Are you sure the AI doesn`t make weapons to counter yours or am I understanding wrong?

In my game, if I make missiles, it won`t be long before the AI comes out with ecm missile contermeasures that zap my missiles before they can do anything.

If I may chuck my 2c:

I think the absolute first priority of making singleplayer interesting (crucial for a game like GalCiv3) in strategy games, is to make the AI Fun to play against. I cannot stress that enough. I realise there are criticisms of the Civ5 AI, but I also think one thing Firaxis did pretty well was give the AI's a sense of personality in most layers of the gameplay. With the matches of GalCiv3 I have played, I think there is a bit of that going on, but could be improved. I will add, though, that I feel like I haven't played enough and still don't understand the game fully. As has been stated here, the AI's ability to execute moves on a local basis is hamstrung and I have definitely noticed that. I'm convinced that once the AI is capable of accounting for localised variables, it will become much more interesting to play against, even at low difficulties.

Also, I happen to think having a space Ghandi would be pretty freakin' cool.

What I really hate, and has ruined the re-playability of previous strategy games for me, are things that would likely be described as cheating/gratuitous handicaps. My best example is Star Wars: Empire at War, where the strategic layer of the game had some really cool, 4x-lite ideas. However, you would quickly find that careful manoeuvring of strategic assets and fleets would become extremely dull and repetitive when you discovered that the huge stack of death that you fended off would automatically spawn over the world it came from, and the AI would auto-spawn stacks of buildings and units on worlds it just conquered. This made the game a complete grind, and is a pretty gratuitous example of how to quickly kill off interest in future re-playability. I don't have a problem with giving an AI handicaps, but that has to come with interesting behaviours that are fun and satisfying to play against as well as avoiding the perception that AI doesn't adhere a cheesy different rule-set. I really hate that. 

I followed the development of GalCiv3 and picked it up at around launch in large part because of what the devs hoped to achieve with AI. Making it interesting and fun to play against and that the AI would be capable of 'learning' through collecting player data. Honestly, I think at this stage in GalCiv3' life cycle and hearing that it has had a very successful launch, I think the AI is definitely above average and I am looking forward to seeing future improvements. 

Yeah. I had the Krynn and Iconians waltzing warships and constructors throughout my ZoC (another thing that annoys me), colonizing worlds and capping resources they hadn't actually scouted. 

I know it's getting updated, but it feels really silly having your AI opponents getting pissy with you for having ships in their ZoC, but they would constantly waltz through your borders with the middle-finger raised. Last game I played, I just got so annoyed with them that I went about killing all of them off like house pests... Not even a Drengin.

In my latest game, I ended up being the aggressor, but I knew they were going to DOW me, so I did a pre-emptive attack (no, it was like I surrounded planets first, I just DOWed first).  Thought I was weak, so I ramped up my military very quick.  The AI didnt even have a lot of ships to try to figure out what they were in love with.  Well, their first ships that were "new" designs were in fact designed to counter mine.  Well played AI.  This may have happened simply because I was the aggressor, and built A TRUE war fleet first.  Perhaps they countered me. Thats the key, is that human and AI need to both "continue" to evolve.

In my previous post, the Yor simply NEVER adapted.  Several hundred missile boats over a few hundred turns.  They never had many defenses for my energy beam weapons, and continued to use Missiles even though I ONLY had missile defense on my ships.  Now Yor did have hundreds of ships compared to my 30 or so.  It might have been the AI just needed to "try" to use them and get killed before making any new ships (it did not "swarm" me with even half its ships).  Yes, refit is too expensive, but at some point you need to realize that what you have is WORTHLESS and either scrap or refit.  Sending them into the slaughter doesnt do much good, if you have historical info that this has been happening the last 100 turns. 

I would rather see something like it has 2x the portion of the total score for remaining factions.

So if there are 10 factions with a total score of 1000, then any faction of 200+ would get a negative (200/(1000/10)) = 2x and it would escalate from there. 

