To be honest, the AI is in a disastrous state at the moment. Even at the highest available difficulties, the AI is inefficient, suicidal and lacks a decent strategy to actually win the game.
If you browse through the posts, you can see more and more beta players complaining about that and in my opinion this is the worst point of the game currently. One can forgive if there is some small graphics bug, or if the UI is cumbersome, maybe even seldom game crashes. But one completely looses interest in the game quickly, if the AI is not able to challenge you and make such stupid things like
Sorry for being cynical, but you devs/Stardock invested a lot of time and money in the game development and now (are forced to) release too soon. I fear the game test scores will be much worse due to that fact and your ROI too. In the old Ascendency-times it was kind of ok that the AI was dumb, but nowadays in view of e.g. self-learning neuronal networks getting common (there are even open distributions of the available for C++/Python etc.), this has to change.
I'm looking forward to your reply and beg you to focus on AI-improvement, please!
Thanks for your honest replys and
I'D LIKE TO APOLOGIZE FOR THE LURID TITLE I'VE CHOSEN. Good that you renamed it...
The reason was, of course, to grasp your intention (and get more feedback from other players) because I think the game is already great and with a little AI improvement it can easily get the best 4x available for the next years. You put already so much effort into it, so let's try to solve the "last" tough issue.
1) ad better AIs - on an ordinary CPU:
I'm not aware of better AIs, sry. But that doesn't mean it's not possible to make a really good one. Sad that (deep) learning strategies are computationally too costly.
2) Suggestions:
- on harder difficulties, I'd like that the AI really plays as good as it's possilbe without cheating. Cheating doesn't really help because players will recognize and dislike.
- as far as I understand, you (Frogboy) have to "kind of script" the AI based on Feedback. As it is impossible that you yourself try each and every strategy, you could maybe launch a contest (after game release), where people send you their games. You can then analyze the best games and implement the best strategies from these human players. To get to know the best players, maybe you could introduce kind of a online "high-score" list.
- ad trade: a human player only trades, if he gets a "significant" advantage for his strategy. The AI algorithms evaluating if a certain trade for the AI is significant should be made "tougher". And if a certain trade is not in line with the AIs winning strategy, it shouldnt trade at all, just to be on the safe side. E.g. if the AI plays military it can trade in military techs, but not give them away.
- ad trade II: I'm not sure, if the AI factions are trading with each other at all. If not, it could resolve the "round-trading-issue" because then sooner or later all available techs are quickly spread to all others (or at least to most techs are spread to almost everybody). Thus the gap between the human trader and the AIs is closed a littlle bit.
- Fleets: a human player usually tries to avoid ship losses and waits until the victory is (almost) certain or only bad/old ships are lost. The AI should calculate its chance before battle and only engage if that is also the case.
- Colonization: as said by others, too: AI has to make use of adjecent tile boni and group thing together. Even if it needs to do some terraforming before. That maybe difficult to program maybe...
Frogboy, definitely appreciate the insight.
Here are my thoughts on those questions:
1) The way buildings are right now, its almost always better to build rather than wait. I would much rather build a few research labs than wait for one to upgrade to a higher tier. And since the buildings upgrade later, why wouldn't you start building early?
2) To the question of "soften" vs "kill", again its almost always better to kill than soften in the game right now. Fleets provide orders of magnitude greater strength than their individual components. To me the default should be "create the best fleet you can, then go kill. If you can't kill, defend a starbase or planet to gain the benefit of their defenses".
In my own warmaking, weakening a fleet or a starbase is just not worth it compared to building up a kill fleet.
One of the good things here is that the data driven nature of the AI will allow players to create lots of AI tweaking mods.
I think understand you want the AI to be as good as possible but am puzzled why you would not recommend buying the game on the basis of the current AI when you can't recommend a game that has better AI instead. Are you suggesting people don't buy any strategy games because all AI is sub what a skilled human player can do?
Frogboy's reply confirms what I have always believed about having the game ready to release.
Most of the Founders and many other beta players are experts at the game. Being experts they are great for finding bugs and making great suggestions but they are useless for advice on when or how to release the game. 73% of players are using the beginner setting. That says it all for me.
The one thing that new players hate are crashes. That is the only thing that they must minimize by release day. All the rest is going to be great with the 73%, and they are working on making the other 27% happy as well.
I don't want to get argumentative here but, which game AI are you comparing us to? The reason I ask is that there are still people out there who think GalCiv II's AI isn't good.I agree with the points you're making. Don't get me wrong. But even the much weaker AI in the current game is more than enough to handle the majority of people -- on normal. I realize this is small comfort to people like you (and me) who can crush it at higher levels right now. That's what I'm working on improving here. The colony management AI is something we've asked to have more time spent on. The answer I got was that the AI is waiting for better techs before filling in the slot. The problem is, how long should it wait? Clearly too long! My recommendation / request is that we look for specific things we can do to make the AI better. Because that we can do.
