding edge of what is possible. Sometimes that blood comes from being too ambitious and sometimes from missing docs. Either way, it seems a small price to pay.
I certainly hope not. Most of the research papers have been related to refactoring TCL code, rather than on any correspondence between the models and reality.
You don't need a computer model to tell you that people don't like when you bomb their neighborhoods, kill their friends, and threaten their access to food and water. You definitely don't need a model to suggest that maybe if there are also some riots, people will like the outcome.
I don't quite understand the criticism... Yes, this is not (yet) a model that predicts the future. But when has "we're not there yet" ever been a good reason to stop?
Societies are comparable to the weather, in that they are dynamic systems highly susceptible to instability when simulated (-> chaos theory). That's why it took from the 60s until the mid-90s before weather prognosis consistently started to beat "it's going to be like today".
Societies are even more difficult, because the individual units of simulation are already quite chaotic: while the result of increased pressure on a fixed amount of air has been measured to whatever precision you could want, and has always followed a rather simple formula, doing the same with human behaviour is still far in the future.
But that doesn't mean such models are useless. Just think of the fight over the results of raising the minimum wage. Some people should initially lose their job. Others have more money. How are they spending that money? Does that spending create jobs? What's the effect on inflation? etc...
Having an agent-based model forces you to describe every assumption you're making, free of ambiguity. Maybe if we ask economists to build something like that, they'll at least grow an appreciation for what they don't yet know.
If you use a Taylor series to model a sine curve, you can get arbitrarily good results over an arbitrarily large range, yet the model will always fail catastrophically when the modeler wants to figure out the value that will give sin(x) = 2.
This seems like a system designed for the analyst who wants to make sin(x) = 2, and will blame the participants when reality doesn't match the model.
People don't do things like that because they lack understanding of the consequences, they do so because they don't care about the consequences if their other objectives are met.
The weakness of scientism is its its agnosticism; the scientific method can test hypotheses but has nothing to say about the selection of premises. If you (as a hypothetical dictator) think it's perfectly OK to inflict suffering on people you deem inferior, science can be applied to tell you how to make their lives as miserable as possible.
Philosophy has fallen into a social ditch in that it seems mainly to talk about itself and doesn't have the predictive power of science, but I think it's irrational to expect intellectual tools to provide you with answers to ethical questions.
