4 ms·

Based on reading an article about detailed work on making a model more accurate you’ve concluded that models are useless? Just because models aren’t infinitely

by ximeng 5y ago

Based on reading an article about detailed work on making a model more accurate you’ve concluded that models are useless? Just because models aren’t infinitely perfect doesn’t mean they don’t have some predictive power. Yes you might drop a ball and it gets destroyed by a nuclear explosion before hitting the floor but it doesn’t mean a model that says it will likely hit the floor is useless! At large scale climate change isn’t that complicated - sun hits earth, heats it up, some energy escapes. We can get a pretty good picture from models and the details just add to that.

adamrezich 5y ago
> Based on reading an article about detailed work on making a model more accurate you’ve concluded that models are useless? no—I, knowing basically nothing about climate science and having been taught to more or less blindly trust climate models as accurate predictors of the future, read an article that said we have no idea how to model clouds with anything resembling accuracy, because clouds by themselves are crazy complex things. this led me to realize that this should have been intuitive knowledge without even reading the article, which led me to truly consider the scope of the climate system w.r.t. how it could possibly be modeled by computers, which led me to realize that the idea that it's possible for a data-driven computer model to predict anything about the future with any kind of reliability is pretty silly to begin with.
- ximeng 5y ago
  This just seems like a ridiculous position to hold: data-driven computer models can’t predict anything with any reliability?! Pretty much all predictions about the future rely on data-driven models! Yes models are not perfect, but the answer’s not to close your eyes and hope!
- lukifer 5y ago
  The right way to think about models at this scale is complexity isn't "Accurate" vs. "Inaccurate"; it's to use probabilistic/Bayesian reasoning. No map is a perfect representation of the territory, and no model can encapsulate the entirety of the earth. That doesn't mean maps and models are useless. A couple counterpoints from former climate skeptics: - Richard Mueller is a prominent physicist who founded an institute to debunk anthropogenic climate change, only to change his mind after thorough examination of the data. https://youtu.be/Sme8WQ4Wb5w https://youtu.be/Sme8WQ4Wb5w - Jerry Taylor is a libertarian and former climate debunker for the Cato Institute; a noteworthy pivot moment was speaking to a risk analyst for a financial institution, whose job it is to make bets on complex systemic risks in the context of uncertainty. He claimed that from a financial lens, AGW was a slam-dunk case for mitigation, not because of the certainty of any one model, but of the full probability distribution in aggregate. https://www.reckonings.show/episodes/17 https://www.reckonings.show/episodes/17
  - adamrezich 5y ago
    sure, that's fine. the problem arises when politicians state that things must be done by x date or y year or else we're going to have a CLIMATE CATASTROPHE, because the models say so. models can't predict anything like this
    - lukifer 5y ago
      I agree there is a moral hazard / perverse incentive when it comes to alarmism, especially in politics (eg, specious references to "mushroom clouds" in the run-up to the Iraq invasion). But as good Bayesians, we also have to consider the opposite case: crises and "black swan" events do happen. How do we filter signal from noise? We probably could have benefitted from more "enlightened alarmism" re: COVID in Jan/Feb 2020, when billions in preventative measures could have saved trillions to world GDP (and millions of lives). It's a double-bind politically: given a non-trivial long-term risk, how does one frame its urgency? A calm and measured tone spurs little action (the unsqueaky wheel doesn't get the grease); a tone of urgency and panic bypasses badly-needed rational faculties, and/or flips the "bullshit detectors" based on the history of proverbial boys crying wolf. While I'm not opposed to FDR-ish green infrastructure plans, the answer from economists is undramatic and uncontroversial [0]: tax the externality to establish a price signal, and return the dividend back to citizens, whose consumption behaviors then change in response to those signals [1]. Of course, we've had to contend with decades of FUD from vested O&G interests, no different than the tobacco cancer debacle [2], despite having had credible evidence for AGW since the late 70's. I'm happy to concede the potential for scientific groupthink and perverse Big Government incentives on the green side, so long as we're also acknowledging the enormous financial incentives for muddying the waters (uncertainty = inaction = status quo = profit). [0] https://www.wsj.com/articles/economists-statement-on-carbon-dividends-11547682910 https://www.wsj.com/articles/economists-statement-on-carbon-... [1] https://en.wikipedia.org/wiki/Pigovian_tax https://en.wikipedia.org/wiki/Pigovian_tax [2] https://www.barbarafreese.com/industrialstrength-denial https://www.barbarafreese.com/industrialstrength-denial

