15 ms·

Science without Validation in a World without Meaning

127 points by nkurz 6y ago

mturmon 6y ago
A very nice article -- more pessimistic than I'd be -- about the limits of outsider understanding when predictions come from complex models that few understand.
- webmaven 6y ago
  The pessimism seems to comes from the proposition that no one really understands models beyond a certain complexity, even insiders. Even whoever created the model.
mjparrott 6y ago
I could use a TLDR on this one ...
- guscost 6y ago
  We can't easily validate models of reality that are complex/stochastic, because any experimental result could be a statistical anomaly. In particular this is a problem for models of long-term and/or global phenomena like climate or public health, since we only get one "experimental run", so to speak. Therefore: > Confronting the problems of complexity, validation, and model uncertainty, I have previously identified four options for moving ahead: (1) dispense with modeling complex systems that cannot be validated; (2) model complex systems and pretend they are validated; (3) model complex systems, admit that the models are not validated, use them pragmatically where possible, and be extremely cautious when interpreting them; (4) strive to develop a new and perhaps weaker scientific epistemology.
  - dang 6y ago
    I haven't read the article, but #4 is the interesting item on that list. #3 is the Goldilocks option, the one that sounds between-the-extremes like the just-right porridge that Goldilocks ate—but it's not realistic, because "be extremely cautious when interpreting them" exceeds what the human mind is capable of.
    - guscost 6y ago
      It may be possible for a person to succeed with #3, but I have a hard time believing it could work in a community of people who disagree. For #4, the closest thing I can imagine is a utilitarian approach, which of course I’m going to prefer as an engineer, and which (I think) ultimately reduces to #1. That is a tough problem I don’t expect to see solved in my lifetime, but I’d be happy to be proven wrong.
      - btrettel 6y ago
        On reflection, option 1 includes like Taleb's philosophy. At first I wrote it off as leading to arguments like "no one knows so let's do what I, an 'expert' say", which I hate. But it's written generically enough to include rules like Minimax, where you don't need model estimates to make a decision. This rule is very conservative, though. I think there are a fair number of people who succeed at option 3. I can think of a few people who I believe do option 3 well, but as you indicated, they're individuals. I think that the US Department of Energy is probably the closest organization that I'm aware of to succeed at this. (Or at least some DOE labs; I can say from personal experience that groups at both Los Alamos and Sandia take this fairly seriously, though I don't think it has fully permeated their culture.) I'll have to think more about this. Option 3 is my basic approach, I'd like to think that I succeed at it, e.g.: https://news.ycombinator.com/item?id=23397785 https://news.ycombinator.com/item?id=23397785 A major problem with option 3 is that it runs the risk of people claiming to do option 3 but actually doing option 2. I think this happens regularly but it's due to ignorance, not malice. Ultimately I think we need to change scientific standards and the STEM curriculum before option 3 becomes tractable on a large scale, but even then I'm not sure it'll work because it'll always be easier to claim to do option 3 while actually doing option 2. I agree with the writer that option 4 is more of a long-term goal. I wish the comments on this article focused more on #4 than what was discussed...
        webmaven 6y ago
        > (Or at least some DOE labs; I can say from personal experience that groups at both Los Alamos and Sandia take this fairly seriously, though I don't think it has fully permeated their culture.) It has not. In the current funding environment it can't, given that the way research grants and contracts are awarded will always constrain organizational culture.
snowwrestler 6y ago
An article by an electrical engineer, published in a political journal, about climate change... I had low hopes, which were met. IMO few categories of professions have a harder time understanding cutting-edge science than engineers. That is because they think they know science because they use similar mathematical and technical tools, when in fact the professions do the exact opposite of one another. Engineers use what is known and understood to construct systems that can be validated. Scientists investigate unknown systems to try to construct knowledge and understanding. The worst thing, professionally, for an engineer is to not know or understand your work. The worst thing, for a scientist, is to spend too much time on things that are already known and understood. Engineers love to point out how scientists don't know, and can't prove, whether their climate models are accurate. Scientists know that that is the whole point of building such models. Working without knowing whether you're correct is not epistemological conflict. It's the fundamental condition of being a scientist. Yes, it would be better if we knew everything and only consulted systems that are provably correct. But we don't. And the only way to expand what we know is to spend a ton of time doing things that we don't know. We haven't found any other way to do it. Writing hand-wringing articles about the state of science from the sidelines does not advance human knowledge. You can't expand the map by standing inside the border and complaining about how hard it is to see past it.
