Return to “Dev Logs”

Post

Re: [Josh] Monday, May 21, 2018

#16
I feel that there could be a difference between upgrade and greenfield build in that upgrades can be easier to reason about. Surrounding an existing asset, there may be current equilibrium but potential room to change the equilibrium in a favourable way. Your mining ship is profitably mining an ice field, to the point that you have spare cash. There is demand for more water, there is more potential supply in the ice field, so why not buy a new mining ship? Similarly, pirates are beginning to take a bite out of profits, so why not buy a new patrol vessel? While these investments may be expensive, there is an existing set of flows to hang reasoning around, the location is well known, and the capital spend is incremental on existing asset values..

For greenfield, one approach might be follow the upgrade process but loosen the location criteria - this might 'organically' come about if an AI has pressure from excess capital but no local investment flow opportunities. And another approach (fudge) might be an entrepreneurial character trait - some AIs are more ready to randomly invest in new locations, new technologies etc.

regards,
Charles
Post

Re: [Josh] Monday, May 21, 2018

#17
Thinking further, an excess of cash beyond expected day-to-day spend could create a pressure to invest.

In new territory, with big opportunities, this should (could?) lead to rapid growth. In mature territories, with many AIs, there may be an overall pressure to invest but the capital could be fragmented across too many actors to be able to make larger investments. Perhaps this could create the pressure to aggregate into larger looser factions to make the really huge investments?

Regards,
Charles
Post

Re: [Josh] Monday, May 21, 2018

#18
I really don't know much about economics, but had a thought about the AI based on something I read the other day. This could all be useless information/rambling, so... fair warning.

I wonder if you're at the point where you should push out the rest of the AI functions so that it can take those into account for making these value based decisions, as they will have some kind of weight in the choices made. If an AI has amazing military abilities, faction followers, or maybe with their scout ships and patrol ships, understands that there aren't really any/many pirates within a system (or maybe they own the system) or even neighboring ones; would those things not change how or when they decide to do things in the economy game? I wonder if the economy is important as a foundation, but to polish it, the entire rest of the game should be in place, it should almost be the last thing polished/worked on?

Continuing this AI related thought,

The creator of Rimworld, Tynan Sylvester posted something on twitter a couple days ago, and I don't know if it's a quote or something he came up with:

"When we talk about AI, "human level" intelligence is a meaningless term.

Human intelligence isn't a "level". It's just one particular combination of various quirky sub-capacities and reward functions.

It's not even the same for each person.
"

With that in mind, it got me thinking, what if the the AI had a numerical seed for its intelligence, where each value was how well it could predict, perform, or understand things in each category/field. Or maybe this is already how it works?

IE: An AI has a seed number that is 48176... based on where the numbers are in that order is the AI's intelligence. They could be Military, Economics, Science, Diplomacy, Industrial, and I'm sure there are more that could be added or I'm missing.

4 = Military
8 = Economics
1 = Science
7 = Diplomacy
6 = Industrial

Based on that order, this particular AI seed would very likely be a trader of some sort, with ties to production. Having sub-par military strength and not being knowledgeable in the sciences, it does have a good understanding of the economy, with stronger than average diplomatic abilities backing it which allows it to do well as a trader. Since it also has an average understanding of industrial production, it might also favor building production space factories once it has the funds, to further increase it's overall profits. If it's industrial intelligence was something closer to 3, it would most likely favor buying more trade ships than opening up industrial ventures for profit.

The numerical intelligence level from 0-9 would then change what formula it has access to with regards to it's particular field. This is hard to explain because I don't know any formula's to do this stuff, but just imagine that a level 1 in economics would have a simple math formula that would be:

2+2=4

Whereas a level 8 in economics would be more complicated like:

x³ - 12x² + 48x = 64
(I just googled "long math equation that equals 4" but I didn't bother double checking if this actually equals 4, so hopefully it does and I don't look like an idiot because it's way too late to math.)

Both of those equal 4 (see italics above), so both of the AI would come to the same conclusion, but one of the AI would have access to a lot more information in getting to the final conclusion, which could greatly impact what they do or how they handle whatever it is they are thinking about? If that formula is the simple question of buy at station A, sell at station B, make profit, the first AI will buy at station A and fly straight to station B. The second AI would have more knowledge of the area, and knows that if they buy at station A, but take a different route than straight to it, they can eliminate risk, possibly go faster because of a warp rail, acquire a second resource on the way to increase profits at a station directly beside station B, and hire an escort based on the risk vs reward. Or at some point the formula takes into account what other assets the AI has to assist more with the job, or how they should outfit their ship for the voyage. Maybe the first AI (intelligence level 2) ignored the newspaper that day (knowledge/news of the system could be part of the formula for a higher level AI intelligence) and didn't realize that there were pirate sightings directly between station A and station B. In the long run, the AI who "thinks" and plans out to be more efficient in regards to all the variables presented, because they have more information to base a decision on, would ultimately make more money and "grow" faster, wouldn't they?

