3 ms·

I think you can make the case that, for example, going out, taking a stick from a forest, breaking it and record the sound will be much more economical than try

I think you can make the case that, for example, going out, taking a stick from a forest, breaking it and record the sound will be much more economical than trying to simulate a realistic stick breaking sound -- in particular in developer time trying to model sticks and wood breaking sounds in a convincing way (suppose we need say an art asset for a game). It's something complexity theory isn't well prepared to tackle, at least on first look. Complexity theory doesn't take into account the cost of devising an algorithm itself. And complexity theory usually only is interested in asymptotic behavior -- different in both ways from what we have (n=1 and algorithm cost). I think simulations win only when there are significant gains to be had over just doing it in real life, i.e. when you are only interested in a simple aspect that can be summarized by a model, or when for some reason the real life event uses a lot of energy or costs significantly. An extremely accurate simulation of wood and wood breaking, and sound, I think would be quite computationally costly -- the wood is already very complex (not as complex as the base atomic reality -- you probably don't need to model every single atom -- but still very complex). Sticks are cheap, and breaking a stick is easy. Breaking the stick doesn't seem to take much energy, and realistically stick breaking itself (although perhaps not stick collecting) comes out ahead energetically! It gets quite complicated quickly, and it really depends on what you want to model, but I think it's interesting to see when and if simulations are advantageous (for example, you could take the cost of raising a tree into account, or not! is the stick "free" because it would grow and decay anyway?). I think in the long term many things can be simplified into what we're interested, so models increasingly win. But realistically (in our limited lives and limited universes :)) I think there will always be a place for physical experimentation, taking pictures of sticks and recording the sound as they break. (and hopefully our relationship with trees and sticks itself still lasts for a long time...) I say that as someone who really enjoys modelling nature! :) I think modelling and trying to recreate natural process (sometimes with an artistic touch as well) is a great way to understand and appreciate the wonders of the universe. Note: Also, even with ideal reversible 0-cost computation, there's likely to be some kind of polynomial time overhead to 1:1 simulation of reality. This can be enough to justify experimentation instead of simulation, in some cases, when your model itself isn't a significant simplification over reality.

riwsky 3y ago
Complexity theory doesn’t handle it, but that’s why I didn’t use complexity theory. Economics handles it fine: no one’s ever paid you or anyone else to do a wood-breaking simulation at a sufficient physical level of detail to capture the sound accurately, so the comparative savings on it aren’t economically valuable.
- chaboud 3y ago
  Sound effect libraries have long included the sounds of wood breaking/cracking, generally recorded (rather than simulated). A couple of jobs ago, we sold sound effect libraries that cost 2-5x the cost of the software applications that they were used with (hundreds to thousands of dollars). I think you’d be hard pressed to find economists (beyond freshman college students who recently discovered Nietzsche second hand at a party) arguing for a purely direct monetary expression of value, but even they would have to admit that recordings of broken sticks in snow covered forests have non-zero value.
  - riwsky 3y ago
    You misunderstand me. It doesn’t matter how much the sound library is worth: if you weren’t using simulation to build it, then it’s irrelevant for the economic value of the simulation. Like, you can turn lead into gold if you run a nuclear reactor—and gold is valuable—but running this process is way more expensive than just buying gold, so it’s not valuable.
    - chaboud 3y ago
      That’s classic cost benefit analysis, and there are plenty of examples in audio of trade offs where the simulation is sufficient (or more flexibly applicable). Just because the cost is high for a given application doesn’t mean something has no value. Bringing the cost down is the classic engineering conundrum. Let’s take reverberation as an example: - Folks used to record in big rooms for big room reverb. It’s nice, nuanced, and comically impractical. Place mics around the room and capture the effect live, but even your close-mic’d take is going to have the room in it. - So some studios started piping audio out to a speaker in a secondary room and piping a mic from that secondary room back to the board. Boom. Configurable reverb, at your service. - But you can’t take it with you, so spring and plate reverb devices showed up. - Enter computers and DSPs, where simplified reverb models could be applied in real-time. - And then on to impulse capture and convolution (e.g., pop a balloon in a church, record, convolve). - And then beam tracing and other techniques approaching physical simulation (but orders of magnitude less computationally complex). And all but piping to a second room were used in making the libraries of sounds that we sold. However, throughout history, each of those techniques went through times then the cost benefit balance led to using some other technique(s) to achieve the desired result. That didn’t render those techniques worthless just because money hadn’t traded hands. It’s why economists describe markets as mechanisms for price discovery. What something had sold for is a way to understand value, but it’s not the sole definition of value (even monetary value)

riwsky 3y ago

Complexity theory doesn’t handle it, but that’s why I didn’t use complexity theory. Economics handles it fine: no one’s ever paid you or anyone else to do a wood-breaking simulation at a sufficient physical level of detail to capture the sound accurately, so the comparative savings on it aren’t economically valuable.

chaboud 3y ago
Sound effect libraries have long included the sounds of wood breaking/cracking, generally recorded (rather than simulated). A couple of jobs ago, we sold sound effect libraries that cost 2-5x the cost of the software applications that they were used with (hundreds to thousands of dollars). I think you’d be hard pressed to find economists (beyond freshman college students who recently discovered Nietzsche second hand at a party) arguing for a purely direct monetary expression of value, but even they would have to admit that recordings of broken sticks in snow covered forests have non-zero value.
- riwsky 3y ago
  You misunderstand me. It doesn’t matter how much the sound library is worth: if you weren’t using simulation to build it, then it’s irrelevant for the economic value of the simulation. Like, you can turn lead into gold if you run a nuclear reactor—and gold is valuable—but running this process is way more expensive than just buying gold, so it’s not valuable.
  - chaboud 3y ago
    That’s classic cost benefit analysis, and there are plenty of examples in audio of trade offs where the simulation is sufficient (or more flexibly applicable). Just because the cost is high for a given application doesn’t mean something has no value. Bringing the cost down is the classic engineering conundrum. Let’s take reverberation as an example: - Folks used to record in big rooms for big room reverb. It’s nice, nuanced, and comically impractical. Place mics around the room and capture the effect live, but even your close-mic’d take is going to have the room in it. - So some studios started piping audio out to a speaker in a secondary room and piping a mic from that secondary room back to the board. Boom. Configurable reverb, at your service. - But you can’t take it with you, so spring and plate reverb devices showed up. - Enter computers and DSPs, where simplified reverb models could be applied in real-time. - And then on to impulse capture and convolution (e.g., pop a balloon in a church, record, convolve). - And then beam tracing and other techniques approaching physical simulation (but orders of magnitude less computationally complex). And all but piping to a second room were used in making the libraries of sounds that we sold. However, throughout history, each of those techniques went through times then the cost benefit balance led to using some other technique(s) to achieve the desired result. That didn’t render those techniques worthless just because money hadn’t traded hands. It’s why economists describe markets as mechanisms for price discovery. What something had sold for is a way to understand value, but it’s not the sole definition of value (even monetary value)