To be fair, the way the AI colonizes planets works somewhat like it's finding them (it appears to prioritize those it has 'seen'). Its... kinda opaque and I haven't figured out the mechanism yet, but it doesn't rush out colonizers to all planets everywhere on the map as if it knows where they are.

Frogboy, could you please confirm or deny the assumption above that CPU cycles is a limiting factor?

With computer chess opponents, for example, you can simply set the search-depth to N+1 to get a more challenging game, but computational cost increases exponentially with every extra move ahead the AI is asked to plan. Does the GalCiv3 AI engine have this property, or does it simply lack things to do that are computationally expensive compared with available time? (I know map size increases computational cost, but do you have ways to "think harder"?).

I would like to suggest that this is an important issue. What you asked for in your original post is, frankly, bug reports. That's great, and I welcome the clarity in what would constitute a useful bug report on the AI subsystem, but that's only one part of improving the AI. There's also the question of higher-level design issues.

Incidentally, I play on Challenging and find it a challenge. That probably puts me above much of the player base, but below the skill of most forum jockeys. But I can certainly discern AI behaviour that 'undermines' the feeling of the game for me even at this level: for example, an AI that adapted more flexibly in its ship design (building shields and dropping armour when I've got beams) would feel much smarter than one that simply has a better production bonus. 

You said that 90% of players play on Normal or below. Firstly, if that's the maths then thank you for even bothering to engage with the rest of us. That figure demonstrates how vastly unrepresentative the forums are of the player base. Having said that, forum players are your biggest fans and advocates. So one of the things you need is ways to improve what forum players perceive as good strategy which are cheap to implement (in terms of your own investment). This is why I'm asking about CPU cycles.

I'm going to go out on a limb and speculate that while most players play on Normal or Easy, most players also have modest CPUs, but the forum jockeys, as well as playing on higher difficulty modes, also tend to have more powerful CPUs. And if they don't, they would be willing to upgrade if you give them a reason to do so.

Most players don't just play on Normal or Easy; they also have crappy graphics cards. But the biggest fans have ninja cards. That's a way of transferring the cost of good graphics from expensive-to-dev-but-cheap-to-customer to cheap-for-dev-but-expensive-for-customer. The archetypal example is anti-aliasing: really quite cheap to implement (at least compared to creating another tier of high-res textures) but very computationally expensive. So players can be told "if you want better graphics than X, go buy a better gfx card and turn the settings up to 11". Most players don't; but the biggest fans spend $1000 on a gfx card, and wax lyrical about how Crysis is the best ever...

What you want to be looking for is a way to make the same commercial dynamic work for you in the AI subsystem. Also, you've got Moore's Law working for you: what may have seemed expensive at launch quickly becomes normal.

So we're looking for techniques that are easy to implement and give a noticable change to how the AI feels, even if the cost is more computationally expensive than you previously thought feasible (within reason).

Let me suggest a couple of ways this approach could result in a change to the feel of the game, for the better:

* you could split the CPU-cycle budgets between the AI unevenly. So (for example) Drengin could be given a low CPU-cycle budget, but a big positive handicap in production, and they will feel like a big dumb juggenaught; other races the opposite

* you could open up new decision subsystems that are computationally costly but allow more sophisticated player interaction. For example, instead of just weighting tech value for trade by research point cost, have each opponent calculate their current build-order, and weight value by how far a tech helps them reach the end - or calculate current and alternative build-order strategies, and weight by whether this allows a new strategy.

Of course, if only 10% of the player base is using these, then you're not going to want to invest heavily: but again, I suggest, implementing without optimising is OK for 10%; just ask them to throw money/CPU at the problem.

Anyway, I don't know if it is feasible to trade off better perceived AI for a requirement for faster CPUs. The main point of this post is to prompt that line of enquiry, and to suggest that both for the satisfaction on your fans, and for efficient use of your own limited resources, it would be a good idea to assume that there is a good correlation between the most demanding customers and those who are willing to buy CPU power if you can find ways to use it.