Ah, that makes sense. I think the AI should at least build up the planet using like 25% of it's manufacturing budget to create sub optimal buildings to get something. Maybe have a primary goal and a secondary goal. The primary goal is what is wants to eventually achieve the secondary goal is what it can achieve right now. The weights should be heavy when they can build the primary project but lighter for the secondary...I donno. I don't think they should wait until they get all the optimal techs...or at least modify the research waits so they go after the techs they need.
This.
First thanks for replying and I guess I'm a better player than I thought I was. I don't think I am comparing your AI to anyone else's I can't think of an AI that really impressed me and I should have been clearer that I do think the current AI is a good start although if it is already good enough to challenge most players its a lot more than that, mostly I've been thinking that it needs more polish. I think colony management is definitely what needs the most work and I guess I shouldn't understand what the AI is thinking waiting for an improvement but most of the basic improvements come pretty quickly and I would think some bonus is better than nothing so might as well build something.
As others have mentioned, the AI should begin building structures on it's planets IMMEDIATELY. Even if it is something as simple as build 1 manufacturing structure and then 1 research structure, until at least 50% of the planets tiles are covered. Even this very simplistic approach to AI colony management would vastly improve the overall AI strength. After all, what is the real primary source of power in the game? Your colonies of course. If the AI doesn't do anything at all to develop it's colonies, it horribly gimps itself. Even the wrong "something" is better than nothing.
Keep up the good work. The deadline is coming up fast, but I know you can push a decent product out the door by the 14th.
P.S. And clean up the "empty" techs in the tech trees before release please
Most of these things aren't really AI questions but ones of judgment -- human AI data judgment.
On point 1, the main problem looks to me at how it is using it's tile and its bonus.
On point 2, I need to do more testing, especially with your recent change in unreleased patch #2, but it seems to me the AI (well, some races) is pretty good at going for ascension or research victory, less so for conquest. Other problem wich you seem to have fixed is a race going at war with everyone while struggling with one race. Say, Drengin vs Altarians, they are pretty equal, but then Drengin goes to war with Thalans and me. But apparently, this is fixed, so again, more testing when we get the patch
On point 3, so far, it was hit and miss. But I see, again, that patch #2 has correcte many mistakes, namely that the AI was ignoring my stationned ships in his evaluation of my strenght. That led the Krynn to declare me war every 10 turns or so, asking for peace as soon as they could, rince&repeat until I've destroyed them.
I would recommend contrary to the posts above, letting the AI cheat. The AI should get some production and money bonuses. It is struggling with colony management and getting its empire set up quickly.
The other issue as raised is that the AI needs to be building up their production buildings right away.
2) To the question of "soften" vs "kill", again its almost always better to kill than soften in the game right now."
Also agreed.I didn't realise the vast majority play on beginner. In that case release seems fair enough. Curse my too great love of 4x games! Here are some further thoughts on the AI:1. Multiple wars could use work. If I end up in multiple wars I'll almost always pay more than usual to go to peace with the weaker faction. Then take on the stronger one and crush the weaker one when I have more free resources. I think 'attack strongest opponent first' is usually vital. If you can always keep ahead of your strongest opponent, you'll win. Of course if that opponent is too strong, picking on the weakest opponent and claiming resources can be fruitful. But if I'm doing that, hey, now's a good time for my rival to attack me!2. Trading tech values needs more computation. If I'm playing as the Korx, maybe don't give me your race-specific economy tech? Or at least charge me more, otherwise my economy will become horrendously powerful. Here's a thought. Try to work out what victory condition the player is going for (doesn't have to be perfect, humans aren't). Apply a multiplier, say 1.5, to that sort of tech. 3. How much do you actually want those bells and whistles I'm giving you? If you've spent the whole game investing in missiles, you're near end-game and I offer you a laser tech, it's probably worthless. Unless of course your rival has great chaff and no shields.4. What techs am I near? If I'm about to research medium hull, I probably shouldn't go to war now. I'll wait to build some of those nice big ships, then a few turns later I'll be in a much better position.5. Can we use player stats from previous games? Maybe as an option? A sense of persistence would be cool. I know the Drengin are warmongers. This might be more of an expansion thing but if, say, the AI knew, "90% of the time he/she pursues an influence victory," they could try to guard against that.
