Agree with 1 and 3. Is there no way to tweak their fleet composition by weight? Or maybe put in fewer invasion templates for their ship design?



With regards to 2, here I'm going to show my ignorance. I thought being sieged is a starsystem status. Is it possible to change their build priority when their system enter siege status? (Would only make the system being sieged build ships, but better nothing I guess?)

You can pretty much play them off on one another. If any of them declare war on you, ally with a neighbor and stomp them.

Commander wrote:
I've sort of come to the conclusion that, if the system isn't full on population, the AI dedicates everything to Food, if the system isn't fully developed in terms of system improvements, everything is dedicated to Industry. If both of those are satisfied, everything is dedicated to Dust.

Sorry, but your theory doesn't quite meet the reality here.



The exact exploitation-picking behaviour is described in PlanetInfrastructure.xml:



Food: You got this one right. The AI will focus on food as long as there's still population to grow or when the systems has an A.I.S.

Industry: If the AI is Sower it will build this everywhere, if not, then it will build industry on the planets that give a bonus to industry-exploit like Lava, Tundra and Jungle.

Dust: If the AI is in financial-troubles it will use this everywhere. If not, it will turn exactly those planets to produce dust that get a bonus for that exploit like Arid, Desert and Terran.

Science: AI will build it on planets that have a bonus on Science like Ocean, Arctic and Barren.

regarding science and industry output the AI usually picks the optimum configuration in my matches. Sometimes it even uses exploits I "dont get " but double-checking the numbers makes the AI right. Just the food-to-industry conversion messes with the AI a lot.

How does the food-to-industry conversion mess with the AI?

It simply changes all planets to food once it has it because of the multipliers you can accumulate for food make it extremely powerful.



I wanted to express, that I simply don't feel that the system- and exploit-management is the achilles-verse of the AI.

We put a lot of effort into it by modding it and telling Meedoc how we think so he can apply this into future changes on his side.

I think the AI's weaknesses mostly come from other issues.



What I also find interesting: When you look at the general-forums you'll find threads from people saying they shelved the game because they can't win on newbie and here's people discussing why the AI even with it's boni on endless can't compete. ^^

One of the reasons why I don't like the bonuses and penalties of the difficulty system is how it throws a wrench into understanding the situation. People can't quite pin things down, simply because the bonuses obfuscate what is and isn't broken about the AI. Heck, the only reason why I play Endless is because I hope that the AI would play smarter than Newbie - but I am going to guess that "smarter" play is dependent on the bonuses that the AI receives, not from better programming. That is to say a "Newbie" AI assumes it is in a weak position, while an "Endless" AI thinks it has a strong one. Problems arise for the latter when forced into a corner.



This is pure speculation, unfortunately.

You already have harder AI available in the Super Nova mod.



This is what it looks like 50 turns in to a Serious game with 8 factions using a custom faction with all the best traits on a good starting system in an excellent starting constellation.





They've colonized the central star systems long before you get there.









You're in sixth place and will have to come from behind.









You'd better hurry if you want to make treaties.









They've got significant fleets you will have to contend with.









In addition, the Super Nova factions are smarter about colony development, research priorities, ship design, fleet composition, and even targeting strategies.



Finally, they cheat less than the vanilla AI.

^That's looks interesting, thanks! Will check it out. I prefer to constantly be playing catchup, and feel that I win microbattles through human thinking rather than the progressive steamroll.

if you define "smarter" by "being better then I am at nearly everything (research, production etc)" then the regular AI is already smart enough. For me "smart" means clever use of resources and actions taken. If I do everything right and am heavily focused on research and STILL the AI manages to colonize the central hub long before I even get there....thats not "smart" for me.



My personal view only of course.

MTB-Fritz wrote:
if you define "smarter" by "being better then I am at nearly everything (research, production etc)" then the regular AI is already smart enough.

I will be very interested if that "cheat less than the vanilla AI" is true.



I can imagine how an experienced player can enhance the AIs if he has the ability to modify all the strategies the AIs use.



