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!
The single greatest AI issue right now is its economic management.
It's just really poor at developing a strong economy, which causes it to fall substantially behind in terms of research and production to the average player. Subsequently the AI is still running around with medium ships with 20 attack when I have huge ships with over 400 attack. It's a whitewash. Surely this isn't that hard to fix?
The second biggest issue is its lack of military tactics / sense. but if you fix the economy, the ships the AI is using should be closer in power to the player, and so this will seem less of an issue
I think if you fix those it will actually be in a reasonable state.
Edited as I was an ass to the Original Poster and I have removed what I posted here. Thanks Halimac.
In any event the AI is VASTLY improved over Beta 5 and VASTLY improved over GC II and will continue to get better. Did you review the new beta patch 2 notes? Here let me post some of them for you...
AI
Improved logic for when the AI keeps ships as planet defenders or ejects them to go chase down threats
If the sponsor is under threat, the AI will rush build (or try to if it can)
AI will reduce the weight priority of a ship project to build based on the number of turns it will take to build that ship
AI will tend to focus only on its primary opponents and is less likely to go after "that guy" (reduces spreading himself thin)
Rewrote the surrender code
AI much more likely to surrender if it is basically out of the game.
AI can now try to bribe other AI's to declare war on players
Lots of new checks to make sure the AI behaves as you would expect based on their relations with you and each other
AI checks relatiosn before trading with each other (no more tech trading with people they don't like)
Eliminated the use of the wealthslider setting in AIdefs in determining balanced budgets. Balanced budget request to AI supercedes its wealth slider preference (translation: the AI now has logic to control its spending and won't be constrained by the xml guidance values if it needs to do something extreme)
Shared borders is no longer a negative issue if they have an open borders treaty
AI won't suggest a player attack another player if that player is stronger militarily
If the AI is losing a war it will build ships even though it might go beyond the maint' cap
AI more prone to building scouts to explore early on
AI greatly reduces freighter building priority if it is losing a war
At higher difficulty levels the AI is more likely to aggressively rush build
AI determines how many ships it should have in orbit based on how good the planet is (instead of caring about the logistics)
Reduced the number of ships a planet thinks it needs in order to feel defended
Reduced some of the bonuses the gifted AI gets
Gave a range bonus to the higher difficulty levels
Rebalanced the defense importance modifications (how important a given object is to the AI)
Increased the valuation the AI gives to building assault shiups
AI takes into account the cost of a ship when deciding whether to build it
AI keeps track of enemy faction power
AI is made aware if it is starving its military of funding
AI is much less likely to defend starbases and shipyards with new units than preivously
Modified the rate at which the AI starts diplomacy with the player
AI is more sensitive to military buildups by players
AI doesn't demand tribute or tech unless it has a stronger military
AI now decommissions shipyards that it isn't sponsoring anymore (most likely because it lost the planet that was sponsoring it)
I agree that the AI management of it's colonies are very poor.
The AI build very few buildings on planets and build them very haphazardly when they do. The AI does not seem to understand how to use buildings that boost other buildings and standard buildings together and have no clue how to specialize worlds.
Currently the AI seem to only manage the focusing sliders on empire levels. I just monitored my AI empire yesterday and all planets used the same slider settings or at least very close to it. It was basically around 50/35/15 (manufacturing/science/wealth) and 60/40 or 70/30 for military/social spending. This make the AI develop their planets extremely slowly and also stop developing them after they build a couple of buildings.
Most smart players set their focus into 100% production on new colonies until they built up the base of that colony starting with the industry. Despite the different terrain on planets giving you bonuses in different ways these are hardly something you bother with. Every planet are designated to one area... be it industry, science, wealth, influence or military. Given how the mechanics work there are no reason to do otherwise so the AI should be taught the same.
An efficient and smart player rarely have a slider on anything but 100% in one direction or another... almost no reason not to from an efficiency standpoint.
For me, there are two AI changes that would dramatically improve its competitiveness in short order.
1) Planet Building Updates: Summarized in a few basic blocks:
a) Specialize planets: No markets and research labs on those manu planets.
Use adjacency bonuses: Put power planets in the middle of rings, but the tech capital in the middle of a big ring of research labs, etc.
