Digital Twins are by their very nature numerical models.
The physical world is made up of non-linear dynamical systems.
In iterated numerical models, all non-linear dynamical systems are extremely sensitive to initial conditions and tend to chaotic behaviors past certain parameters -- in ocean/atmosphere systems, this produces hurricanes. In biological systems, this produces population booms and busts. In airplane design, this produces turbulence which can shake a new design to pieces. Almost all real world systems are non-linear, thus prone to the same problems.
The problem of numerical chaos can not be eliminated from such models.
In 2014 Schmidt and Sherwood published a paper, A Practical Philosophy of Complex Climate Modelling, Euro Jnl Phil Sci. The authors propose to “explore specifically to what extent complex simulation in climate science is a new ‘pillar’ of inquiry as opposed to an expansion or cross fertilization of existing notions of theory and experiment from the point of view of model developers and users of the simulations.”
It’s an extremely bold proposition that climate models might themselves be an independent source of incremental Information, separate from prospective data acquisition, i.e. experimentation. It’s an extraordinary claim, requiring meticulous proof.
But climate modelers have granted their models this special “‘pillar’ of inquiry” status completely outside the frame of the ordinary empirical methods that form the core of the scientific method--- specifically, prospective data acquisition. So, you see this extremely odd and telling phenomenon of climate modelers calling their climate runs experiments and describing the output as data.
Fascinating to see how this expands, but I tend to think more hesitantly regarding a lot of modeling studies, especially ones that will inevitably drive policymaking. There are obviously a lot of interesting implications from a scientific perspective. But, how well can these models and simulations be vetted before any of their results influence decision making? These models are going to inherently have to make smoothing assumptions and cannot accurately model every aspect of an open system that is our planet's climate. If a true digital twin existed, my hesitancy would go away. But, we seem to be many centuries away from the computing power required to run simulations on an entire planet.
Smart people have failed to model the human body adequately for goals such as time-related prediction of illnesses.
When there is a usefully predictive human model, we might have demonstrated the skill to make a useful earth model, even though it would be less beneficial. Geoff S
Fascinating literature of critique, I enjoyed this entree into it. Dirty bathwater indeed, but readers should be careful too: thar be babies. There are actors with interests in not having quantitative frameworks at all: those who want to dump externalities for instance.
In mild defense of complex, overly detailed (in the sense of wrong-in-detail) mechanisms within models: Data assimilation can sometimes gain value from even overly simplistic mechanistic linkages. Observations are always specific. If a model doesn't contain that specific thing (even if it is part of a cluster of chaotic mechanisms that together are a source of unpredictability), then observations can never be used to confront the holistic framework embodied in models (even ones that cannot be reliably integrated forward in time to make temporal predictions). That holistic framework has value even when it does not comprise a literally accurate forward prediction model, I would contend. Well, that is my mild defense, even as the overall critique seems valuable and on-target in many ways.
Sorry for commenting twice, but I'm reading Neil Gorsuch's book "Over-ruled", and stumbled on a gem. Gorsuch believes Friedrich Hayek's greatest contribution lay in the discovery of a simple yet profound truth: "Man does not and cannot know everything, and when he acts as if he does, disaster follows."
It takes a chest full of hubris to move from "All models are wrong. Some are useful" and "Climate is a non-linear chaotic system. Therefore long-term predictions of climate are not possible." to the idea of a digital twin of earth. To even attempt it assumes that all relevant physical relationships are already completely understood. Never mind about the failure of the models to predict this year's hurricanes. Never mind about the inability of the models to deal with clouds. Wow. I am awed by the arrogance of the modelers.
I was going to make much the same comment as you but decided to read down first.
I posted a line here recently from a Ryan Maue weather trader post where he laments that they clearly haven’t figured out the hurricane model, so what is the purpose of the earth digital twin?
GIGO.
The digital twin in manufacturing works because the physical parameters can be measured.
Earth as a complete system? We seem to understand 5% of it, maybe, on a good day, 1% if you are Piltdown Mann.
At a meta level, my comment on this is the same as on almost all of the posts here. A problem is identified, but the policy implications are left hanging.
What kind of modeling _should_ de done for policy making?