I would like to see the number of players per difficulty and how much TIME they play. I would guess a large portion of players who play at normal or below also play less overall time than those at higher levels. The higher level players are also probably veterans, play larger maps and play longer. It would be a good stat(s) to know.

And turn the game into a stat padding excerise?  No thanks.  I know stardock gets ALL the stats they need already, and thats whats important.

Looking at the dates of the last two posts, this seems to be a thread that was dead, so I'm not sure if replying to it will have any value.

My AI programming is very limited.  I am not a programmer, and do not have a background in programming.  But I have programmed AI for NWN for many years, if that lends me any credibility.  And I am a frequent player/modder of Civ III if that, too, will lend me any credibility.

Off the top, I'm really quite happy with where GC3 stands right now, my opinion including the map editor and AI.  I more fear downgrades from added content than I look forward to upgrades.  Does the AI make bad decisions about where it places improvements on planets?  Yea, but not game breaking.  Particularly not game breaking at the 90% levels I imagine, although I've never played them.  So what then is the AI discussion really about?

Firstly, you, Frogboy I mean, are hesitant to use cheats to give the illusion of a stronger AI.  Secondly, it seems the desire for improved AI is to make the AI into a tougher opponent and to prevent the AI from taking actions that are just patently foolish to most players.  Okay, we have our goals.

Now we need our perspective.  From what perspective will we be regarding our efforts and determining whether or not our efforts have pushed us closer to success?  From the player's?  Or from the designer's?  My experience in NWN has taught me that the best mods are mods that DO NOT give the player everything the player wants, mostly because the players are of the type that want immediate satisfaction and don't realize that too much immediate satisfaction actually breaks down those elements of the game that make it fun - namely challenge/reward.  However, that particular of the discussion seems almost irrelevant here, because both the developer and the players want the same thing: an AI that models as much as possible the behavior of the best players.  Or at least creates the illusion of doing so while having a similar result.

So, keeping the goals and perspective in mind, I thought to myself, 'how would I go about doing this?'  In order to answer that, I first had to break down my gameplay.  Immediately I saw the first hurdle - my gameplay never changes.  The only element that would change my gameplay in a significant way is the layout of the map, more specifically, the answer to the question - how close is my nearest opponent?

Aaah, but there is second major factor that SHOULD impact my gameplay, but actually doesn't, and I have almost fallen into a trap.  That is because I don't play with any victory conditions.  My ideal game is an unending one.  For most people, and most certainly the AI, victory conditions are going to determine gameplay.  Some players let their path to victory evolve as the game goes on, some decide what victory condition to aim for based on the map, and some predetermine the victory they are after before the map has been loaded.

So on turn 1, since we can equate gameplay to strategy, the AI should of course already have a desired victory condition.  To do this, the AI needs a way of assigning values to the map, and determining which type of victory condition may be best.  If it were me, since we do not want an AI faction playing the game without a chosen victory condition, I would also program in an equation that could at some point inform the AI that its chosen victory condition is not going to work, and it needs to pick a new one.  That equation is not only important in its existence, but in its timing as well.  If that equation is already coded into the AI, then simply re-timing it so that the AI changes strategies sooner will go a long way towards making the AI more challenging and realistic.  And, again, if it were me, I would include 2 variables for that timer: a relatively small random one, and a relatively large one based on the personality of the faction.  Smarter factions, and higher difficulty levels for that matter, are going to change strategies sooner.

I personally also think that the chosen victory condition should be affected by the faction's traits and personality.

Now, once the faction has chosen its victory condition (VC), how will it achieve that goal?  It should never, never put 100% of its resources into it.  A player never would.  So how much?  Again, it depends on the map, and the largest factor is the projected power of other factions.  Military power firstly, but as a fairly decent player, once I have determined that I am not in range of an invasion, the largest 'fear' factor for me is not military, but tech.  I am far more concerned about other civs being ahead in techs than I am about being in last place with military strength.  What should be a significant factor, but rarely is, is whether or not I am behind in colonies.  THAT indicates to me that ALL of the AIs need to be expanding faster.  So... military, tech, and expansion.  Of the three, its pretty clear right now that the AI focuses on military the most, and that should not be the case, because a strong military is the weakest asset of the three to have, until enemy warships are in your ZOC.  So, regardless of VC, I would reprogram the AI to focus less on military and more on techs and expansion.