With those two things, the AI would get significantly more resources, and automatically make it more competitive.
2) Link up fleets (no more stray craft)
I know there was an update of this in the last version, but I still find the AI doesn't clump its fighters into fleets fast or often enough. Its too easy for me to pick off stray ships and bash down half formed fleets.
Especially when an AI has moved into my borders, it should be full fleets all the way....at least until it becomes much more sophisticated in its tactics.
There are of course a million other things you can do, but those two alone I think would greatly increase the AI's punch.
Quoting Larsenex, reply 2Play on Insane with Godlike AI and 30 ai and come back to me and tell me its 'too easy'.
Larsenex, be nice. Everyone, including you and me, are entitled to an opinion. I think many of us in the forums were surprised by the release date, thinking the game needed a month or few longer to bake. The AI IS problematic right now. With the latest patch, at turn 185, my score is almost 1900 and the nearest AI major power is in the 400s. As I've noted in other threads, Challenging difficulty is simply not right now. I think one major reason is simply the AI not being able to adapt its strategy to the addition of pirates into the game. At about turn 180, one of my Explorers watched as a major power tried to move an unarmed colony ship through a pirate infested area. The AI has to be smarter than that (against itself). I am hoping this is an easy enough fix before the game is released and everything will be okay.
Theeannon, my opinion, I am holding off on "final judgement" until I have the actual release game. I will post reviews publicly only then. Stardock is a business and I assume the release strategy is part of a larger business strategy. Like the Civilization franchise, I imagine this game will become great over time. I know that, even in its current state and with all its flaws, I do have fun playing, though I have been a fan of this franchise since the first Galactic Civilizations 2 release (sorry, I don't go all the way back to the original).
I think everyone should keep posting their opinions of the AI, because otherwise this game is not going to get great.
My opinion.
i have to (sadly) agree with the op. I made the same observations, especially reading it's economy.
Just a short reply:
0) I'm for sure not the only one; see above or link below for an older thread...
1) I played on insane and godlike with 5 major AIs (the rest "occisional"). I built 1 RnD planet, 3 Manu planets (incl. Homeworld; which has RnD, too), 1 Wealth, 4 trade routes, 2 shipyards; only a two starbases); traded a bit of my "not so important techs" for x credits per turn (as explained), got lots of money; can thus buy quickly new buildings --> get in tech lead (or 2nd) quite easily and also by far 1st in wealth; also if sliders on 100% military for a few turns, can quickly build up a good enough fleet (traded in military techs from minors by also giving them some "unimportant techs"); then I stopped because it would now be easy to overrun the AI, which has almost no ships (no defending ships on planets), so too simple...
Of course, it's necessary to micromanage efficiently, i.e. adjust sliders such that you can ramp up a few factories within a few turns (actually with all the trading money I get, I can easily buy 3-4 factories on the new planets; I have a per turn income of 700 or so, just by gaving away useless techs...).
2) In beta 5, I played a larger game. Same story.
3) Maybe that changes for 30 factions, but it has to work also for smaller games, don't you agree?
Suggestioned Improvements:
1) AI has to improve as the guys above suggested its colony management. They have to specialize planets and group buildings appropriately.
2) Fleet solutions have been suggested.
2) !!! Trading: as I already mentioned in a previous thread (Extremely dumb AI-trading behaviour spoils whole game: https://forums.galciv3.com/463156/page/2/#3539608), it's too easy to get all the techs from others and get tech leader by "trading round-whise techs" (this is still valid for beta 6) Also, as desribed above, it very easy to get lot of money via "per turn credits for useless techs" (especially that shouldnt be possible with minor factions).
4) AI should have a descent strategy to actually win the game and focus on one or two victory possibilites:
E.g. the plan could be for a certain faction to win via defeating all others, thus they have to research military tech, build Manu planets a few (not 100 shipyards), forget about the rest (except maybe more food, a little entertainment etc.) and attack quickly others with a good enough (!) fleet.
Or for another faction: focus on ascention points and persue only that, but quickly and efficiently. The AI seems not have a good focus on "winning the game quickly"...
For all those haters...