I have read the details of Super Nova, is there any version that only enhances the AIs?

As a modder, I'm limited to tweaking the values exposed in the moddable xml files. Only the devs can really make the AI smarter.



I started with no AI cheating. Then I added what I had to in order to make the game challenging and fun. I've been adding more and more cheating as I discover weaknesses in the AI and try to alleviate them.



The Heavy Nova mod enhances the AI without changing the game nearly as much. It's pre-disharmony but as close as you're likely to get to just enhancing the AI.

I haven't been looking into AI in detail so please correct me if I'm wrong.



As far as I can tell the game AI basically use a set of weights and an RNG to determine what action to take next and how to react in a given situation. Its not unlike a robot that has a set of pre-programmed actions. You push this button, it punch you in the face, etc. The problem is that any such AI is will inevitably predictable because the player controls a major part of the stimuli that the AI uses to make decision. Leading to the perception of a dumb AI.



Seems to me like if you really want the AI to play well, you need a predictive AI, as opposed to the reactive AI above. Something that actually predicts predicts how good the situation becomes for it if it chooses different options. Such an AI would be hard to make since you need to teach it the value of its prediction. But I don't see another way to cover for a reactive AI's natural weakness, short of massive cheating.

That's true, but there may be simple things that can be done to make the current AI much more effective.



These are the three biggest problems I see that I can't mod away:



1. The AI doesn't group its ships into the largest possible fleets.



/#/endless-space/forum/28-game-design/thread/13008-improving-ai-fleet-management



2. The AI doesn't build ships aggressively enough, especially after its initial fleets have been destroyed. I recently jailbroke into a game where the Cravers were at war with three other races, their fleets were decimated, their systems were being invaded. None of their star systems were building warships. Some were even on industry to science. They had plenty of money and morale and threat. I don't understand why they were not building ships.



3. The AI doesn't use invasion ships properly. They should use combat fleets to establish space superiority, then use invasion ships to conquer systems. They should protect their invasion ships with combat fleets. Their invasion fleets should retreat when attacked. Instead, the AI attacks with invasion ships unsupported. I had to mod the invasion ships into half combat to alleviate this; but it's not ideal.

KrayZee wrote:
I want to add that the addition of Bombers and Fighters to the game made the AI even weaker, lets face it, they are pretty much useless, they need to be at least 5 to 10 times as strong as they are right now in the beta patch.

the AI however uses them way to much, even building dreadnoughts around them with nothing but defence and bomber modules (why do they take up 2 special module slots anyway?)

I'm not sure if it was added into the recent patch but the AI does use the invasion mechanic quite often. one time I lost 3 systems in 1 round because of that.. though it was very easy to get them back, because the AI sends siege fleets without escorts

I noticed that.



In my first match(Endless level) after Disharmony released, I found that the AIs kind of like to equip these tiny little bugs, trying to advertise them or what, and they fly around like beheaded flies after their mother ships get shot down.



But it occurs from time to time, I didn't see it much in my last match.



I said that in other threads, too, and I agree with some people who really see the big picture: Disharmony has some issues but not the minor ones which really don't matter at all.



It has a good blueprint but the AIs don't keep up with it. Only the players know how to master it, and it's a pity because we do need good opponents in single player.



But I believe the dev can make that happen in no time.



3 one round invasions in one match? And that is what we call improvements.



I played 5 or 6 Disharmony Endless matches with different factions, finished 3 of them, gave up the other 3 because of the hero management bug in Beta(you have to exit to desktop and then restart the game to access the hero management panel, which is annoying, so I quit the Beta).



And the AIs only have done that one round invasion to me once and bombed me once totally!



If they keep working on it, I think there will be some progress.

There are settings to specify fleet composition by weight, but the game seems to disregard them. I think it builds invasion ships based on the defense of the system it wants to invade, but I'm not sure.



In Super Nova, I've eliminated all the invasion templates except for an invasion Corvette to get them started and an invasion Cruiser for the high end. It helps.