- pdonis 6y ago
  > Working without knowing whether you're correct is not epistemological conflict. It's the fundamental condition of being a scientist. This is true, but it's also true that, when making public policy, we should be thinking like engineers, not scientists. Public policy is not about constructing new knowledge. It's about working out conflicts between different values and priorities based on our best current knowledge. To the extent the article is talking about public policy, I think it makes a valid point that the limitations of our knowledge in areas highly relevant to public policy are very often not recognized or taken into account when making public policy.
  - snowwrestler 6y ago
    There are no sidelines in public policy. There's nowhere to wait to see how things turn out before making a decision. So, there are no neutral decisions. "Waiting until we're more sure about something" is not actually waiting, it is an affirmative decision to disregard what we think we know right now. That can be a valid choice, and something to debate, but it's fundamentally different from the feined neutrality of the concept of waiting to see. To make such an argument requires attacking that knowledge directly, showing specifically how it's wrong. Not just vaguely complaining that it's not good enough yet. This is actually well-understood in public policy when it comes to other areas like economics or defense; leaders act to address needs in real time, making the best decisions they can with the information available to them. "A good plan violently executed now is better than a perfect plan executed at some indefinite time in the future.” ― General Patton Engineering, as a discipline, cannot cover all of life. It largely constrains itself to situations that are understood, and can do so because there are other complementary disciplines that create the knowledge it uses. But in public policy there's not a complementary Earth or society that we can wait for.
Hnrobert42 6y ago
A little about the source: https://en.m.wikipedia.org/wiki/American_Affairs https://en.m.wikipedia.org/wiki/American_Affairs
- Hnrobert42 6y ago
  I got downvoted for linking to a document about the source. WTF?
guscost 6y ago
> Four conditions must be satisfied to have a valid scientific theory: (1) There is a mathematical model expressing the theory. (2) Precise relationships, known as “operational definitions,” are specified between terms in the theory and measurements of corresponding physical events. (3) There are validating data: there is a set of future quantitative predictions derived from the theory and measurements of corresponding physical events. (4) There is a statistical analysis that supports acceptance of the theory, that is, supports the concordance of the predictions with the physical measurements—including the mathematical theory justifying the application of the statistical methods. This definition appears rigorous at a glance, but it is deficient. We cannot properly test a theory if it only predicts things which we already expect to happen. Popper said that scientific theories must instead make "risky predictions": "Confirmations should count only if they are the result of risky predictions; that is to say, if, unenlightened by the theory in question, we should have expected an event which was incompatible with the theory–an event which would have refuted the theory."
cgiles 6y ago
I work in molecular biology research, and I think this is a great article that strikes at the heart of many problems in the field. I can't comment on the climate change stuff, although I wish he hadn't included it because it was almost certain to distract people from the overall point. The problem is that there are no remotely comprehensive, predictive, and mathematical models of what goes on inside of cells. It is pure empiricism: you run an intervention, and see what happens. Write it up in a paper. All well and good, except there are no viable models of what is happening inside that are predictive in the sense of being able to know what an intervention will do until you test it. We really need that if we want to develop treatments for molecular diseases that are more than marginally better. The Santa Fe Institute, systems biology people, and others were working hard on this problem at the turn of the century, but progress has stalled. It's too hard. We don't know how to do it. A new "mathematical epistemology" that could handle this problem would be a huge step forward, if it is possible. I can see why the author would extend this idea to things like economics or climate science. The thought in systems research was that, perhaps, different fields share similar underlying "complex systems" mechanisms, and if we can solve the problem in one area, we may have insights for how to do it elsewhere.