Anyways, I hope I've explained my thought process well enough. I don't know if this helps at all or if I'm just spewing nonsense again, or if it even makes much sense to do it that way in the first place, but it seemed like an interesting idea. If everything I just typed means nothing than I hope my rambling will at least lead to some kind of eureka breakthrough thought process for Josh or Adam.

Thanks for reading.
Post

Re: [Josh] Monday, May 21, 2018

#19
Love these latest Joshposts. Theorycrafting, ho! :D

JoshParnell wrote:
Mon May 21, 2018 3:14 pm
You see, by taking our purchase of a new asset into the domain of the quantum -- that is, by making the smallest change that it is possible to make to the system, what we have actually done is converted the problem of reasoning about a new asset into the problem of reasoning about an existing one, making the assumption that it is effectively 'free' to purchase a single quantum (minimal discrete unit) of any given asset (this assumption is important and I will probe it further later).

Is this clever?

Yes. Yes, this is really, really clever. It's one of the practical lessons of systems design: if you want to build a big system that works, start with a small system that works, then see if it scales up. It might not... but starting with a big system is a great way to spend time making something that doesn't work, and can't ever be made to work.

JoshParnell wrote:
Mon May 21, 2018 3:14 pm
But what if the station is in a suboptimal location? Sure, maybe it was viable to put it at X, but what if having put it at Y would have made everyone's life even easier? I claim that it doesn't matter! Here's the beauty: sooner or later, another perturbation will come along, and if it's better than our station, we will slowly-but-surely lose business to it. Sooner or later, optimality will be evolved naturally through competition. Even if our station wasn't optimal, it was good enough to survive, and that meant that it provided value to the system. That's all that matters.

You're describing what's (often pejoratively) called "economic Darwinism," or more nicely "evolutionary economics": evolving an economy toward a state of perfection.

Like Darwinism, the evolution you describe has two parts: survival of the fittest through a selection process, and mutation -- which you're calling "perturbation" -- to introduce a local randomization that might lead to a higher overall maximum state.

In more practical terms, this sounds like generating random events: a sudden craze for tulips on Planet Zedcrom-4, or the appearance of a Space Dragon in Blipverts Sector. I'm curious whether this feels right to you when you prefer that everything come from somewhere... but then where do the perturbations you're considering come from?

One other interesting thing is that this could connect to the thinking you're doing on factions via some of the notions of evolutionary psychology: things a species learns to do because individuals discover a survival advantage in them and live to pass those algorithms to their descendants... but then the world changes and what was a survival advantage behavior is now vestigial superstition, or even actively anti-survival. What if NPCs could express such behaviors, and communicate them -- what does this do for factions in terms of tropisms toward various forms of capital expenditure decision-making? (Maybe nothing; I'm just thinking out loud again.)

JoshParnell wrote:
Mon May 21, 2018 3:14 pm
If a day comes when the system is so finely-tuned that the suboptimality of its location actually matters, a perturbation will come along and unseat it, eventually growing into the station that will replace it. Such is the nature of competitive evolution. A business that provides something fundamentally new has an easy time growing. It is only later, when adequate competition comes along, that the new market is pushed toward efficiency.

And the term of interest here is Pareto optimality: the moment when an economic system is so ordered that it's no longer possible for anyone to improve without someone else being made worse off.

(Pareto improvement may still be possible, though.)

JoshParnell wrote:
Mon May 21, 2018 3:42 pm
Cornflakes_91 wrote:
Mon May 21, 2018 3:22 pm
faction mechanics?
you thinking again how to solve faction personell without the worker/executive split?

That split no longer exists in such a distinct way; all AI players are 'executives' now (and ships' autopilots are the 'workers'). All AI players must make decisions about managing their assets, but those in positions of power will naturally have more assets to manage.

Excellent. That split always felt strongly artificial to me. Whatever the internal implementation benefit might have been, I just never saw value as a player.