Great stuff!
here's the best way to format these as I can tick these off in my notes as things to fixate on:
What is the bad behavior you're seeing and what is the context you're seeing it in.
Suggestions on HOW to handle it won't do much since most of the time, especially if it's obvious, it's already doing it.
Let me give you an example of what I mean:
The AI lets me park a big fleet on its doorstep, transports and all. There is a ton of code already in there to look for this and react. Why isn't it?
All I need to know is that you see this behavior. You can generally presume that anything obvious is already built in and that there is some other reason for it happening than "they need an algorithm to detect threats!!!!"
Korx? Em, I meant Iridium Corporation of course. Hah
I want to start by giving a massive shout out to how awesome this game is already, and how my expectations are somewhat inflated by my experience in gc2.
My first experience in 99% of 4x games was that the AI on normal was terrible and I would normally win my first serious game. There would be setbacks and I'd do dumb stuff as I learned, but the AI for these games was normally nowhere near aggressive enough and made lots of dumb errors.
I was a latecomer to gc2. I picked the game up with the twilight expansion in a bundle in a steam sale because I was bored and looking for a 4x to burn some time. I distinctly remember creating my first game on normal... and getting absolutely crushed.
The AI felt like next level to most games. I remember playing 3-4 more games getting crushed before I finally got the hang of it, and managed to scrape together an epic win. I eventually was able to beat it reliably across many more difficulty levels - but it required a level of understanding of mechanics and strategy light years ahead of other games of its time.
I really think it was the main reason gc2 is widely considered to be one of the best 4x's ever. Other than the ship designer, the concepts weren't that innovative... but the Ai - while not perfect - was so far ahead it just made the game legendary.
Hence why I am holding gc3 to such a high standard. The problem with being a sequel to an awesome game is huge expectations!
So the AI definitely needs work. I want future newbies to enjoy their first experience of being utterly destroyed by an AI! While somewhat chastening, it's also really addictive. After all, who wants to lose to a computer?
But I really think that by far it's biggest weakness is economy management. It's just really poor right now.
I think what it needs is a general planet building strategy. Start with factories and 100% production, scaling it back to optimise waste. Specialise. Build planets in a ring. Hold off building a shipyard unless either needed (drengins feeling hungry) or it reaches 1 turn build time. Rules like that. Build factories on research / wealth world's to shorten upgrade times and increase uptime. Build economic starbases, and place them in roughly the right places. Have a tech strategy. Etc. Basic rules-based stuff like this for Developing a strong economy.
For insights into how to build a good economy, and the rules used, talk to some of the top players here. The people who have written posts about the economy and how to manipulate it.
That will be the key to a release-worthy Ai. Because at the end of the day , even if the AI makes the occasional dumb decision with its ships, or makss the occasional bad trade - if it micromanages it's economy well it should be able to match most players with overwhelming force and present a real challenge that I know will entertain newbies and hardcore players alike.
Build planetary improvements in a ring with a capital or solar power plant etc etc in the Middle*
Can do. Here are a few I've noticed:
Colony Building
1) Not specializing buildings. Money, manu, and research buildings on the same planet, which it should be only 1 type (maybe a few manu buildings just to speed things along, but you should not see money and research buildings on the same planet....even if a tile gives a big bonus to it.
2) Not building colonial hospital. Or if it is, not putting the farms it builds around it.
3) Adjacency buildings (fusion plants, coordination centers) put in the edge of areas instead of the middle of rings. In general, these kind of buildings should always have at least 3 connections.
4) Not moving the slider as strongly towards a specialty as it should.
5) Mixing in influence buildings with other types. Generally its better to make "influence planets" than to mix them in
Warfare
1) Fleet Separation: In my last war with the Thalan, this is what he would build:
a) 1 full fleet
2-3 "half fleets" (using about half the logistics that they can afford)
c) 6-7 single ships, incapable of doing any real damage on their own.
These should be converged into all full fleets.
2) Attacking Starbases when you can't win (pirates should to be really bad about this in 6.0, I've noticed improvement in 6.1). Further, they don't do it to "soften it up". They will attack it, and no other ships will attack it, even if they are only a few turns around from the starbase.
3) Not attacking Starbases aggressively. Once the AI has a "kill fleet", it won't target starbases aggressively. Best example I have is a full Thalan fleet (all logistics used) nearby one of my starbases that was near 2 of their planets, and it didn't take it out.
Nice discussion, we had the same discussion on the Civ 5 AI on Civfanatics some years ago. There's more to an AI then just making it play the game efficient though. Should GalCiv 3 have a fun AI or a Cutthroat AI?. A Cutthroat AI will for sure make the game harder to win but not necessarily more fun.