First off the game is still in beta, late beta, but still beta. Does the game have it's faults, yes, is the game fun, yes, will the game be ready for release? I have no doubt it will be a very good game on release, and most of the larger issues still lingering will be squished.
Outside of the above argument...
What game company these days releases a game with no bugs? The answer is none to very little as all games have some bugs.
What game companies over the last couple years have flopped major releases by releasing games not ready on release date? EA
How can Stardock help address your concerns (other than adding features that were never part of the game design)? This may help to resolve your issue...
How could one be part of the solution versus part of the problem? Make a suggestion on how to fix the issue versus just tearing down the team and the game.
I hope you change your mind about the game once you see the finished product on or shortly after the release date
I don't see the op as "hate" just blunt.
the message is "the state of the aI is hindering enjoyment of the game"
you may agree or disagree, but there's no spite there.
If you don't want to buy the game because of the bugs just wait and keep this game on the wishlist. Stardock has great after release support and the CEO is an AI guy who loves to tinker with the AI well past development. He even released AI updates somewhat recently for his 15 year old game. GalCiv2 had the best AI for it's time that you could find. I would recommend at the very least keeping this on a wishlist and checking back after a patch or 2. jThey have stated they plan on updating this game for like 7 years.
If he is playing the beta to provide feedback, then he has already bought the game. No need to keep it on a wishlist.
I tend to agree. I'm worried that the AI has lots of fancy features whilst struggling to play the basic game. There are lots of nice touches, conversations, attention to trade, etc. But, as you say, planet specialisation for manufacturing/research is a problem. Trade is a major problem... sure it's getting better but, one example is open borders. Open borders = 1 trade good + 1 gp.Works for me every single time. At least have some randomisation or something. Also, there's a more difficult problem. If I don't think the deal is good, I won't trade. The AI seems to use flat values for everything. For example, if there are lots of barren planets near me, it should recognise that colonisation tech is worth more to me.I understand 4x AI is incredibly difficult. The older I get, the more I struggle to find 4x where the highest ai-doesn't-cheat difficulty is a challenge. Asymmetric games like AI War are pretty good at addressing this. I'm certainly not saying I'm amazing at games, there are far better 4x players out there, but I'm sure they're all forced to play on AI-cheats difficulties.TLDR: The AI has lots of bells and whistles. But sadly I think the fundamentals, like colony management, need major work. That's not meant to insult SD, the progress has been great, but I think it stills needs major effort before a release that is, I worry, too soon for reviewers. (BTW, I've spent countless hours on GC2, if that helps. My major problem with that AI, which was great for the time, was that racial tech trees in TOA made an enormous mess).
I hadn't considered that pirates were having an impact on the civ growth. Civ 5 has issues with that, on the difficulty level I play the AI sometimes becomes broken because it loses settlers. In one game I played my neighbour only had 2 cities but I found 3 of their settlers in barbarian camps near by, which ended up killing them as I was able to very quickly outpace them
For me, the greatest single "issue" that this game has is the AI colony management. On every planet I culture-flip from the AI, it has built absolutely nothing. Nada. Zip. Zilch.
I assume, apart from its insane military spending, this is the reason the AI is ususally constantly broke.
Get this issue resolved and the game will be truly great.
Even now, this is already a bloody fine game, despite some other kinks.
I agree with your points, but I disagree that the game isn't ready for launch, there's still a huge patch (beta 6 patch 2) in the oven that made a lot of improvements to the AI. And SD will keep supporting the game for at least a year, probably more, as they did with Galciv2 (and other projets too, I think, though I'm only interested in Galciv series and Start8 ).
Now, about tech trading, that shouldn't happen with 6.1 though... The AI is supposed to detech when it will go bankrupt by giving you too much money, and as far as I can tell, it is working on harder than normal AI. And it shouldn't give you lots of stuff for crappy techs, unless you have a huge diplomatic bonus, wich tends to happen in late game.
1) I think most of the game is great: graphics, lots of features, UI, sound, ship designer, tech tree, faction design. Very good!
2) As I understood, this whole beta testing effort is because you want to receive feedback from players to be able improve the game before release. AI is, for me, the issue that needs to be improved most. That's why I posted the issue at all, trying to help you in creating a very good (as opposed to "only good") game because I've the impression that the dev team is very capable of actually achieving a much better AI. Again, great job until now (except AI).