- memexy 6y ago
  What's missing from current mathematics to make predictive models for biology? I did a search for "neural network cell simulation" and got a few hits, e.g. https://ieeexplore.ieee.org/document/8805421 https://ieeexplore.ieee.org/document/8805421. So it seems that people are working on the problem of predictability (or at least augmenting the researcher's/experimenter's ability to do some analysis ahead of time based on simplified models).
  - cgiles 6y ago
    > What's missing from current mathematics to make predictive models for biology? Well, I think that, no joke, there is a Nobel Prize waiting for anyone who knows the answer to that. I think this is the next big paradigm shift needed in biology, not to mention several other fields. Who is to say that the problem is strictly mathematical, though? It could be that the math exists, but no one knows how to fit existing data into it, or it could be that there is not enough data, or the right kind of data, to make such a model yet. It could be that both the data and algorithm exists, but we need to turn the Earth into computronium to run it. Who knows? > So it seems that people are working on the problem of predictability I'm sure they are. They have been for decades. The last time I did a systematic review of this area was before the resurgence of neural networks, so I can't really say what is the latest progress, or whether the progress in ANNs can inform this problem. I suspect it's very possible. The situation right now, as far as I know is that: A) most biologists don't even know this is a problem, and B) those who do, don't have any idea what the solution is, or if one even exists (note the author of the linked article was pessimistic on that point).
btrettel 6y ago
This article meanders too much. The basic point is this: > While stochastic models present us with numerous difficulties, an even more perplexing conundrum faces contemporary scientists and engineers who wish to model highly complex systems involving hundreds or thousands of variables and model parameters. Owing to their sheer number, many model parameters cannot be accurately estimated via experiment and are left uncertain. As a consequence of uncertainty, for each different set of possible values for the unknown parameters, there is a different model—possibly an infinite number of models. > Confronting the problems of complexity, validation, and model uncertainty, I have previously identified four options for moving ahead: (1) dispense with modeling complex systems that cannot be validated; (2) model complex systems and pretend they are validated; (3) model complex systems, admit that the models are not validated, use them pragmatically where possible, and be extremely cautious when interpreting them; (4) strive to develop a new and perhaps weaker scientific epistemology. At the moment I am in favor of option 3, though option 4 might be more appealing in the future. This isn't an easy option to take. Recently I had an article accepted for publication where I basically argued that no models, including my own, were truly validated because none fit a non-naive data set well. (In another paper I argued that most data sets used for validation are too easy to match because they don't cover the parameter space well.) A reviewer recommended rejection, basically saying that because the model isn't validated, it shouldn't be published. So much for being intellectually honest! The paper was eventually accepted after I made it more clear that none of the popular models work that well (some are absolutely terrible in my view), and that my model improves on the status quo in a few ways. Note that this situation isn't exactly the same as that described in the link. In the case of my article, I think we can get enough data to validate a model. We just don't have that data at present.
- mjburgess 6y ago
  I tend to come down on (1) in many cases, eg., social psychology. I'd be open to more of this sort of modelling if (4) was a realistic and completed project. The reality of (4) however is saying, to the public, things like: there may either be very limited climate change, or the entire world will be destroyed as we know it. I don't see the current public understanding of science, esp. via journalism, as fit for such realities.
Gatsky 6y ago
"Medicine is a form of engineering in which the physician takes some action (drug, surgery, etc.) to alter the behavior of a physical system, the patient. Its underlying science is biology. Hence our ability to characterize medical knowledge depends on our ability to represent and validate biological knowledge." This is really very wrong. The 'physical system' is conscious and suffering, which excludes Medicine from even the most vaporous definition of engineering. Real progress in medicine is much more akin to exploration (in the Christopher Columbus sense) than engineering. Anyway, this is a typical engineer writing about biology. They always make the same mistakes: 1. Ignoring evolution. 2. Treating mechanistic models as first class citizens with disdain for data (eventhough the latter in the form of DNA is the sine qua non of all life) 3. Failing to understand that "The simplest complete model of an organism is the organism itself."[1] as von Neumann noted. [1] https://onezero.medium.com/the-future-of-computing-is-analog-e758471fbfe1 https://onezero.medium.com/the-future-of-computing-is-analog...