My gut says this way of aligning decision-making and action is likely to be better for design, for implementation, and for player comprehension of interacting with NPCs. But my gut sometimes gets confused, so it and I look forward to more info here when/if appropriate.
Post

Re: [Josh] Monday, May 21, 2018

#20
JoshParnell wrote:
Mon May 21, 2018 3:14 pm
1. What does 'minimal discrete unit' look like for stations, ships, warp rails, etc? Can all assets be made granular?
2. What of the fact that it's not actually free to purchase things, no matter how small? Doesn't this necessitate thought in our 'random' investment, even for 'quantum'-sized ones?
3. Just because an investment is viable does not make it the best, or even close to the best, use of our money?
4. I don't want to see all of this clutter in the game world. I don't want to live in the same sandbox in which AI players are relentlessly experimenting with tiny, bad ideas.
I think the need for the perturbations to be 'quanta', and free, is quite restrictive. Maybe it is possible to make it work even for the large disruptive investments, your experimentation will tell.
If not, maybe the first 'quantum' of large investments could be simulated speculatively. If the expected benefit is not good enough, reject that perturbation. Over time, the AI periodically think about a few ideas of investments, and at some point they may find a good one and act on it.
It would solve 1. and 2., and greatly improve 4.. I believe 3. would not be changed, but as you said the growth or decay should naturally lead towards the optimal configuration.

Ultimately, the quality of the AI will depend on the heuristic function that suggests all the perturbations. Do you already have ideas of how to implement it?
Post

Re: [Josh] Monday, May 21, 2018

#21
Flatfingers wrote:
Tue May 22, 2018 2:14 am
JoshParnell wrote:
Mon May 21, 2018 3:42 pm
Cornflakes_91 wrote:
Mon May 21, 2018 3:22 pm
faction mechanics?
you thinking again how to solve faction personell without the worker/executive split?

That split no longer exists in such a distinct way; all AI players are 'executives' now (and ships' autopilots are the 'workers'). All AI players must make decisions about managing their assets, but those in positions of power will naturally have more assets to manage.

Excellent. That split always felt strongly artificial to me. Whatever the internal implementation benefit might have been, I just never saw value as a player.

My gut says this way of aligning decision-making and action is likely to be better for design, for implementation, and for player comprehension of interacting with NPCs. But my gut sometimes gets confused, so it and I look forward to more info here when/if appropriate.

it is artificial, yes.
but i do see some value in it to prevent mental and logistical burdens that dont provide interesting choices (as far as i can see).
its not interesting to have to think about every single fighter pilot defecting because e thought that going now is a good idea.
or having to bother with setting up work contracts for said fighter pilots

what is interesting is if an admiral decides to do eir own thing and takes eir part of the fleet with em.

its more interesting to manage a handful of high impact underlings than having to manage every. single. creep. in the universe.

personell management with such kind of granularity stops being fun (for the general gamer i suppose) when you have to manage hundreds of people.

so i'd personally do some hybrid solution.
with a limited amount of "smart" underlings (the "inner circle" or lieutnants or however you want to call it) per "proper" executive combined with a whole lot of dumb workers who dont have their own ambitions but to execute orders

(like i rambled at you in that PM Flat, if you remember)
Post

Re: [Josh] Monday, May 21, 2018

#22
Flatfingers to the rescue with links so that I know what my thoughts are actually called, ho! :D
Flatfingers wrote: You're describing what's (often pejoratively) called "economic Darwinism," or more nicely "evolutionary economics": evolving an economy toward a state of perfection.

Like Darwinism, the evolution you describe has two parts: survival of the fittest through a selection process, and mutation -- which you're calling "perturbation" -- to introduce a local randomization that might lead to a higher overall maximum state.

In more practical terms, this sounds like generating random events: a sudden craze for tulips on Planet Zedcrom-4, or the appearance of a Space Dragon in Blipverts Sector. I'm curious whether this feels right to you when you prefer that everything come from somewhere... but then where do the perturbations you're considering come from?

Ah yes, no economy is complete without tulip mania :lol: I do feel torn about random events. As you say they're contrary to the goal of everything being causal where possible. "Josh does not play dice" :P ([/sacrilege]) On the other hand, it could really be a source of flavor. My current plan for events is that we will have them only at the colony level. Since colonies are already our 'blackboxes,' the only major place in the game where conservation, causality, etc. can be fudged, I think it's fine for random events to occur on colonies. But as for something like "large pirate fleet discovered in the Bohr Hydrogen Cloud," this kind of thing must happen organically.

Flatfingers wrote:One other interesting thing is that this could connect to the thinking you're doing on factions via some of the notions of evolutionary psychology: things a species learns to do because individuals discover a survival advantage in them and live to pass those algorithms to their descendants... but then the world changes and what was a survival advantage behavior is now vestigial superstition, or even actively anti-survival. What if NPCs could express such behaviors, and communicate them -- what does this do for factions in terms of tropisms toward various forms of capital expenditure decision-making? (Maybe nothing; I'm just thinking out loud again.)