I'll toss the same question here: Should the AI play the game to win or play the game so its more enjoyable for the player? Its really hard to make the AI do both good.
You should always aim for the best AI you can make, humans will always find ways to beat it with today's computing power. There are always easier setting where they player actually get bonuses which can be used if you are a beginner.
The AI should always build up their colonies as early as possible at least until there are only a few spaces left, say three or four perhaps even fewer when you consider early terraforming of a couple of spaces. If the AI have to destroy an improvement later on that might be so.
The AI should consider approval before population since having a high approval is way more important than more population in almost all scenarios. Low class planets should consider having no population improvements at all until terraformed to a certain size.
One thing that often also is a waste is upgrading buildings too early, many of the second tier building are too expensive both to build and in maintenance for what you get from them so waiting until later to upgrade would be a good thing to teach to the AI.
Example...
You have a planet with five basic research improvements that give you 165% research bonus and two basic industry buildings that give you 80% extra industry. The RAW output is 12. It will cost you 45 production to upgrade one research building to the second tier and all five will cost a total production of 225 production from a colony that produce 22 production a turn... thus is take about 11 turns to upgrade all the improvements. The gain will be +25% research from 165% to 190%. During these 11 turns you could produce 352 research points, it will take 117 turns for the upgraded research improvements to start break even after they are upgraded. It is often worth more to gain a few techs extra during those 11+117 turns than wait 128 turns for the same amount of extra tech.
I NEVER build up my planets with more than tier one improvement until I stop colonizing new planets, with some exclusion for wealth buildings. One or two industrial buildings is enough on none manufacturing worlds, money also help you build. The basic buildings is good enough for a very long time as long as you expand, money should go to helping new colonies.
I think the AI cold be taught to ignore upgrading buildings until some threshold is met and it can start concentrate on improving their colonies from the core and out slowly.
The same way it can be good to have one or two manufacturing improvements on non manufacturing worlds it can also be good to have either one or two research or wealth buildings on a manufacturing world. Even the AI will want to cut production of ships at some point and concentrate on research or wealth. I find that it actually is worth having at least one such building in a good spot and loosing one industry will not matter much.
AGreed with Jorgen. While he and I differ in exact strategies (I will tend to upgrade a bit more than he does), I think what's really important is that you implement a simple rules based system for developing a good economy. Doesn't have to be perfectly optimised, but Does need to be functional and competitive with the player which it is not right now.
For example, when an AI colonies a world, it should decide up front a specialisation for it, based on planet bonuses, planet class, number of planets in the system, and finally tile bonuses. Basic logic is:
1) is this a single or multi planet system? How close is this planet to others? If close, it's a great candidate for a manu world, since shipyards can pool production from multiple world's. If solo, it's a great candidate for wealth or research world's - since these do not benefit from being near other world's
2) how many of each specialisation do I have already? The Ai should be aiming to roughly stick to a ratio between the specialisations, based on what victory condition they are currently aiming for and their race characteristics. Ie drengin conquest, altarian influence/tech, etc.
3) does the planet have any bonuses? ghost world etc.
4) what tile bonuses are present?
5) what class planet is it? I generally reserve high class planets for my primary and secondary specialisations. Ie, if I'm prioritising research, then I want to focus on research and prod (for expansion, military etc).
Based on those factors (and maybe more), the AI can then make a good specialisation decision.
Then it needs to choose what to build where. Ahead of time. Again, as mentioned, the AI should always build a few factories first with 100% production to speed up setup times.
When choosing, the AI actually has an advantage!!! It can plan out the world much better than a human, as it can.model where to place the terraforming improvements. Humans can't really do that since they can't see what tiles are unlocked until after they've researched the techs, usually in midgame.
Where there is a big cluster of tiles, build amplifying buildings (solar plant, etc) in the middle, then core buildings around the outside. Where the clusters are smaller, build non core buildings. Factories if not a manu world, farms, approvals. Again all clustered.
Just implementing that stuff alone would be huge. The AI would become so much more competitive.
I mean, humans can kind of guestimate it based on how much land in the tile... but its in no way a perfect science. Unlike the AI who can calculate up front...
Behavior: AI fields tiny attack only ships.
In early game, the AI puts out tiny hulls with attack only ratings of 2, 4, or 8. They are glass cannons at best, not even good enough to handle pirates. I think they should wait for a balance of offense and defense, or at least a minimum of defense, before fielding ships. And they should prioritize the defense a bit more as part of that. It is possible they already have those kind of evaluations, but they don't appear to be working by what I see.
I agree with the poster who said that the AI does not use its logistics strength well. Also with the colony management issues. Both are described above beyond my ability to improve.