The earth’s climate is a chaotic system affected by many independent and dependent variables that have yet to all be identified, much less their interactions/relationships and effectual strengths determined. The predictions of the various climate models vary greatly and they all continue to be wrong.
The documentary film, “The Most Unknown” follows nine scientists that are at the forefront of their disciplines. It is a very well made and thought-provoking film. It drives home the point that as science pushes the envelope of human knowledge, it is becoming clearer just how little we know relative to all that is to be known/reality. We are becoming knowledgeable enough to understand that we know next to nothing about our universe, our world, our bodies, our psychology, etc.
I don’t think it was the intent of the makers of the documentary (but it may have been) to point out just how arrogant humans are regarding our perceived level of knowledge relative to all there is to be known. But it shows the hubris to be ridiculously large. One of the scientists in the film, regarding the complexity and vastness of unknowns in his field of study, said he feels like a frog trying to understand Einstein’s theory of relativity.
I am not a scientist, so when the James Webb telescope was aimed into space and turned on, and pretty much every scientist on Earth said, “Holy shit!” I suspected we still have a lot to learn.
All very ture, but what tis the alternative to making decisions on the basis of the lease wrong knowledge we have at the time? Ptolemy was wrong, but they could still predict eclipses.
I’m not sure who should be credited with this saying but I think it makes a lot of sense, “The wise man learns from the mistakes of others. The ordinary man learns only from his own mistakes. The fool learns nothing from mistakes.” It is our human condition to have to make many decisions based on little or incomplete information. However, is it not foolish to make an important decision based on information that is demonstrably wrong just because it is all you have? In most of these type cases you would have better odds of making a good decision by flipping a coin.
I think the deeper message here is not so much that scientific models can or can’t predict the behavior of vastly complicated chaotic systems, but the concept has already been hijacked and polluted by politics. If we had statesmen in power, decisions would be made with Hippocratic oath style concepts in mind. Alas, we have politicians in power that do not let a crisis (real or manufactured) go to waste regarding an opportunity to acquire more power.
This is an excellent post underscoring how far off the rails the modelers can take us if we let them.
Does the paper underlying the post represent the conclusions of a conference or some other series of discussions? I am interested in knowing what brought the 11 authors of the paper together.
This posted reminded me of a recent essay in Aeon on chaos theory and causation as seen through two lenses, the deterministically physical and the [reality/illusion] of human agency.
“Classical physics not only presented an orderly universe among Newton’s followers, but it also instilled a profound sense of mastery over the natural world. Newton’s discoveries fostered the belief that the Universe, previously shrouded in mystery, was now laid bare, sparking an unprecedented optimism in the power of science. Armed with Newton’s laws and revolutionary mathematics, leading thinkers felt they had finally unlocked the secrets of reality.
…
And yet, the 20th century witnessed a dramatic shift with the emergence of relativity, which redefines our understanding of space and time; QM, which revolutionised our understanding of the subatomic world; and chaos theory. The orderly and predictable world of Newtonian physics, the dream of a mechanical universe ready to unveil her innermost workings, was, happily or not, something of an illusion. In the 20th century, science revealed a far more intricate, less predictable and, indeed, chaotic universe.“
Almost 300 years ago, poet Alexander Pope in his Essay on Man expressed the same idea of an orderly universe and sense of mastery with this famous couplet-
Modeling the effects of already emitted CO2 and policy changes to reduce future emissions has always been the way to make decisions mitigation and adaption policies and investments.
Whether a single model on a concatenation of separate more local and regional models is best for these purposes sounds like a secondary question, although I agree that concatenation seems like the better route.
Digital Twins are by their very nature numerical models.
The physical world is made up of non-linear dynamical systems.
In iterated numerical models, all non-linear dynamical systems are extremely sensitive to initial conditions and tend to chaotic behaviors past certain parameters -- in ocean/atmosphere systems, this produces hurricanes. In biological systems, this produces population booms and busts. In airplane design, this produces turbulence which can shake a new design to pieces. Almost all real world systems are non-linear, thus prone to the same problems.
The problem of numerical chaos can not be eliminated from such models.