Although I don't envision full behaviors being able to evolve through generations of successful AI players in the sense of actually evolving algorithms, I do envision the AI personality traits as evolving in such a way as to tune the weights/branches of the AI algorithms. Ideally, traits will control AI behavior to such a degree that some personality vectors will be 'significantly more advantageous' than others (predicated on the environmental conditions, of course), and will therefore naturally survive with higher frequency. Ideally, at a large scale, I envision this causing certain environments to naturally favor, e.g., aggression, while other environments may favor intellectualism. To ensure that this plays out, it will be necessary to think carefully about the interaction between various generating algorithms and the AI psychology. In some sense it necessitates the careful crafting of 'biomes' in space :geek: If it works out, though, we could indeed see entire regions where factions are partial to, e.g. expenditure on research, perhaps due to highly-constrained resource availability driving the need for ever-more-efficient use of the existing resources, or perhaps entire regions of military inclination, etc.

Flatfingers wrote:And the term of interest here is Pareto optimality: the moment when an economic system is so ordered that it's no longer possible for anyone to improve without someone else being made worse off.

(Pareto improvement may still be possible, though.)

Nice, thanks.
Cornflakes_91 wrote:
Tue May 22, 2018 8:47 am
it is artificial, yes.
but i do see some value in it to prevent mental and logistical burdens that dont provide interesting choices (as far as i can see).
its not interesting to have to think about every single fighter pilot defecting because e thought that going now is a good idea.
or having to bother with setting up work contracts for said fighter pilots

what is interesting is if an admiral decides to do eir own thing and takes eir part of the fleet with em.

its more interesting to manage a handful of high impact underlings than having to manage every. single. creep. in the universe.

personell management with such kind of granularity stops being fun (for the general gamer i suppose) when you have to manage hundreds of people.

so i'd personally do some hybrid solution.
with a limited amount of "smart" underlings (the "inner circle" or lieutnants or however you want to call it) per "proper" executive combined with a whole lot of dumb workers who dont have their own ambitions but to execute orders

(like i rambled at you in that PM Flat, if you remember)

This is my thinking as well and it is (somewhat) how I have things structured. You do not need to manage 'every single fighter,' because you don't have to hire a pilot for every ship you buy. A ship can execute orders on autopilot without requiring high-level AI. It helps to think about ownership, the definition of 'player,' etc. as you would in an RTS, where each player has many units that don't require 'hiring' or any special consideration. They are simply dumb, order-following units. In that sense you could view them as the 'workers,' but really that's just lore -- the point is, your units follow your orders, period. In, for example, the command interface, sending an order to a unit simply broadcasts an action that the unit's 'autopilot' executes.

So, if you want to 'hire' another player, it is more like forming a team/alliance in an RTS. Theoretically you will get assistance from their assets, and have some loose control so that you can coordinate with them ('send some ships to escort me to my destination'), but the other player's assets are still under their full control, and it is open to them at any point in time to break the contract and 'take their part of the fleet with em' if they feel it's necessary. This is where personality comes in, and you should probably make sure you bring along a relatively brave friend if you anticipate the odds being stacked against you...

Hired 'managers' / 'commanders' is still an open problem.

---

Gotta get back to it, but perhaps will reply to others at a later date :wave:
“Whether you think you can, or you think you can't--you're right.” ~ Henry Ford
Post

Re: [Josh] Monday, May 21, 2018

#23
Perhaps I'm missing something, but I feel like this problem is being approached from the wrong direction. Unless I'm misunderstanding (which I may well be), Josh is trying to justify purchases pre-purchase - which makes sense, as it's something people do in real life all the time! However, people don't unify their approaches for that in real life, so can you really do it ingame as well?

Assuming I'm not missing anything:
Spoiler:      SHOW
For stations, justifying a new station with a method that takes changing dynamic systems into account is fairly simple: Run "fake ship" calculations based on all points of interest in the system:
1. Timer ticks off once every x minutes; the AI has all the requirements for building a station, so it can actually think about building one.
2. decide where to place the station. Choose a random spot that's near either a planet or an asteroid field.
3. start a "fake ship" in each asteroid field with the goal of depositing resources, and see if it would visit the station, or one that already exists, or a planet, etc. All market prices should be "averaged" - because, after all, this station will exist throughout all market price fluctuations, and you want to find the average.
4. start a "fake ship" at each planet with a goal of transporting resources, another for buying weapons, etc, and see where it goes. Use the same rules as above.
5. start "fake ships" at jump gates with various goals. Continue counting how many hypothetical visits our station gets.

...and so on. If the station "scores" fairly high - if ships are likely to visit this station over other available places, then the cost is justified. "Justified how?" Simple again. Say you have a station where there are (or were) 10 visits a day when it was brand-new. Our new hypothetical station gets 11 hypothetical daily visits. Therefore, if the stations are roughly the same cost, and the first station was worth building in the first place, the second station is also worth building. Each time you add a new station, you see if it scores higher than the highest-scoring station.