3) Patches: of course, there will be patches, but these should not be for key game issues (such as AI). No doubt, there are real-llife (economic) reasons that don't allow devs to postpone a release over and over again. If it's simply not possilbe to do that, I'm waiting for AI-patches after the release, of course.
Until now, thanks for your efforts and good look for the last (hopefully also AI) issues and hopefully lots of people will buy the game and have great fun with it.
You weren't around for Elemental were you? Same things were being said in the forums for that beta too. The likelihood of GC3 being released with a lackluster AI is now approaching certainty.
I don't mean this critically, most games have this problem. And it's probably a smart financial decision not to delay release any longer. Most people who buy the game aren't going to play it long enough to notice how clumsy the AI is, which will mean that the AI is good enough. If you understand the mechanics of the game and hearken back to the challenge of GC2, it's likely to be a glaring issue.
Part of the problem is that it's incredibly difficult to design an AI algorithm for a game that isn't complete. Brad and his team could spend thousands of hours honing strategies for the game and then have an engine change or balance pass invalidate all their work. Brad has talked about that specific challenge on these forums.
And yet it doesn't appear that Stardock made any adjustment their release schedule for the game to accommodate that reality. They should have planned to spend at least a month in Beta after the game was completed hammering on the AI. To be fair, nobody does that. Nobody sits on a finished game for a month and especially not to work on a "feature" when 80% of purchasers wont even notice the difference. Making games is a business.
Greetings.
I appreciate the feedback. I will ask the team to look at the things you mentioned. I changed the title to be less inflammatory.
That said, I have a few thoughts on this:
Re current AI:
Most of the more significant AI work isn't going to go in until the end of this week due to the amount of QA they require.
Re the argument that a month should be spent just hammering on the AI and nothing else
That would be a disastrous choice to make. According to our stats, 73% of players play on beginner. BEGINNER. And even in Beta 6, if you bump up the AI to the harder levels, it's more than enough to defeat most players -- and that's without the changes we're putting in to make it better. No studio is ever going to just go idle for a month before release just to work on AI. And you guys who know me know I'm obsessed with game AI.
Re trade screen abuse
This is a tough one to handle. This is why most games cripple this stuff. As a practical matter, all we can do is wait for posts (like this) that find an exploit and nail it.
Ok I removed most of what I (previously) posted.
Obviously the OP has issues with the Ai.
What I really love to hear is not so much the criticisms of the AI but show me a game with BETTER ai than what we have now. Tell me the game and how awesome its AI was at launch. I would like to purchase it today and make an comparable response. I say this because it seems to me the OP has played games with far better ai and is making comparisons based on that. Yes I was around for Elemental. A disaster. I am 100% sure that Gal Civ III will not have that issue, at all. Brad has declared on dozens of posts that the AI is an ongoing work and will continue to 'need' updates for years after release.
Again I want to hear about a superior ai and game to what we have now.
I’ll be honest I don’t share Frogboy and your obsession with the AI for a couple of reasons;
1) I don’t really play competitively; I want an AI just good enough to keep the game interesting as I conquer the universe, not good enough to have a great chance of beating me. The AI is already good enough to give me that.
2) A good human player will always beat a good AI unless it employs machine learning techniques superior to anything we have today, even deep learning systems are useless at long term planning and I do mean really completely useless. Game AI in this sort of game doesn’t learn at all. When you play the AI you are really playing against how well Frogboy could tell the AI in a precise step by step guide how to play.
3) The AI has to work within a memory and CPU cycle budget in a game which limits how much and how complex the evaluation it can do is, particularly in a game that can have > 100 of AI players.
You are right many people playing the beta are well above average strategy game players, and sadly this means they will never really be happy with any form of AI using todays techniques on today’s hardware, other games offer challenge by simply cheating the AI’s don’t really play by the same rules you do making it easier to programme a greater challenge.
Now the good news;
1) The feedback you are giving here on precisely what you don’t like about how the AI is playing is useful to Frogboy in improving the AI, though laying out these issues as you have and actually finding a way to have the AI solve them without cheating or causing other illogical behaviour are too completely different things and will take time.