Modders can add rules to prevent building an improvement based on star system status like besieged. Ships can't be controlled directly. They have settings that define threat levels and the maximum % of star systems that can be building ships and they work somewhat. The problem is that there are a lot of behind the scenes factors that also seem to affect ship construction.



In my games, once a factions initial fleet is destroyed, they don't build more ships in significant quantity and it's just a long boring slog to invade their worlds. I've tried changing every setting I can find and so far nothing will make them build ships once they are in that state.

LetoSilvermane wrote:
These are the three biggest problems I see that I can't mod away:

What an amazing insight!



If you are the devs, the AI problems will be probably solved now.



I am not questioning the competence of the devs, but it is clear now why they can do it but haven't done it.



The new title must be one of the reasons. Especially it is a very small company, and I don't expect it can produce tons of great works like other big ones.



And now my solution is waiting, which is not recommended, but seems to be a reasonable one.

LetoSilvermane wrote:
2. The AI doesn't build ships aggressively enough, especially after its initial fleets have been destroyed. I recently jailbroke into a game where the Cravers were at war with three other races, their fleets were decimated, their systems were being invaded. None of their star systems were building warships. Some were even on industry to science. They had plenty of money and morale and threat. I don't understand why they were not building ships.

Because you talked about "jailbreak," I have a question for you.



I notice that X-Com: Enemy Unknown definitely "cheats" because we can save and load, and it helps discover something about the AI of the game.



The enemies in Enemy Unknown will suddenly spawn if the player does something very foolish.



Sometimes what players do is not very stupid, but the way the enemies spawn is quite threatening.



Why do we know it "cheats"? Because if you load the game, you will know the enemies who suddenly spawn are not "waiting" there.



They really "spawn" for the purpose to surprise the players, especially when the players expose weaknesses.



Of course the devs of the game think it is a way to improve the competence of the AI against players.



And sometimes the AIs in Endless Space do the same thing.



First I thought that a fleet suddenly shows up and attacks my undefended or weak fleet or star system because the AIs detect the weakness and then mobilize the fleet.



And it is like how players do.



But after loading the game, I realized that the fleet which caught me unguarded was probably "not there" at all.



It could spawn on purpose.



Because some players talked about if the AIs of Endless Space really "cheat," I think only people like you can answer it.

Pirates will spawn on uncolonized star systems after most of the galaxy has been colonized.

That's not really cheating though. The idea is that the inhabitants of that world turned to piracy since they had no strong central government policing them.

If you park a ship in orbit around the system, it will prevent pirates from forming.



I've never seen Endless Space spawn a faction fleet though.

What the AI in Endless Space does do is use buyout to buy ships quickly, even on a blockaded system. Then it launches the fleet and it appears over that system.





P.S. XCOM Enemy Unknown was a great game. I created a mod for it too:



http://xcom.nexusmods.com/mods/73//?

LetoSilvermane wrote:
Pirates will spawn on uncolonized star systems after most of the galaxy has been colonized.[/URL]

I know the pirates and buyout thing, but I thought you can "jailbreak," so maybe you have seen the AIs do something dirty.



But that is probably not the case.



I hope XCOM(which is great) can have a better multiplayer coop mode, like what Mass Effect 3 has(and my dream is that Endless Space does, too).



The multiplayer of XCOM now is too...hmm, simple, like the toys from McDonald's.



You buy the game, and a lousy multiplayer mode for free, unfortunately.

How does the food-to-industry conversion mess with the AI?

It simply changes all planets to food once it has it because of the multipliers you can accumulate for food make it extremely powerful.

And thats where the problem lies. Often enough when I select production value for the AI he even terraforms the planets to lava in order to max out numbers but once I research the appropriate tech he terraforms all planets to Ocean (!!!) and uses industry exploits, some of the food improvements simply get ignored (like more food on planets with a moon) in the end he lies at 60-70% of whats possible with the "correct" setting of planets and exploits (still being jungle with food exploit)