This entire calculation probably took so little CPU relative to everything else going on that you probably didn't notice. If we spread them out (which we absolutely can), we'll have even less FPS impact.


New colonies? You can do the same thing there as with stations. See how likely it is to be used.


Ships are a different beast entirely, though. You can't place a hypothetical ship and see whether other ships will visit it. :P Instead, build ships if you have no down time for any of your ships - if it's mining ships or transport ships. If it's fighters, you can check to see how often they're used. If it's anything larger - things meant for large battles - see how effective each was in the last few major combat events. If it was fully effective, buying another could be useful. If it didn't have much to do, or didn't deal much damage relative to its cost, buying another is probably a bad idea. If it did well but was taking too much damage, buy ships meant to protect it.


As to warp rails, currently they don't seem to actually have any monetary "benefit" - they only affect time taken to travel. The algorithm there is a little more complex.

1. Timer ticks down, and you have the ability to build warp rails.
2. Count all your current ships going between two major points in a system - say, you have 50 miners traveling back and forth between a single planet and an asteroid field. Now, tally up how much time is wasted on the entire trip between them. (if the trip is 15 minutes, 15 minutes x 50 ships = 750 minutes total)
3. Figure out how much total time would be taken up if there was a warp rail in place. (Let's say it increases ship speed by 10 times.) 1.5 minutes x 50 ships = 75 minutes total, saving you 675 accumulated minutes per trip.
4. How much time did the last warp rail in the system save for the faction that created it, when it was created? Figure out if the current warp rail will save more for you. If yes, you can continue.
5. See if there are any other opposing factions in the area that would be benefited by the new warp rail more than you would be. If there are, building that warp rail is probably a bad idea.

If everything is a go, you can build the warp rail.


Working out the balance:
"But Tal, how do I decide whether to build a warp rail, station, or ship?" Yeah, that's the core issue, right? There probably will be no clever solution there. What you'll have to do is add multipliers to each "type" of score (combat ship, economy ship, station, warp rail) and then manually balance them against each other with those multipliers, so that the outcome of them is "equal" - that is to say, that the game is able to nicely choose when to make what. You could even have those values affected a bit by AI personalities!

(the values in question being, 1. Hypothetical ship visits to a hypothetical station, 2. How much profit-per-economy-ship you're making, 3. Damage dealt per ship, and all the other mess 4. Minutes saved by hypothetical warp rail)


It's not an elegant solution, but it works, and it's not too far removed from how an actual player might go about it, either.

Josh-ninja'd!
Have a question? Send me a PM! || People talking in IRC over the past two hours: Image
Image
Image
Post

Re: [Josh] Monday, May 21, 2018

#24
Thanks, Josh! Appreciate the additional insights.

And Cornflakes, your thoughts on the exec/worker design are also well-taken, as well as Josh's more recent way of framing it. It's just that... TBH, it sounds kinda boring.

I get the point about control. I understand very well that for many players of a game with a strong "accumulation of wealth tokens" component, the game would feel broken if NPCs can't be counted on to do what they're told to do. (I've written about this expectation before, which I think is personality-driven, such as in this article from 2006 that -- in the section on crafting -- observed that players of economic dominance games really, really don't like random failures in mass production systems.)

But these gamers, while their preferred kind of fun is worth supporting, aren't the only kinds of folks who may enjoy Limit Theory. What about the Explorers? For these players, autonomy in NPCs isn't a bug, it's a feature.

Let's say you're in a dogfight with Bob The NPC. In "worker" mode, Bob fights to the death (preferably his). So do Sue, and Pat, and every other NPC. They're cannon fodder, just following orders, the same way every time.

But now suppose Bob has more autonomy to assess a situation, and to express behaviors within the context of each particular situation that reflect Bob's unique vector of personality traits. What if most NPCs are middle-of-the-road and will just follow orders, but some can surprise you? Maybe Bob flees after taking a little damage. Maybe Pat sees you and closes aggressively on you as quickly as possible, all guns blazing.

Am I alone in thinking this kind of universe feels vastly more alive and interesting than one in which all pilots always just do as they're told?

For those concerned this would feel too random, I give you the original Thief: The Dark Project by Looking Glass. One of their insights was to give guards multiple states of awareness of your presence, and -- vitally -- to communicate these states to the player:

  • Glimpsing you: "Hmm... did I see something?"
  • Seeing you: "Hey, stop right there, you taffer!"
  • Chasing you: "I know you're around here somewhere!"
  • Losing you: "Bah. Got away. Back to my patrol."

