On Tue, Jul 30, 2019 at 8:19 AM James Cook <jc...@cs.berkeley.edu> wrote: > > On Tue, 30 Jul 2019 at 06:57, Aris Merchant > <thoughtsoflifeandligh...@gmail.com> wrote: > > First off, I’m very happy you published this. It’s very interesting and it > > adds a lot to the discourse. I will now proceed to state why I think it’s a > > bad idea. (Upon retreading your post, these amount to explanations of why > > concerns c and a are perhaps more serious than you think.) > > Thanks for writing this. I will respond to your points in more detail > below, but a summary: > > * You're right, this would make the rules more complicated. However, > it can be separated into a small change that changes the way we think > about timelines, but doesn't remove the references to "gamestate", and > a follow-up change that may get less support that would indeed make > the rules somewhat longer. > > * I disagree that this creates multiple timelines.
It does, but this is minor compared to the rest of it. > * I don't understand why you think this needs a high-power rule, and > propose that we test it by creating a mini-game with power-0.5 rules. I think it would, but this is also pretty minor; you'd only need a one or two paragraph rule. The problem is that, given that we haven't traditionally allowed timeline manipulation, you'd need something to tell high powered rules that it works now. > > You haven’t actually solved the gamestate problem, just moved it back a > > layer. The problem is that a report is fundamentally a declarative > > statement of how things are, and you’re changing that to make it > > imperative. This removes the generality of reports, since we have to define > > exactly what each report is changing. Switch reports are easy. What about > > asset reports? Do they “create, destroy, and transfer assets such that they > > are in the manner described”? At that point, you may as well be saying “set > > the assets so they are in the manner described”, which doesn’t seem like > > much of an improvement over “set the gamestate such that it is in the > > matter described”. I really fail to see how this is a particular > > simplification? It seems like you’ve made things more complicated, not less > > so. > > Yes, now that you mention it, my preferred way of implementing this > idea for reports about assets would be to explicitly say that the > assets are changed to be as reflected. Rule 2034 would also need to be > changed, and every time we add another kind of things to ratify, the > rules would get longer. > > So, I would like to separate this into two parts. > > 1. A smaller change to the way we deal with time: > > AI: 3.1 > Text: > In R1551, replace the first sentence of the second paragraph with: "If > a public document is ratified within 14 days after it is published, > then at the time it is published, it is Implemented. When a document > is Implemented, the gamestate is minimally modified to make the > ratified document as true and as accurate as possible." How does this help? Specifically, how is it any better than what exists now? > 2. Remove all references to "gamestate", and talk about actions rather > than documents being ratified. As you say, this would make the rules > somewhat longer. We could do 1 and not 2. > > > Also, you’re still doing timeline manipulation. Specifically, if I’m > > understanding you correctly, you’re causing a timeline divergence. There’s > > the timeline where the action is ratified in the future and the timeline > > where it isn’t ratified in the future. Worse, no one knows which timeline > > we’re on until 4 days after the ratifiable action happens. That means that > > any actions that depend on whether the ratifiable action succeeded or not > > have indeterminate success. This means that any CFJs about those options > > are indeterminate. At any given time, there will be no current summation of > > what the game is. Instead, it will be in a superposition of multiple > > possible states. This is, frankly, very hard to reason about. While you may > > technically be right that it isn’t strictly any more complicated than the > > current state of affairs, it would mean that there’s never a right answer > > about how the game is at the current moment. It removes our ability to have > > a shared idea of how things are now. > > Is this kind of uncertainty fundamentally different from any other? If > a player posts a hash of a pledge, then later reveals the text of the > pledge, there is similarly a period of time during which the Agoran > public does not know what the content of the pledge was. CFJs about > such a pledge have the same trouble in that unless the player is > motivated to reveal their pledge to the Judge, the content cannot be > known. It would be unconventional to say the game is in ally > "superposition of multiple states" during that period of uncertainty. Yes, it is. The difference is that current uncertainty is (or at least will be, when G.'s private contract thing comes about) limited to carefully controlled areas of the game state. We don't have uncertainty about what people CAN and CANNOT do, only about obligations. Obligations are prohibited from having any direct effect on the rest of the game; they only effect the blot system, and some have made strong arguments that they shouldn't even directly effect that. You'd be vastly expanding the range of things we might be uncertain about. -Aris