NCAR/UCAR proved this in its paper called "The Community Earth System Model (CESM) Large Ensemble Project: A Community Resource for Studying Climate Change in the Presence of Internal Climate Variability". (available in pdf: https://journals.ametsoc.org/downloadpdf/view/journals/bams/96/8/bams-d-13-00255.1.pdf)
"The Black Swan" by Nicholas Nassim Taleb, and the sequels, should be required reading for those who think we can model the future.
In 2014 Schmidt and Sherwood published a paper, A Practical Philosophy of Complex Climate Modelling, Euro Jnl Phil Sci. The authors propose to “explore specifically to what extent complex simulation in climate science is a new ‘pillar’ of inquiry as opposed to an expansion or cross fertilization of existing notions of theory and experiment from the point of view of model developers and users of the simulations.”
It’s an extremely bold proposition that climate models might themselves be an independent source of incremental Information, separate from prospective data acquisition, i.e. experimentation. It’s an extraordinary claim, requiring meticulous proof.
But climate modelers have granted their models this special “‘pillar’ of inquiry” status completely outside the frame of the ordinary empirical methods that form the core of the scientific method--- specifically, prospective data acquisition. So, you see this extremely odd and telling phenomenon of climate modelers calling their climate runs experiments and describing the output as data.
The author adopts a very nuanced tone to talk about a subject that is, at best, totally wrong and, at worst, horribly insane.
Wrong because the mass of parameters and co-factors, including human conjectures, makes it impossible to develop any plausible and useful scenarios.
Insane because, however wrong and useless, it will be misused for evil purposes.
Fascinating to see how this expands, but I tend to think more hesitantly regarding a lot of modeling studies, especially ones that will inevitably drive policymaking. There are obviously a lot of interesting implications from a scientific perspective. But, how well can these models and simulations be vetted before any of their results influence decision making? These models are going to inherently have to make smoothing assumptions and cannot accurately model every aspect of an open system that is our planet's climate. If a true digital twin existed, my hesitancy would go away. But, we seem to be many centuries away from the computing power required to run simulations on an entire planet.
In honor of climate modeling I’m rewatching The Day After Tomorrow
Smart people have failed to model the human body adequately for goals such as time-related prediction of illnesses.
When there is a usefully predictive human model, we might have demonstrated the skill to make a useful earth model, even though it would be less beneficial. Geoff S
Fascinating literature of critique, I enjoyed this entree into it. Dirty bathwater indeed, but readers should be careful too: thar be babies. There are actors with interests in not having quantitative frameworks at all: those who want to dump externalities for instance.
In mild defense of complex, overly detailed (in the sense of wrong-in-detail) mechanisms within models: Data assimilation can sometimes gain value from even overly simplistic mechanistic linkages. Observations are always specific. If a model doesn't contain that specific thing (even if it is part of a cluster of chaotic mechanisms that together are a source of unpredictability), then observations can never be used to confront the holistic framework embodied in models (even ones that cannot be reliably integrated forward in time to make temporal predictions). That holistic framework has value even when it does not comprise a literally accurate forward prediction model, I would contend. Well, that is my mild defense, even as the overall critique seems valuable and on-target in many ways.
Sorry for commenting twice, but I'm reading Neil Gorsuch's book "Over-ruled", and stumbled on a gem. Gorsuch believes Friedrich Hayek's greatest contribution lay in the discovery of a simple yet profound truth: "Man does not and cannot know everything, and when he acts as if he does, disaster follows."
It takes a chest full of hubris to move from "All models are wrong. Some are useful" and "Climate is a non-linear chaotic system. Therefore long-term predictions of climate are not possible." to the idea of a digital twin of earth. To even attempt it assumes that all relevant physical relationships are already completely understood. Never mind about the failure of the models to predict this year's hurricanes. Never mind about the inability of the models to deal with clouds. Wow. I am awed by the arrogance of the modelers.
Hubris is another good word.
I was going to make much the same comment as you but decided to read down first.
I posted a line here recently from a Ryan Maue weather trader post where he laments that they clearly haven’t figured out the hurricane model, so what is the purpose of the earth digital twin?
GIGO.
The digital twin in manufacturing works because the physical parameters can be measured.
Earth as a complete system? We seem to understand 5% of it, maybe, on a good day, 1% if you are Piltdown Mann.
At a meta level, my comment on this is the same as on almost all of the posts here. A problem is identified, but the policy implications are left hanging.