- Ma8ee 6y ago
  > "The simplest complete model of an organism is the organism itself." But that is trivially true about everything, including golf balls. The interesting question isn't if the model is complete, but if it is sufficient (or complete enough) to make predictions. And in the case of golf balls we can make quite good predictions with a very simple model, mostly because we are only interested in if it ends up in the hole or not. But a complete model of a golf ball would need to include the exact dynamics of the different layers and the turbulence around it, which we just can't do. Luckily, we don't need to. So what is a sufficient model of a living organism? That of course also depends on what we want to know. I am quite certain that modern egg farmers have, for their purposes, quite sufficient models of hens. They know how fast the hens grow, the effects of different amounts and types of food, how much heat they produce, and of course how many and how big eggs they lay, etc.. And it is of course possible that we need to know the exact wave function for the complete system that is the cell to model it sufficiently enough to, say, synthesise new drugs. But I don't think so.
  - Gatsky 6y ago
    It is definitely not trivially true about anything. A golf ball can be modelled easily. You are adding external complications to the system. An electron can be impossible to model if you inject it into a complex system. Anyway the point is a golf ball will not evolve into a sentient being who will then fly to mars and write poetry, which is what cells have done. This shows you can’t even put a meaningful bound on what you need to model for an organism.
kdkdk 6y ago
I think the system the author is looking for is probabilistic programming. It allows you to write programs/models with uncertain factors. Then at the end you can observe a few variables and it automatically computes the expectation and variance all the input variables. Check out psisolver.org for example.
ur-whale 6y ago
Feynmann: "It is whether or not the theory gives predictions that agree with experiment. It is not a question of whether a theory is philosophically delightful, or easy to understand, or perfectly reasonable from the point of view of common sense" Then proceeds to invent string theory. Don't get me wrong, I have a lot of admiration for Feynmann,but if he did indeed say the above, that strikes me as rather inconsistent : string theory is exactly what he preaches not to do : widely successful because of how beautiful it is, but unfalsifiable (as of yet).
N1H1L 6y ago
I read the full article. And it feels like much ado about nothing. One, the author for some reason is uncomfortable with a purely mathematical description of the world - without giving reasons beyond that humans cannot physically comprehend what the equations represent. Two, the whole piece disregards the advances in emergent phenomena, complex systems and statistics. Yes, we do not understand how individual electrons look like in copper, but statistical descriptions of copper dimensions, purity and grain size are enough to get an exceptionally accurate idea of the electrical behavior of a copper wire. This extends for exceptionally complex systems like lungs (smoking will significantly increase your cancer/emphysema risk), or planetary systems (increasing CO2 concentrations will increase surface temperatures). Three, There is also a weird bias against modeling. I am an experimental scientist, but I work closely with modellers, as all experimentalists do nowadays. Personally, I believe, that this bias against modeling often has a political component due to anthropogenic global warming. But in reality, there is a very vibrant dialogue between modellers and experimentalists, and hybrid scientists, people who are trained in both fields are the rage right now for hiring committees. Models have also improved dramatically - density functional theory is exceptionally accurate for smaller systems, while ReaxFF for dynamic systems and molecular dynamics simulations for more complex systems become better every year. The whole point about complex systems, applies to modelling too. I do not need to know the location of every molecule in the north Atlantic to predict the path of the hurricane - high pressure ridges and sea surface temperatures will give me really good results.
mensetmanusman 6y ago
This is a great article, but I wish it would have really dealt with the idea of complexity. Nature is not ‘unintelligible’ it is ‘complicated’ because of the large number of discrete interactive constitutive units. In fact, if you plot # of constitutive quarks/atoms being worked with versus uncertainty, you see an interesting layout of the sciences. e.g. take this plot by xkcd ( https://xkcd.com/435/ https://xkcd.com/435/ ) and think about how much matter is under observation (i.e. increasing complexity to the left). Somewhere on the left would be climate in this example (as in the article).