Now imagine that NPC pilots in LT can similarly communicate their states. The cowardly Bob can squeak, "Aaaughh! To hell with orders, I'm outta here!" The aggressive Pat can howl, "My name is Pat Montoya -- you killed my father, prepare to die!" and so on.

The point is that when actors communicate their state, we wrap those communications in intentionality. Instead of perceiving actions as random, we see feelings and reasoning that seem plausible, like things that real people would do.

In LT, I believe this would mean much higher tolerance by players for NPCs not all being mindless order-following drones. Not only does more NPC autonomy breathe more life into the game world, it might even be a source of useful perturbations -- isn't this exactly what a high rating in the Lawless personality trait ought to produce?

This is the kind of stuff in my head that makes me prefer an analog range of possibilities over a binary exec/worker distinction. Every NPC has some level of autonomy; what makes one a worker is simply one's current role. One who's Lawful, and survives, can be promoted to higher levels of responsibility. NPCs who also score high in Creativity can be promoted even higher (or create their own factions) to give exec-level direction to subordinates. Meanwhile, the Lawless NPCs will eventually bail on an organization after going rogue one too many times, because the universe needs explorers (Creative/Lawless) and pirates (Consuming/Lawless), too.

So with this kind of framework in mind, and setting aside for a moment the question of implementation/testing time -- why would not imposing a hard worker/exec split make Limit Theory less fun, rather than more fun as I believe it would?
Post

Re: [Josh] Monday, May 21, 2018

#25
Talv, I absolutely walked down that path during my thinking. I have many pages on such hypothetical 'parallel universe' simulations. Certainly that is a higher-quality approach, and clever as well since we can defer some of the cost across the entire standard game simulation. However, after pretty extensive consideration of it, I am of the opinion that it exacts either too high of a performance price, too high of a complexity price, or yields too little of an accuracy gain to be worthwhile (there is a trade-off to be had among these, depending on how one implements the machinery of hypothetical state). I could go into detail on these points, but it is a lengthy discussion and for now I need to push toward testing my methods. I just wanted to mention that I have indeed evaluated the general technique you outlined!

In fact, part of my development toward the mechanism I outlined was due to considering hypothetical simulation, and recognizing that a 'very small version' of a thing carries many of the same benefits of a hypothetical thing.

Anyway, we will see how things pan out soon :geek:

(I see Flatfingers has now posted some more material that I must look through..)
“Whether you think you can, or you think you can't--you're right.” ~ Henry Ford
Post

Re: [Josh] Monday, May 21, 2018

#26
Talvieno wrote:
Tue May 22, 2018 12:17 pm
Perhaps I'm missing something, but I feel like this problem is being approached from the wrong direction. Unless I'm misunderstanding (which I may well be), Josh is trying to justify purchases pre-purchase - which makes sense, as it's something people do in real life all the time! However, people don't unify their approaches for that in real life, so can you really do it ingame as well?
People don't sure.
But we aren't dealing with people. We are dealing with businesses.
You can expect a much higher percentage of their expenditure to be considered carefully, and then applied incorrectly or non-optimally.

Individuals will tend to buy things that make them happy or make their life easier.

Businesses buy whatever the current buzzword is, at the lowest price they can.

I propose dealing with the AI at three different scales.

A) Individual Scale. This AI is one guy, maybe as many as 5 ships that follow him, and spends mostly to improve what he has, or replace things with better versions. He doesn't care about building stations, or system Infrastructure, he only cares about himself.
These guys do only what is most expedient at the moment. And will happily split into two AIs when they get particularly rich.

B) Small Group Scale. This AI controls a dozen to a hundred ships. He is focused either on finding and developing a single system, or exploiting in a few systems maintaining a single base, and will often spend to convert a good place into a better place. But will happily move on from an area if opportunity lies elsewhere, or he is forced out.
These guys tend to own a single station or pair of stations, and will often bulk produce a small quantity of goods.
These guys split only when their system is fully exploited sending out a new small AI with a dozen ships and some cash to continue the expansion cycle.

C) Large Group Scale. This AI spaces 100+ ships. He tends to focus on controlling space. He will grab systems and attempt to brutally control everything in them. Letting systems go only when they become particularly bad. He is barely a step down from an Empire, just lacks the scale to be one. He is quite interested in policing and taxing any activity in his area. These guys are the factions that vie for actual control of the universe, and will tend to control a vast number of assets, but only few will be upgraded to a high enough level to be useful for high end players.
These guys only split once their number of systems gets too large. Splitting in half and forming an alliance with their new split. Eventually these alliances will degrade and new empires will properly form.
°˖◝(ಠ‸ಠ)◜˖°
Toba - A Development Dump
Post

Re: [Josh] Monday, May 21, 2018

#27
Awesome Work. Truly.

