5 ms·

Ultimately all descriptions of physical reality are incorrect.

analog31 7y ago
"All models are wrong, but some are useful." -- George Box If we think our models are right, we will push them until we find out where they are wrong. That's what we do, over and over again. Pushing classical physics to its breaking point spawned quantum mechanics. We know there's a breaking point for present day physics too.
- visarga 7y ago
  In machine learning we have many more models, all being imperfect. The ease of trying out a new model makes them cheap. Hell, we even do AutoML and search the space of billions of possible models. They are all approximations, we don't 'understand' reality. We can only do useful things with them, but there's a limit to what the models can teach us about the metaphysics of the world.
  - MiroF 7y ago
    No offense, but the word "model" is used very loosely in ML to describe what is essentially a black box function that can be computer on a computer. Models in Physics involve experimentation, validation, interpretability
egdod 7y ago
There is a description that’s correct. We may never find it, but it exists.
- lidHanteyk 7y ago
  Do you have evidence in favor of the idea that, say, there exists a finitely-axiomatized formal system which describes reality? We've been searching for a while, and evidence so far suggests that it's not going to be that easy. Tarski's Undefinability is a non-trivial barrier. With mathematics, we can describe hypotheses of reality, but there may be details and nuances to reality which we're simply not able to observe and thus not able to empirically consider.
  - egdod 7y ago
    > finitely axiomatized formal system I said “description.” You’re adding additional constraints that may or may not be warranted.
    - lidHanteyk 7y ago
      I'm merely helping you catch up on the past century of improvements in logic. What would your description consist of, if not a finite listing of axioms? I bet that you'll complain if my description includes a susbystem for arithmetic, in accordance with Gödel's Incompleteness.
      - egdod 7y ago
        Why would a description of physics necessarily include a formal system or arithmetic?
        lidHanteyk 7y ago
        It necessarily includes a formal system. That much has been hammered into the ground already, and you'll discover it for yourself as soon as you attempt to formalize your description. All existing attempts to describe physics have encountered some sort of incompleteness. Newtonian physics yields ordinary differential equations, which can encode Turing-complete problems. [0][1] Quantum mechanics yields Hamiltonian operators, which can encode Turing-complete problems too. [2][3] Both of these paradigms predate Turing's work; physics was Turing-complete a long time before you came along to look at the problem. I can empathize with your original point: Surely reality exists! It seems so real! But reality can be real without existing. Ultimately, our models are only hypotheses about reality, and what they have shown us here is that even our models are so complex as to admit problems that are not going to be solved by mere assertion of existence. [0] https://link.springer.com/chapter/10.1007%2F11494645_21 https://link.springer.com/chapter/10.1007%2F11494645_21 [1] https://sapientia.ualg.pt/bitstream/10400.1/1008/1/05-GCB-stable.pdf https://sapientia.ualg.pt/bitstream/10400.1/1008/1/05-GCB-st... [2] https://www.nature.com/articles/nature16059 https://www.nature.com/articles/nature16059 [3] https://eprints.ucm.es/38062/1/spectral-gap_supplementary.pdf https://eprints.ucm.es/38062/1/spectral-gap_supplementary.pd...

"All models are wrong, but some are useful." -- George Box If we think our models are right, we will push them until we find out where they are wrong. That's what we do, over and over again. Pushing classical physics to its breaking point spawned quantum mechanics. We know there's a breaking point for present day physics too.

visarga 7y ago
In machine learning we have many more models, all being imperfect. The ease of trying out a new model makes them cheap. Hell, we even do AutoML and search the space of billions of possible models. They are all approximations, we don't 'understand' reality. We can only do useful things with them, but there's a limit to what the models can teach us about the metaphysics of the world.
- MiroF 7y ago
  No offense, but the word "model" is used very loosely in ML to describe what is essentially a black box function that can be computer on a computer. Models in Physics involve experimentation, validation, interpretability

egdod 7y ago

There is a description that’s correct. We may never find it, but it exists.

lidHanteyk 7y ago
Do you have evidence in favor of the idea that, say, there exists a finitely-axiomatized formal system which describes reality? We've been searching for a while, and evidence so far suggests that it's not going to be that easy. Tarski's Undefinability is a non-trivial barrier. With mathematics, we can describe hypotheses of reality, but there may be details and nuances to reality which we're simply not able to observe and thus not able to empirically consider.
- egdod 7y ago
  > finitely axiomatized formal system I said “description.” You’re adding additional constraints that may or may not be warranted.
  - lidHanteyk 7y ago
    I'm merely helping you catch up on the past century of improvements in logic. What would your description consist of, if not a finite listing of axioms? I bet that you'll complain if my description includes a susbystem for arithmetic, in accordance with Gödel's Incompleteness.
    - egdod 7y ago
      Why would a description of physics necessarily include a formal system or arithmetic?
      - lidHanteyk 7y ago
        It necessarily includes a formal system. That much has been hammered into the ground already, and you'll discover it for yourself as soon as you attempt to formalize your description. All existing attempts to describe physics have encountered some sort of incompleteness. Newtonian physics yields ordinary differential equations, which can encode Turing-complete problems. [0][1] Quantum mechanics yields Hamiltonian operators, which can encode Turing-complete problems too. [2][3] Both of these paradigms predate Turing's work; physics was Turing-complete a long time before you came along to look at the problem. I can empathize with your original point: Surely reality exists! It seems so real! But reality can be real without existing. Ultimately, our models are only hypotheses about reality, and what they have shown us here is that even our models are so complex as to admit problems that are not going to be solved by mere assertion of existence. [0] https://link.springer.com/chapter/10.1007%2F11494645_21 https://link.springer.com/chapter/10.1007%2F11494645_21 [1] https://sapientia.ualg.pt/bitstream/10400.1/1008/1/05-GCB-stable.pdf https://sapientia.ualg.pt/bitstream/10400.1/1008/1/05-GCB-st... [2] https://www.nature.com/articles/nature16059 https://www.nature.com/articles/nature16059 [3] https://eprints.ucm.es/38062/1/spectral-gap_supplementary.pdf https://eprints.ucm.es/38062/1/spectral-gap_supplementary.pd...