127 points by nkurz 6y ago

A very nice article -- more pessimistic than I'd be -- about the limits of outsider understanding when predictions come from complex models that few understand.

webmaven 6y ago
The pessimism seems to comes from the proposition that no one really understands models beyond a certain complexity, even insiders. Even whoever created the model.

mjparrott 6y ago

I could use a TLDR on this one ...

guscost 6y ago
We can't easily validate models of reality that are complex/stochastic, because any experimental result could be a statistical anomaly. In particular this is a problem for models of long-term and/or global phenomena like climate or public health, since we only get one "experimental run", so to speak. Therefore: > Confronting the problems of complexity, validation, and model uncertainty, I have previously identified four options for moving ahead: (1) dispense with modeling complex systems that cannot be validated; (2) model complex systems and pretend they are validated; (3) model complex systems, admit that the models are not validated, use them pragmatically where possible, and be extremely cautious when interpreting them; (4) strive to develop a new and perhaps weaker scientific epistemology.
- dang 6y ago
  I haven't read the article, but #4 is the interesting item on that list. #3 is the Goldilocks option, the one that sounds between-the-extremes like the just-right porridge that Goldilocks ate—but it's not realistic, because "be extremely cautious when interpreting them" exceeds what the human mind is capable of.
  - guscost 6y ago
    It may be possible for a person to succeed with #3, but I have a hard time believing it could work in a community of people who disagree. For #4, the closest thing I can imagine is a utilitarian approach, which of course I’m going to prefer as an engineer, and which (I think) ultimately reduces to #1. That is a tough problem I don’t expect to see solved in my lifetime, but I’d be happy to be proven wrong.
    - btrettel 6y ago
      On reflection, option 1 includes like Taleb's philosophy. At first I wrote it off as leading to arguments like "no one knows so let's do what I, an 'expert' say", which I hate. But it's written generically enough to include rules like Minimax, where you don't need model estimates to make a decision. This rule is very conservative, though. I think there are a fair number of people who succeed at option 3. I can think of a few people who I believe do option 3 well, but as you indicated, they're individuals. I think that the US Department of Energy is probably the closest organization that I'm aware of to succeed at this. (Or at least some DOE labs; I can say from personal experience that groups at both Los Alamos and Sandia take this fairly seriously, though I don't think it has fully permeated their culture.) I'll have to think more about this. Option 3 is my basic approach, I'd like to think that I succeed at it, e.g.: https://news.ycombinator.com/item?id=23397785 https://news.ycombinator.com/item?id=23397785 A major problem with option 3 is that it runs the risk of people claiming to do option 3 but actually doing option 2. I think this happens regularly but it's due to ignorance, not malice. Ultimately I think we need to change scientific standards and the STEM curriculum before option 3 becomes tractable on a large scale, but even then I'm not sure it'll work because it'll always be easier to claim to do option 3 while actually doing option 2. I agree with the writer that option 4 is more of a long-term goal. I wish the comments on this article focused more on #4 than what was discussed...
      - webmaven 6y ago
        > (Or at least some DOE labs; I can say from personal experience that groups at both Los Alamos and Sandia take this fairly seriously, though I don't think it has fully permeated their culture.) It has not. In the current funding environment it can't, given that the way research grants and contracts are awarded will always constrain organizational culture.

snowwrestler 6y ago

An article by an electrical engineer, published in a political journal, about climate change... I had low hopes, which were met. IMO few categories of professions have a harder time understanding cutting-edge science than engineers. That is because they think they know science because they use similar mathematical and technical tools, when in fact the professions do the exact opposite of one another. Engineers use what is known and understood to construct systems that can be validated. Scientists investigate unknown systems to try to construct knowledge and understanding. The worst thing, professionally, for an engineer is to not know or understand your work. The worst thing, for a scientist, is to spend too much time on things that are already known and understood. Engineers love to point out how scientists don't know, and can't prove, whether their climate models are accurate. Scientists know that that is the whole point of building such models. Working without knowing whether you're correct is not epistemological conflict. It's the fundamental condition of being a scientist. Yes, it would be better if we knew everything and only consulted systems that are provably correct. But we don't. And the only way to expand what we know is to spend a ton of time doing things that we don't know. We haven't found any other way to do it. Writing hand-wringing articles about the state of science from the sidelines does not advance human knowledge. You can't expand the map by standing inside the border and complaining about how hard it is to see past it.