The flow economy is going to be like nothing else we've seen in a game.
The ability for first-person dogfighting or RTS control will be truly fascinating.

However, and I may be mistaken, but for the year I have been following the development of this amazing game I have not come across a discussion about the macro side of gameplay.
And since Josh is starting work on factions I believe it might be time to discuss such things. Macro gameplay is a very, very critical part of this game concept. The flow economy and the dynamic ability for the A.I. to develop means that the environment the team is developing will be ripe for player intervention.

With procedural generation giving infinite possibilities and playgrounds to conquer, we are going to need the tools and mechanics to be able to manage massive operations.

So these are my questions for the devs. How are you considering Macro mechanics? Because the game will become limited and un-enjoyable if we must micro-manage the acquisition, production, assignment, armament, maintenance, etc, of every..single...asset. Which from what has been demonstrated, will be many, many, assets.

When or if the player chooses to create a faction/corp/etc will there be an AI similar to what runs all the other factions created for the player? I know that some sort of AI will have to be involved in order to run things. But I'm curious as to how much of the tasks it will run.

In my mind I think the AI will have to accomplish the same tasks as any of the other Factions, however, it will have to do so with player guidance. Which are where my questions come from. How much will the AI do? Will the player be able to dictate what it does? Give it direction? Decide on how aggressive it is or how passive?

The AI has four major areas right? Economy, Military, Expansion, Industry. The AI must be able to manage and adapt to the economy so that it makes money. It does this by investing in assets. How will the player manage such things? Is it something where the player will tell the AI to build X number Ship class Y with Z percent or amount of GDP or profit? Whereas the AI will take that percentage or amount of its GPD or profit from last "cycle" or "unit of time measurement" and use it to buy the actual ships or raw materials to use in factories to make the ships? What about infrastructure? Will the management AI be able to autonomously decide where to put stations using the method you described in this dev log and/or via player request? Will it have to use guidelines for maximum investment like I said above i.e. use X percent of profit investing in infrastructure (Stations, Factories, Warp rails if we can even build them in the first place, etc)? How about military investments, can you tell the AI to spend a certain amount of money on ships and defenses. i.e. build X number of Battleship class Y with Z percent or amount of GDP or Profit for X task or fleet (Patrol, Fleet Y, Fleet Z, Escort, AI choice, etc). Will our management AI be able to give suggestions based off of trends it's noticing like a rise in ore prices so it recommends we build more miners to capitalize on that? Or if we are taking heavy losses in a conflict and arent reinforcing enough so it suggests we invest more in military production. And lastly, if a player doesn't want to worry about any of that stuff can they simply tell the AI to do what it wants and have it make all Related decisions without any of our intervention.

The AI will need to be able to defend its self and expand. How will it handle military? Will we have to manually order every single military ship or structure or can we have the AI build and form our fleets and military might on its own or with guidelines sort of like what I listed above? If we let it, can it dynamically decide where to assign military assets based on trends or perceived threats or will the player have to manually order every single fleet or individual ship in our empire to a task or fleet? If we let it can our management AI expand through conquest? Will we have to order every ship into combat and decide every RTS maneuver or can we give it a goal like invade X system with Y aggressiveness where aggressiveness determines the amount of military commitment and behavior of ships and strategy in situ? Will a management AI even be able to handle the logistics of a full-scale invasion with thousands of ships fighting across multiple systems? And once again, If the player isn't interested in such things can they simply tell their management AI how aggressive or expansionist to be and let it do its thing without any player intervention at all.

How will a management AI decide to expand and where to put things? Will the player have to assign it to place every single station or tell it to guard an asteroid field or can it, if the player allows it, to use whatever mechanic the other AI's use to determine where to put things? Or can it be a mixture of both?

Lastly Industry. Say the management AI places a factory. Can it automatically assign traders or transports to go to the warehouses to pick up the materials necessary or go to the market and buy the materials then transport it back to the factory? Or will the player have to make the trade itinerary for every...single...logistic ship (trader, Miner, Transport, etc). Can the AI decide how to distribute the products produced by the factory? Like, say the factory was made to produce weapon X for construction on Fighter Y. Will the AI be able to manage the logistics necessary to ferry the weapons from the weapons factory to the ship factory? Or will the player have to assign and coordinate every single ship, factory, and distribution channel of all factory or trade/transport systems? Will the AI be able to make the factory chains totally on its own if instructed or allowed to by the player?