by ximeng 5y ago

adamrezich 5y ago

> Based on reading an article about detailed work on making a model more accurate you’ve concluded that models are useless? no—I, knowing basically nothing about climate science and having been taught to more or less blindly trust climate models as accurate predictors of the future, read an article that said we have no idea how to model clouds with anything resembling accuracy, because clouds by themselves are crazy complex things. this led me to realize that this should have been intuitive knowledge without even reading the article, which led me to truly consider the scope of the climate system w.r.t. how it could possibly be modeled by computers, which led me to realize that the idea that it's possible for a data-driven computer model to predict anything about the future with any kind of reliability is pretty silly to begin with.

ximeng 5y ago
This just seems like a ridiculous position to hold: data-driven computer models can’t predict anything with any reliability?! Pretty much all predictions about the future rely on data-driven models! Yes models are not perfect, but the answer’s not to close your eyes and hope!
lukifer 5y ago
The right way to think about models at this scale is complexity isn't "Accurate" vs. "Inaccurate"; it's to use probabilistic/Bayesian reasoning. No map is a perfect representation of the territory, and no model can encapsulate the entirety of the earth. That doesn't mean maps and models are useless. A couple counterpoints from former climate skeptics: - Richard Mueller is a prominent physicist who founded an institute to debunk anthropogenic climate change, only to change his mind after thorough examination of the data. https://youtu.be/Sme8WQ4Wb5w https://youtu.be/Sme8WQ4Wb5w - Jerry Taylor is a libertarian and former climate debunker for the Cato Institute; a noteworthy pivot moment was speaking to a risk analyst for a financial institution, whose job it is to make bets on complex systemic risks in the context of uncertainty. He claimed that from a financial lens, AGW was a slam-dunk case for mitigation, not because of the certainty of any one model, but of the full probability distribution in aggregate. https://www.reckonings.show/episodes/17 https://www.reckonings.show/episodes/17
- adamrezich 5y ago
  sure, that's fine. the problem arises when politicians state that things must be done by x date or y year or else we're going to have a CLIMATE CATASTROPHE, because the models say so. models can't predict anything like this
  - lukifer 5y ago
    I agree there is a moral hazard / perverse incentive when it comes to alarmism, especially in politics (eg, specious references to "mushroom clouds" in the run-up to the Iraq invasion). But as good Bayesians, we also have to consider the opposite case: crises and "black swan" events do happen. How do we filter signal from noise? We probably could have benefitted from more "enlightened alarmism" re: COVID in Jan/Feb 2020, when billions in preventative measures could have saved trillions to world GDP (and millions of lives). It's a double-bind politically: given a non-trivial long-term risk, how does one frame its urgency? A calm and measured tone spurs little action (the unsqueaky wheel doesn't get the grease); a tone of urgency and panic bypasses badly-needed rational faculties, and/or flips the "bullshit detectors" based on the history of proverbial boys crying wolf. While I'm not opposed to FDR-ish green infrastructure plans, the answer from economists is undramatic and uncontroversial [0]: tax the externality to establish a price signal, and return the dividend back to citizens, whose consumption behaviors then change in response to those signals [1]. Of course, we've had to contend with decades of FUD from vested O&G interests, no different than the tobacco cancer debacle [2], despite having had credible evidence for AGW since the late 70's. I'm happy to concede the potential for scientific groupthink and perverse Big Government incentives on the green side, so long as we're also acknowledging the enormous financial incentives for muddying the waters (uncertainty = inaction = status quo = profit). [0] https://www.wsj.com/articles/economists-statement-on-carbon-dividends-11547682910 https://www.wsj.com/articles/economists-statement-on-carbon-... [1] https://en.wikipedia.org/wiki/Pigovian_tax https://en.wikipedia.org/wiki/Pigovian_tax [2] https://www.barbarafreese.com/industrialstrength-denial https://www.barbarafreese.com/industrialstrength-denial