What kind of modeling _should_ de done for policy making?
The earth’s climate is a chaotic system affected by many independent and dependent variables that have yet to all be identified, much less their interactions/relationships and effectual strengths determined. The predictions of the various climate models vary greatly and they all continue to be wrong.
https://www.drroyspencer.com/2024/02/u-s-a-temperature-trends-1979-2023-models-vs-observations/
The documentary film, “The Most Unknown” follows nine scientists that are at the forefront of their disciplines. It is a very well made and thought-provoking film. It drives home the point that as science pushes the envelope of human knowledge, it is becoming clearer just how little we know relative to all that is to be known/reality. We are becoming knowledgeable enough to understand that we know next to nothing about our universe, our world, our bodies, our psychology, etc.
I don’t think it was the intent of the makers of the documentary (but it may have been) to point out just how arrogant humans are regarding our perceived level of knowledge relative to all there is to be known. But it shows the hubris to be ridiculously large. One of the scientists in the film, regarding the complexity and vastness of unknowns in his field of study, said he feels like a frog trying to understand Einstein’s theory of relativity.
I am not a scientist, so when the James Webb telescope was aimed into space and turned on, and pretty much every scientist on Earth said, “Holy shit!” I suspected we still have a lot to learn.
All very ture, but what tis the alternative to making decisions on the basis of the lease wrong knowledge we have at the time? Ptolemy was wrong, but they could still predict eclipses.
I’m not sure who should be credited with this saying but I think it makes a lot of sense, “The wise man learns from the mistakes of others. The ordinary man learns only from his own mistakes. The fool learns nothing from mistakes.” It is our human condition to have to make many decisions based on little or incomplete information. However, is it not foolish to make an important decision based on information that is demonstrably wrong just because it is all you have? In most of these type cases you would have better odds of making a good decision by flipping a coin.
I think the deeper message here is not so much that scientific models can or can’t predict the behavior of vastly complicated chaotic systems, but the concept has already been hijacked and polluted by politics. If we had statesmen in power, decisions would be made with Hippocratic oath style concepts in mind. Alas, we have politicians in power that do not let a crisis (real or manufactured) go to waste regarding an opportunity to acquire more power.
This is an excellent post underscoring how far off the rails the modelers can take us if we let them.
Does the paper underlying the post represent the conclusions of a conference or some other series of discussions? I am interested in knowing what brought the 11 authors of the paper together.
This posted reminded me of a recent essay in Aeon on chaos theory and causation as seen through two lenses, the deterministically physical and the [reality/illusion] of human agency.
“Classical physics not only presented an orderly universe among Newton’s followers, but it also instilled a profound sense of mastery over the natural world. Newton’s discoveries fostered the belief that the Universe, previously shrouded in mystery, was now laid bare, sparking an unprecedented optimism in the power of science. Armed with Newton’s laws and revolutionary mathematics, leading thinkers felt they had finally unlocked the secrets of reality.
…
And yet, the 20th century witnessed a dramatic shift with the emergence of relativity, which redefines our understanding of space and time; QM, which revolutionised our understanding of the subatomic world; and chaos theory. The orderly and predictable world of Newtonian physics, the dream of a mechanical universe ready to unveil her innermost workings, was, happily or not, something of an illusion. In the 20th century, science revealed a far more intricate, less predictable and, indeed, chaotic universe.“
https://aeon.co/essays/does-chaos-theory-square-classical-physics-with-human-agency
Almost 300 years ago, poet Alexander Pope in his Essay on Man expressed the same idea of an orderly universe and sense of mastery with this famous couplet-
"Nature and nature's laws lay hid in night.
"God said, 'Let Newton be!', and all was light.
I am totally confused.
Modeling the effects of already emitted CO2 and policy changes to reduce future emissions has always been the way to make decisions mitigation and adaption policies and investments.
Whether a single model on a concatenation of separate more local and regional models is best for these purposes sounds like a secondary question, although I agree that concatenation seems like the better route.
What if the models have the effect of co2 in the atmosphere completely wrong, bass-ackward, as they say?
Do you really believe that is settled?
If Charles Ponzi were alive today, I'm pretty sure he would be involved in climate modeling.