My last question about the Macro mechanics is if the player wants, can they just let the AI manage literally everything of the player made faction and just play in the aftermath of telling the AI to destroy the universe at all costs?
----
Miscellaneous questions

How will procedural generation be controlled? Will you tell it to generate X number of systems on start? or will it actively generate the frontier as the player and NPCs explore and expand? If so can we tell it to stop generating if we feel we have reached a certain size we want?

Are Factions dynamic? Meaning can they come into existence mid-game. Like if an NPC builds enough wealth can it found a new faction. Or will the number of factions be set in the beginning and only be destroyed until one is left?

Will diplomacy be a consideration in gameplay? like when two weaker factions form an alliance to face a big faction. Will those factions be able to split if one wants the others resources after the war?

And this one is just pure curiosity, but will machine learning be moddable into the AI?

And I know that you guys may be theorizing on a lot of the Macro gameplay and not know what you are going to do just yet. I just want you to know that Macro tools and mechanics will be absolutely critical parts of gameplay. With procedural generation meaning we can play on massive scales and, with a good engine meaning large amounts of stuff; we are going to need extensive and deep Macro tools to be able to oversee it all while still having the game be playable and enjoyable.

Love reading the dev-logs by the way. It always excites me to see a new one posted.
Post

Re: [Josh] Monday, May 21, 2018

#29
Flatfingers wrote:
Tue May 22, 2018 2:34 pm
Now imagine that NPC pilots in LT can similarly communicate their states. The cowardly Bob can squeak, "Aaaughh! To hell with orders, I'm outta here!" The aggressive Pat can howl, "My name is Pat Montoya -- you killed my father, prepare to die!" and so on.

The point is that when actors communicate their state, we wrap those communications in intentionality. Instead of perceiving actions as random, we see feelings and reasoning that seem plausible, like things that real people would do.
Them communicating some intentionality doesnt make it more fun or interesting (in ways that arent shades of disaster recovery) when i have to handle a hundred dudes deciding at the same time that their mission is suicide and break off all on their own.

Its nothing one can interact withwout a massive amount of micromanagement.

A statistics based approach, where morale (or whatever else) is tracked for the group and the individuals are just dice rolls inside that statistic.
You dont need individually intelligent actors for that
Flatfingers wrote: In LT, I believe this would mean much higher tolerance by players for NPCs not all being mindless order-following drones. Not only does more NPC autonomy breathe more life into the game world, it might even be a source of useful perturbations -- isn't this exactly what a high rating in the Lawless personality trait ought to produce?
If you manage to include flexibility without making an empire of any size a literal nightmare to control, you are welcome to do so.

But for now all i've seen was "lets make them all as smart as possible but provide no reason for them to listen to you when you need it"

How is my military supposed to work if i get deserters every time my forces arent sure if they can win?
Because thats whats going to happen if they have intelligence and any kind of self preservation and no overriding cause to act opposing to it.

Or rather, a system thats functionally different from all dumb + statistics rolls to tell what they are allowed to do without explicit input.
Post

Re: [Josh] Monday, May 21, 2018

#30
JoshParnell wrote:
Tue May 22, 2018 12:10 pm

Although I don't envision full behaviors being able to evolve through generations of successful AI players in the sense of actually evolving algorithms, I do envision the AI personality traits as evolving in such a way as to tune the weights/branches of the AI algorithms. Ideally, traits will control AI behavior to such a degree that some personality vectors will be 'significantly more advantageous' than others (predicated on the environmental conditions, of course), and will therefore naturally survive with higher frequency. Ideally, at a large scale, I envision this causing certain environments to naturally favor, e.g., aggression, while other environments may favor intellectualism. To ensure that this plays out, it will be necessary to think carefully about the interaction between various generating algorithms and the AI psychology. In some sense it necessitates the careful crafting of 'biomes' in space :geek: If it works out, though, we could indeed see entire regions where factions are partial to, e.g. expenditure on research, perhaps due to highly-constrained resource availability driving the need for ever-more-efficient use of the existing resources, or perhaps entire regions of military inclination, etc.
I dusted off my login to specifically ask about this, and then saw your response. :) I was a bit worried that all AIs would end up becoming the same, as they all optimized themselves to "perfection", but adding personality traits, and having them be influenced by external forces, sounds like a great way to have this NOT happen.

Have you given any thought to how AIs might influence OTHER AIs? Like maybe, at the border between two different AI "factions", you could have genetic algorithms spin off perturbations, and see what survives? It would be fascinating to see an AI with a domination streak meet up with an AI with a tend toward tech research spin off a nightmare "Borg" AI out to take over the galaxy...

Even though I don't post much, I love reading your updates, always have. Keep up the great work!

Online Now

Users browsing this forum: Zebulon Pi and 1 guest

cron