2) Frogboy worked on GC2 AI for years after release and has already said he intends to continue that tradition with GC3. So AI will continually improve over the life span of the game.
Ultimately I would say personally that the AI is already better than that in many other strategy games I’ve played so recommending not buying the game on that basis seems odd.
One of the things in these kinds of posts is that people assume that AI doesn't already do common sense things. It does. It's just not necessarily balanced.
For example:
Most of these things aren't really AI questions but ones of judgment -- human AI data judgment.
Also, people posting about Beta 5 experiences, which didn't even have AI in it (other than random stuff) isn't really relevant now. Let's stick to Beta 6 or later.
I wrote a post yesterday saying much the same thing as Theeannon, the AI just isn't ready yet. It throws away ships, armed and unarmed, it doesn't fully develop even its home world after 200 turns, doesn't use adjacency bonuses and doesn't try to win. It's still building a shipyard first on most colonies. The AI is just not ready yet.
Larsonex, the only reason you are finding any difficulty is the bonuses a God-like AI gets and the time it takes to find them. Try a game on normal where the AI doesn't get those bonuses to find out how it really behaves.
Seilore, I'm not hating the AI is simply not ready for prime time yet and I don't want Stardock to be embarrassed by releasing the game too early.
In response to Frogboy's post, I'm writing about games I watched yesterday on the latest patch. Not building up its home world by turn 200 on an insane map, ran a soak on a small map that took 906 turns before a winner emerged. I know there's more to AI design than I'm seeing and it might be completely different next week but right now I'm not seeing the AI as ready for release.
I'm afraid you show your lack of knowledge of AI and machine learning here, Neural networks have been recently designed that can learn to play twitch bases arcade games very well. No one has yet managed to teach a neural network to do the sort of long term planning and memory tasks needed for a game like this and if they did it would be way to expensive in terms of compute cycles and memory to run such an AI for one yet alone one hundred AI opponents.
In fact the people who worked on getting their neural networks to play those arcade games have set their sights on trying to crack the issue of games that require strategy and planning next though I believe they will start on much simpler games than GC3. Their current system can only handle games that give immediate feedback as to if a move was right or not and I do mean very quick feedback.
AI in strategy games like this is very much not based on machine learning, it's based on figuring out precise instructions for the machine on how to evaluate everything in the game and select appropriate actions, to beat it you just have to out think the person who programmed it or their ability to tell the machine how to play the way they would. That's true of all strategy games of this kind.
It's also complicated for GC3 by the fact that so much of how the AI evaluates thing is data driven, this makes it easier to change but also more complicated to balance right.
What I like about the AI in this game is that you can create a custom race and give it a distinct set of priorities and personality when you play against it which the AI will take into account. For me it's not important the AI can beat the best human players because for this type of AI that's a pipe dream. What I want is AI's that play with personalities and I'm excited this game allows that,
That said the way the AI evaluates somethings in your examples can clearly be improved, and I'm sure will be over time, much of it could be improved by data changes.
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.
Can I just hold up my hands and say I’m one of those players who plays on beginner, and still find that the game has good enough AI in Beta6 to make my conquering the universe hard enough to be interesting without being hard enough that I lose most the time, which is what I really want.
I play for the experience not the competition and on normal the AI can often trounce me, because I suspect I’m too lazy to learn to play against harder opponents, I like to develop my empire the way I want to and not have to focus all the time on trying to stay alive.
One point of interest I recently completed a course in Machine Learning, and you won’t be seeing true learning AI in games like this for some years to come, because;
1) Someone has to figure out how to make a neural network deal with deferred feedback on the success of its actions before it can even start to figure out a complex strategy game like this. Essentially we need human level learning ability though on a smaller scale e.g. the domain of the game. Then we need to give it lots of experience of playing. Figuring out causality is a really hard issue in a game where you could take dozens of different actions per turn and some of them might not prove successful or disastrous for an indeterminate number of turns.
2) Once we have the algorithms for this maybe in a decade or so’s time we then need hardware that can handle dozens of AI’s running these algorithms and still have the computing power to run the games engine.
So basically you need to wait for the singularity.
There are many great features available to you once you register, including:
Sign in or Create Account