As to the general thread content, I seem to have a very different view. Talk to me in two years or so, and I'll decide how well Brad did. Any sooner is just guessing, really. Thia interchange is just the beginning, which is why it is wiser not to panic or declare urgency over issues that only affect us connoisseurs of the 4X genre. I don't say experts, because I am nowhere near the level some of this is talking about. But I am part of the folk delighting in the nuances. My real problem with earlier posts was that they were all about what was disastrously wrong and none of the mounting delightl. Now we have a mutual conspiracy to make the AI clever. Excellent!
I realize this is a long heated thread and to be honest I didn't read it all. I just wanted to toss my opinion into the ring. Taking into consideration the AI in it's currently released form as of my posting. 1st off do I think the AI is done, or perfect. No and I don't think anyone here is making that argument. Is the AI advanced enough for release? I would say so. It acts well enough to be unpredictable. Are there improvements that can and will be made? Of course. It is possible to predict and react based on play experience. That can be said of any game though. I play alot of 4x so I can say with very little doubt that AI in any game can be bested by having a deep understanding of how it acts. Now with that being said, I do feel the AI is advanced enough to hit the metaphorical shelves. Upon release players are going to purchase this game, and spend hours playing. Those who have been playing beta have already done that. It has taken the experience of playing the AI in it's progressive changes to figure out how to game the AI. It will undoubtedly take a new player longer to develop the understanding that current players have. What I'm getting at is yes in it's current state it is ready to be released. During the time from release to patch 1 or 2 or .... they already intend to keep reworking it based upon testing and player feedback. We fortunate ones that are already playing have had the experience of watching this project grow and develop. We will continue to watch as it continues to do so over the next 10 years give or take. Yes I'm a long time fan and possibly biased, but it is because of that that I know what to expect. This game has not failed to miss a mark. I'm sure at some point there will be set backs but they are running a solid game. It's had/has bugs and work is needed but it's been one of the smoother betas I've ever played.
That is my two cents. Take it for what it's worth. I could list 100's of games I've paid for that yielded far less entertainment then this already has. Play on players.
The game is for multiplayer, so isn't AI just for filling seats between players?
No... the GalCiv is primarily a single player game... all turn based games are. I would like to see any insane map MP game that would ever be played to it's conclusion, or even half way for that matter.
In short... the game is a single player game with support for multiplayer. I doubt more than a few percent who get the game will ever play multiplayer at all.
Didn't Frogboy even claim the game is made for the AI and not the player...
Some troubles in Trading with the AI...
The AI are not asking enough for their infrastructure such as ships, planets or bases.
In a recent game a pretty much Neutral faction only asked about 1000 credits for a mining base they just created close to my capital. It actually even had three resources connected to it. It would cost me 1600 credits to rush a constructor to build that base myself.
The same seem to be try with pretty much anything the AI want to trade.
As said by others the AI need to ask allot more for their trade, at least until they are very friendly with you and the closer you are allied the more you can freely trade.
The AI also need some thresholds such as I will NEVER trade this no matter what. Bases should be like that. A base that are connected with their main cultural border should be more or less impossible to trade for unless you are closely allied and to a steep price. The AI should never give up colony or other support ships unless you are closely allied and they have plenty to spare or you are much weaker than them.
The AI should never give you military technology and/or military ships unless you are much weaker and/or allied. The AI also must consider OTHER threats around them as well, not just the one they are currently trading with. This also hold true for tech trading in general.
In general the worse your relationship the more the AI should ask for any trade you wish to make, trade which they initiate should be better but only be made with things they really desire. So Drengine will ask for additional military specialization technology once the researched their own version and you have something else, they might actually give a good deal on that. If you try to ask for a military tech or some other technology they value more than others it should be very difficult and expensive.
It currently is too easy to abuse the AI i trading without first building up a good relationship with them.
I also think that there should be a system for AI keeping track of deals made between others, perhaps even a way for the player to get notices of AI trades in technology for the same reason. If the AI notice how someone (AI or player) trade lot's of technology they should be more cautious and ask a higher price for their technology and rather trade with someone else than the one how tries to outplay all the other trying to hog all the benefits.
Simple solution is to increase the threshold of each tech trade and place it in some sort of cool-down, thus encouraging trading between other parties in between so technology gets spread a little more even through trade. You can even have the AI tell the player that if they sell a tech to other empires the price of that tech goes down.
Or a system where the price of a technology is cheaper the more empires that has it... in addition to raising the resistance of trading the same tech several times in a short while. That would stop some of the players attempts at abusing the AI in tech trading.
There are many great features available to you once you register, including:
Sign in or Create Account