pdonis 6y ago
> Working without knowing whether you're correct is not epistemological conflict. It's the fundamental condition of being a scientist. This is true, but it's also true that, when making public policy, we should be thinking like engineers, not scientists. Public policy is not about constructing new knowledge. It's about working out conflicts between different values and priorities based on our best current knowledge. To the extent the article is talking about public policy, I think it makes a valid point that the limitations of our knowledge in areas highly relevant to public policy are very often not recognized or taken into account when making public policy.
- snowwrestler 6y ago
  There are no sidelines in public policy. There's nowhere to wait to see how things turn out before making a decision. So, there are no neutral decisions. "Waiting until we're more sure about something" is not actually waiting, it is an affirmative decision to disregard what we think we know right now. That can be a valid choice, and something to debate, but it's fundamentally different from the feined neutrality of the concept of waiting to see. To make such an argument requires attacking that knowledge directly, showing specifically how it's wrong. Not just vaguely complaining that it's not good enough yet. This is actually well-understood in public policy when it comes to other areas like economics or defense; leaders act to address needs in real time, making the best decisions they can with the information available to them. "A good plan violently executed now is better than a perfect plan executed at some indefinite time in the future.” ― General Patton Engineering, as a discipline, cannot cover all of life. It largely constrains itself to situations that are understood, and can do so because there are other complementary disciplines that create the knowledge it uses. But in public policy there's not a complementary Earth or society that we can wait for.

Hnrobert42 6y ago

A little about the source: https://en.m.wikipedia.org/wiki/American_Affairs https://en.m.wikipedia.org/wiki/American_Affairs

Hnrobert42 6y ago
I got downvoted for linking to a document about the source. WTF?

guscost 6y ago

> Four conditions must be satisfied to have a valid scientific theory: (1) There is a mathematical model expressing the theory. (2) Precise relationships, known as “operational definitions,” are specified between terms in the theory and measurements of corresponding physical events. (3) There are validating data: there is a set of future quantitative predictions derived from the theory and measurements of corresponding physical events. (4) There is a statistical analysis that supports acceptance of the theory, that is, supports the concordance of the predictions with the physical measurements—including the mathematical theory justifying the application of the statistical methods. This definition appears rigorous at a glance, but it is deficient. We cannot properly test a theory if it only predicts things which we already expect to happen. Popper said that scientific theories must instead make "risky predictions": "Confirmations should count only if they are the result of risky predictions; that is to say, if, unenlightened by the theory in question, we should have expected an event which was incompatible with the theory–an event which would have refuted the theory."

cgiles 6y ago

I work in molecular biology research, and I think this is a great article that strikes at the heart of many problems in the field. I can't comment on the climate change stuff, although I wish he hadn't included it because it was almost certain to distract people from the overall point. The problem is that there are no remotely comprehensive, predictive, and mathematical models of what goes on inside of cells. It is pure empiricism: you run an intervention, and see what happens. Write it up in a paper. All well and good, except there are no viable models of what is happening inside that are predictive in the sense of being able to know what an intervention will do until you test it. We really need that if we want to develop treatments for molecular diseases that are more than marginally better. The Santa Fe Institute, systems biology people, and others were working hard on this problem at the turn of the century, but progress has stalled. It's too hard. We don't know how to do it. A new "mathematical epistemology" that could handle this problem would be a huge step forward, if it is possible. I can see why the author would extend this idea to things like economics or climate science. The thought in systems research was that, perhaps, different fields share similar underlying "complex systems" mechanisms, and if we can solve the problem in one area, we may have insights for how to do it elsewhere.

