Alright, now that I've had some time to read this thoroughly and digest this Pre-Dark Days style devlog, I predictably have some commentary :p
First off, switching to rates rather than absolute quantities sounds incredibly smart (There's my wonderboy
).
JoshParnell wrote:The point, though, is that the AI has taken the specifics of the simulation and figured out how to craft optimal behavior with them. Nice.
Ummf! That's downright arousing. It makes the
God-Emperor Test sound quite possible
JoshParnell wrote:It's actually interesting to note that the AI is not applying 'rational capitalist' behavior here, but rather 'rational collectivist' behavior; some units are performing intentionally-suboptimal work in order that the whole can be optimal. Philosophical arguments aside, this works out well for our purposes of simulating a predominantly-AI-driven economy
How many AI are actually in these models? If everything belongs to the same AI, then this seems like a fairly effective form of Capitalism if you had multiple AI in competition with one another. Being rationally collectivist makes perfect sense for the internal decisions of a faction or company. As for units intentionally performing sub-optimally, that seems to be more a limitation of the model than anything else. Presumably these sub-optimal units don't have the opportunity to take on other jobs besides mine and trade, they cant go be security, pirates, scouts, etc. If those options were available, the units wouldn't be assigned to sub-optimal trading missions, but be assigned to security detail (assuming it's reasonably affordable to change out that ship's equipment).
JoshParnell wrote:Finally, in a 10,000-ship simulation, the economy is completely over-saturated: ... Economic volatility is gone, equilibrium is here, and the AI is generally much more capable of setting up well-structured economies that take into account all of the nuances of the star system and game constants.
So what you're saying is that a system's economy dynamically stabilizes as its population rises? Is the converse also true, that in perhaps a newly discovered and colonized system on the frontier we can expect significant economic volatility? So the frontiers really be the Wild West?! *Gets cowboy hat*
JoshParnell wrote:if I spend X, Y will happen, and I estimate that this will let me extract Z additional pressure from the total economy." Z is the difficult part: the hard problem of large, one-time expenditures is trying to work out what the net result will be in a complicated economy
With multiple AI competing, each with limited information, would it not make more sense for the AI to predict what the result will be for itself, and perhaps its friends and enemies? Does it really need to know the net effect on the whole economy?
Silverware wrote:Capital ship parts might not be used per day, but per year they would. So an AI could look at the average over a yearly period and dedicate a small amount of production off for it. But only look to keep a small stockpile since it'll likely last a year.
When a war starts and Capital production is pumped up, that same AI can now drop even more investment into the manufacture to cover it, and when the war ends, they can shift it away and still keep just a tiny trickle of production afterwards.
This splitting of supply/demand intervals makes a lot of sense to me. If I were to place an order for 20 Battleships over the next year, the supply/demand rate-per-day would be much lower than if I wanted the same 20 battleships within 1 month. As the saying goes, "You can have it good, fast, or cheap, pick 2" Now suppose it's been 1 month, and not a single ship has been built. The Supply/Demand rate-per-day will naturally increase. If conversely 5 ships were built within the first month, then it would sharply decrease. So long as a production order is given a countdown timer, Should it not naturally draw more attention as the deadline draws closer in relation to the order's completion?
JoshParnell wrote:Yes. I am working out the theory for a scouting job type; it is a planned AI behavior. I have not worked out exactly how it fits into the flow system, but this is one of those "I've given a lot of thought to it this week" bits. Scouting is a lot like patrolling. I think most economic activities will contribute to a 'demand' for things like information discovery and security.
For some reason I get the feeling that scouting is less like patrolling and more like foraging. An explorer selects some semi-random positions in a system to head towards with it's long range scanners on, if it doesn't detect anything unknown, it picks another point; if it picks up an unknown signal on the scanner, it goes to investigate (at least if the signal is of the type the scout is looking for). In terms of fitting it into the flow system, why not give an explorer an expeditionary budget and make time an economic sink, every moment you're exploring, you're losing money from other things you could be doing. However if a system is relatively unexplored, it considers exploration a high risk, high reward investment. There is after all no guarantee that an explorer will discover anything that isn't already known and the expedition will have been for naught, perhaps even bankrupting the explorer, so after a certain point, an explorer who hasn't found anything of note will cut their losses and return to do something else or explore elsewhere.
This brings up the question about AI Theory of Mind and the value of any piece of information. How does an explorer know they have found something few if any others know about as opposed to something that's new to them, but is common knowledge and isn't really a valuable discovery. Perhaps setting up a sort of News Broadcaster would be appropriate, in that if the piece of information is sold to the broadcaster, it is then transmitted to every subscriber to that broadcast instantaneously (provided they aren't too far away). Being the first to get a scoop to the broadcaster results in a handsome payoff, but once it's out there, the value of that information drops drastically (in that area at least, it could still be profitable to then carry that information to another broadcaster far away, conveniently allowing for a natural dissemination of information).
Of course information could not be given to the broadcaster, and be kept relatively secret. However this makes leaking information potentially quite profitable for the leaker (and will likely really piss off the previous owners). Could this make infiltration, espionage, and even undercover reporting viable too?
Challenging your assumptions is good for your health, good for your business, and good for your future. Stay skeptical but never undervalue the importance of a new and unfamiliar perspective.
Imagination Fertilizer
Beauty may not save the world, but it's the only thing that can