4 ms·

> Get it? It's just... insane. The rules of the universe are the rules for all things in it. They apply to everything, at all scales, at all times. This is now

> Get it? It's just... insane. The rules of the universe are the rules for all things in it. They apply to everything, at all scales, at all times. This is nowhere near as certain as you make it out to be. Take Conway's Life. With 4 simple "rules of the universe", you can create a series which to the best of our knowledge cannot be predicted from those rules. Clearly this series is a "thing" within the simulated universe, and clearly it emerges from the universal rules, but the rules don't give us any insight about it! The only known way to "predict" it is to let it run: in other words, it's irreducibly complex. Causality can look different at different scales. It's well established that simple low-level rules can generate extreme if not irreducible complexity. This doesn't mean the bottom-level rules don't apply everywhere; it just means their descriptive/predictive utility is not necessarily preserved when you zoom out. Is a prime-finding algorithm best described by the mechanism of the computational substrate? Any number of substrates could suffice. We can best predict its output by thinking at a higher level of abstraction.

krageon 6y ago
> you can create a series which to the best of our knowledge cannot be predicted from those rules You can absolutely reversibly compute conways game of life state, and you can compute it forward as well (after all, it is a game). That's prediction. > It's well established that simple low-level rules can generate [...] irreducible complexity. Simple rules can create a very complex emergent system, but that doesn't mean it cannot still be reduced to it's component rules. That's what makes them rules and not just guidelines.
- walleeee 6y ago
  > You can absolutely reversibly compute conways game of life state, and you can compute it forward as well (after all, it is a game). That's prediction. Notice I said irreducible, not irreversible. Yes, you can reverse the computation; but there's no known way to "shortcut" the forward process to predict the outcome of the series I mentioned any faster than simply letting it run. Letting it run is not prediction in the sense I mean here. By "predict" I mean foretell in advance what the system will do without needing to let it run. > Simple rules can create a very complex emergent system, but that doesn't mean it cannot still be reduced to it's component rules. That's what makes them rules and not just guidelines. I agree, with a minor caveat about language: you can describe a system which may display complex emergent behavior in terms of its underlying rules, but doing so is not guaranteed to give you useful information about (i.e., allow you to "predict", in the sense described above) the emergent behaviors. In contexts like these, to "reduce" typically means to describe in terms of a lower-level formalism while preserving predictive ability, AFAIK. In other words, the low-level formalism provides full information about the behavior of the entire system, such that you don't need to observe the system to know what it will do. In the Life example this is not the case.
  - krageon 6y ago
    When people do prediction, they also simulate system states given known priors and behavioral characteristics. The distinction you're making is in my opinion not valuable (or at least, you've not demonstrated it's value here).
    - walleeee 6y ago
      Sure, and the point of simulation in the first place is often to understand how the system in question will behave, in advance. The distinction I meant to make is precisely that some processes may not be computable in this way; that they cannot be simulated any faster than the real thing, in other words. That is what is meant by computational irreducibility, AFAIK. Determining which systems this is true of is very valuable, imo

krageon 6y ago

> you can create a series which to the best of our knowledge cannot be predicted from those rules You can absolutely reversibly compute conways game of life state, and you can compute it forward as well (after all, it is a game). That's prediction. > It's well established that simple low-level rules can generate [...] irreducible complexity. Simple rules can create a very complex emergent system, but that doesn't mean it cannot still be reduced to it's component rules. That's what makes them rules and not just guidelines.

walleeee 6y ago
> You can absolutely reversibly compute conways game of life state, and you can compute it forward as well (after all, it is a game). That's prediction. Notice I said irreducible, not irreversible. Yes, you can reverse the computation; but there's no known way to "shortcut" the forward process to predict the outcome of the series I mentioned any faster than simply letting it run. Letting it run is not prediction in the sense I mean here. By "predict" I mean foretell in advance what the system will do without needing to let it run. > Simple rules can create a very complex emergent system, but that doesn't mean it cannot still be reduced to it's component rules. That's what makes them rules and not just guidelines. I agree, with a minor caveat about language: you can describe a system which may display complex emergent behavior in terms of its underlying rules, but doing so is not guaranteed to give you useful information about (i.e., allow you to "predict", in the sense described above) the emergent behaviors. In contexts like these, to "reduce" typically means to describe in terms of a lower-level formalism while preserving predictive ability, AFAIK. In other words, the low-level formalism provides full information about the behavior of the entire system, such that you don't need to observe the system to know what it will do. In the Life example this is not the case.
- krageon 6y ago
  When people do prediction, they also simulate system states given known priors and behavioral characteristics. The distinction you're making is in my opinion not valuable (or at least, you've not demonstrated it's value here).
  - walleeee 6y ago
    Sure, and the point of simulation in the first place is often to understand how the system in question will behave, in advance. The distinction I meant to make is precisely that some processes may not be computable in this way; that they cannot be simulated any faster than the real thing, in other words. That is what is meant by computational irreducibility, AFAIK. Determining which systems this is true of is very valuable, imo