memexy 6y ago
What's missing from current mathematics to make predictive models for biology? I did a search for "neural network cell simulation" and got a few hits, e.g. https://ieeexplore.ieee.org/document/8805421 https://ieeexplore.ieee.org/document/8805421. So it seems that people are working on the problem of predictability (or at least augmenting the researcher's/experimenter's ability to do some analysis ahead of time based on simplified models).
- cgiles 6y ago
  > What's missing from current mathematics to make predictive models for biology? Well, I think that, no joke, there is a Nobel Prize waiting for anyone who knows the answer to that. I think this is the next big paradigm shift needed in biology, not to mention several other fields. Who is to say that the problem is strictly mathematical, though? It could be that the math exists, but no one knows how to fit existing data into it, or it could be that there is not enough data, or the right kind of data, to make such a model yet. It could be that both the data and algorithm exists, but we need to turn the Earth into computronium to run it. Who knows? > So it seems that people are working on the problem of predictability I'm sure they are. They have been for decades. The last time I did a systematic review of this area was before the resurgence of neural networks, so I can't really say what is the latest progress, or whether the progress in ANNs can inform this problem. I suspect it's very possible. The situation right now, as far as I know is that: A) most biologists don't even know this is a problem, and B) those who do, don't have any idea what the solution is, or if one even exists (note the author of the linked article was pessimistic on that point).

btrettel 6y ago

This article meanders too much. The basic point is this: > While stochastic models present us with numerous difficulties, an even more perplexing conundrum faces contemporary scientists and engineers who wish to model highly complex systems involving hundreds or thousands of variables and model parameters. Owing to their sheer number, many model parameters cannot be accurately estimated via experiment and are left uncertain. As a consequence of uncertainty, for each different set of possible values for the unknown parameters, there is a different model—possibly an infinite number of models. > Confronting the problems of complexity, validation, and model uncertainty, I have previously identified four options for moving ahead: (1) dispense with modeling complex systems that cannot be validated; (2) model complex systems and pretend they are validated; (3) model complex systems, admit that the models are not validated, use them pragmatically where possible, and be extremely cautious when interpreting them; (4) strive to develop a new and perhaps weaker scientific epistemology. At the moment I am in favor of option 3, though option 4 might be more appealing in the future. This isn't an easy option to take. Recently I had an article accepted for publication where I basically argued that no models, including my own, were truly validated because none fit a non-naive data set well. (In another paper I argued that most data sets used for validation are too easy to match because they don't cover the parameter space well.) A reviewer recommended rejection, basically saying that because the model isn't validated, it shouldn't be published. So much for being intellectually honest! The paper was eventually accepted after I made it more clear that none of the popular models work that well (some are absolutely terrible in my view), and that my model improves on the status quo in a few ways. Note that this situation isn't exactly the same as that described in the link. In the case of my article, I think we can get enough data to validate a model. We just don't have that data at present.

mjburgess 6y ago
I tend to come down on (1) in many cases, eg., social psychology. I'd be open to more of this sort of modelling if (4) was a realistic and completed project. The reality of (4) however is saying, to the public, things like: there may either be very limited climate change, or the entire world will be destroyed as we know it. I don't see the current public understanding of science, esp. via journalism, as fit for such realities.

Gatsky 6y ago

"Medicine is a form of engineering in which the physician takes some action (drug, surgery, etc.) to alter the behavior of a physical system, the patient. Its underlying science is biology. Hence our ability to characterize medical knowledge depends on our ability to represent and validate biological knowledge." This is really very wrong. The 'physical system' is conscious and suffering, which excludes Medicine from even the most vaporous definition of engineering. Real progress in medicine is much more akin to exploration (in the Christopher Columbus sense) than engineering. Anyway, this is a typical engineer writing about biology. They always make the same mistakes: 1. Ignoring evolution. 2. Treating mechanistic models as first class citizens with disdain for data (eventhough the latter in the form of DNA is the sine qua non of all life) 3. Failing to understand that "The simplest complete model of an organism is the organism itself."[1] as von Neumann noted. [1] https://onezero.medium.com/the-future-of-computing-is-analog-e758471fbfe1 https://onezero.medium.com/the-future-of-computing-is-analog...