Before continuing, I point out that strategy and tactics are two entirely different things, as Frogboy has noted as well.

I am not saying I would program the AI to ignore military.  But investing so much into a pretty crappy military early on has a huge opportunity cost from which no recovery is possible.  With the exception of range.  If I were playing a human opponent, and we both started out in the same cluster or near enough, the first thing I would research is military invasion and rudimentary weapons.  If no other faction is in range, I wouldn't research military for quite some time.  So obviously range needs to programmed in as a factor regarding how many resources to put into military build up.

And yet, a malevolent faction with a conquest VC SHOULD pursue a strong military.  The secondary problem comes in when they have the armada, but no range.  So somewhere there needs to be a line or two of code that tells them to build some constructors to close the distance between themselves and their chosen victim ASAP.  Otherwise they are "all dressed up with no place to go."  Somebody posted above that they hate it when they see AI constructors that have traversed half the map, it doesn't seem realistic, and so forth.  I beg to differ.  Sending out constructors is an action that the AI rarely does, and this gives humans a HUGE advantage.  I start sending out constructors very early on, and even before my expansion stage ends, on an insane map, I can target any other faction in the galaxy, and take any resource I want.  And trade with anyone.

So I would also program in constructor behavior, again with a random factor and personality check, and let fly the constructors across the galaxy. 

Another strategic factor is the use of credits.  Aside from the beginning of the game, where I rush maybe a factory or two, and colony ships, the only thing I use credits for is buying techs.  I don't start upgrading survey vessels until I have upwards of 50k credits, or so.  It is rare rare rare that I upgrade a military ship.  Why not upgrade military ships?  Its easier to produce them, the cost-benefit ratio doesn't pan out.  So why does the AI upgrade so often (and as someone said, particularly when an enemy fleet is so close?)

On tactics:

I don't know if will actually post this reply.  From a programming perspective, tactics is more about making a list of specific things the AI should do and could do.  Ironically, though, while I am unsure whether I will post this, I think its more what Frogboy is looking for than a discussion of strategy.

Maybe a new thread simply called "Tactics" should be started where people can list tactics they use, or bad AI tactics.

One thing that comes to mind right off, as I mentioned above, is the use of constructors.  The AI only seems to use them for resources or planetary bonuses, not for expanding range.

Hmm, maybe I'll go ahead and start that thread.

Lots of posts here so not sure if it's been mentioned, but

AI Declares war when it's not even near you.

Often I get declared on war from Civilizations which I can't even find on the map, which means they are at least more than one civilization away from my border.

Yesterday I had a game where two civilizations declared on me and I could not find them. One civilization I found and they were two(2) civs between my civ and the attacking civ. Later (read after I defeated the other civ) I found the other civ and there were 3 civs still between us and was more than 150 hexes away to their border, let alone starbase or planet/shipyard.

I assume the AI probably spent resources trying to deal with me and could never reach me with the exception of a long range scout.

So my suggestion (which should be a fairly simple check) would be an AI civ should never declare war on another civ which is not within a certain striking range. Take their average war ship (or life support level) and double/triple that range (assume dropping two starbases inbetween) to get to the civ they declared war on. In later games if they have life support researched that range will clearly inclrease, but if they are not spending science to increase life support they can't hope to reach me, yet I can easily build up my fleet, then drop a few bases on the way and take over a planet. Once that's done I usually would have enough defense to protect that planet, so it's a lost cause for them.

As you mentioned I guess you could just give all AI  an handicap of lifesupport range, but why have it in the first place if all AI needs it, giving them more scienes does not mean they will research life support so no help there. It would be simpler to just reduce how far a way a civ can be before it's too far to declare war on.