Ma8ee 6y ago
> "The simplest complete model of an organism is the organism itself." But that is trivially true about everything, including golf balls. The interesting question isn't if the model is complete, but if it is sufficient (or complete enough) to make predictions. And in the case of golf balls we can make quite good predictions with a very simple model, mostly because we are only interested in if it ends up in the hole or not. But a complete model of a golf ball would need to include the exact dynamics of the different layers and the turbulence around it, which we just can't do. Luckily, we don't need to. So what is a sufficient model of a living organism? That of course also depends on what we want to know. I am quite certain that modern egg farmers have, for their purposes, quite sufficient models of hens. They know how fast the hens grow, the effects of different amounts and types of food, how much heat they produce, and of course how many and how big eggs they lay, etc.. And it is of course possible that we need to know the exact wave function for the complete system that is the cell to model it sufficiently enough to, say, synthesise new drugs. But I don't think so.
- Gatsky 6y ago
  It is definitely not trivially true about anything. A golf ball can be modelled easily. You are adding external complications to the system. An electron can be impossible to model if you inject it into a complex system. Anyway the point is a golf ball will not evolve into a sentient being who will then fly to mars and write poetry, which is what cells have done. This shows you can’t even put a meaningful bound on what you need to model for an organism.

kdkdk 6y ago

I think the system the author is looking for is probabilistic programming. It allows you to write programs/models with uncertain factors. Then at the end you can observe a few variables and it automatically computes the expectation and variance all the input variables. Check out psisolver.org for example.

ur-whale 6y ago

Feynmann: "It is whether or not the theory gives predictions that agree with experiment. It is not a question of whether a theory is philosophically delightful, or easy to understand, or perfectly reasonable from the point of view of common sense" Then proceeds to invent string theory. Don't get me wrong, I have a lot of admiration for Feynmann,but if he did indeed say the above, that strikes me as rather inconsistent : string theory is exactly what he preaches not to do : widely successful because of how beautiful it is, but unfalsifiable (as of yet).

N1H1L 6y ago

I read the full article. And it feels like much ado about nothing. One, the author for some reason is uncomfortable with a purely mathematical description of the world - without giving reasons beyond that humans cannot physically comprehend what the equations represent. Two, the whole piece disregards the advances in emergent phenomena, complex systems and statistics. Yes, we do not understand how individual electrons look like in copper, but statistical descriptions of copper dimensions, purity and grain size are enough to get an exceptionally accurate idea of the electrical behavior of a copper wire. This extends for exceptionally complex systems like lungs (smoking will significantly increase your cancer/emphysema risk), or planetary systems (increasing CO2 concentrations will increase surface temperatures). Three, There is also a weird bias against modeling. I am an experimental scientist, but I work closely with modellers, as all experimentalists do nowadays. Personally, I believe, that this bias against modeling often has a political component due to anthropogenic global warming. But in reality, there is a very vibrant dialogue between modellers and experimentalists, and hybrid scientists, people who are trained in both fields are the rage right now for hiring committees. Models have also improved dramatically - density functional theory is exceptionally accurate for smaller systems, while ReaxFF for dynamic systems and molecular dynamics simulations for more complex systems become better every year. The whole point about complex systems, applies to modelling too. I do not need to know the location of every molecule in the north Atlantic to predict the path of the hurricane - high pressure ridges and sea surface temperatures will give me really good results.

mensetmanusman 6y ago

This is a great article, but I wish it would have really dealt with the idea of complexity. Nature is not ‘unintelligible’ it is ‘complicated’ because of the large number of discrete interactive constitutive units. In fact, if you plot # of constitutive quarks/atoms being worked with versus uncertainty, you see an interesting layout of the sciences. e.g. take this plot by xkcd ( https://xkcd.com/435/ https://xkcd.com/435/ ) and think about how much matter is under observation (i.e. increasing complexity to the left). Somewhere on the left would be climate in this example (as in the article).