[Balance] Player vs AI power progression in FE (and all 4x games in general)

CivIV is the best late game AI I've seen as well, but it came nowhere close to holding its own in mid game, let alone late game.  At the highest difficulty, its military could be problem, but only because there is no room for smarts between two stacks of doom.

AI is nowhere near as good as a human being at complex situations, and there's no reason to believe that this will change any time soon.  There has been no revolutionary progress in AI in the last 20-30 years.  We have made plenty of evolutionary advances, due to more processing power, more memory, advances in data structures and algorithms, etc...  But AI can't challenge me in Go, so I do not expect it to challenge me at a fair game in Fallen Enchantress.

Fallen Enchantress can get a lot better, but it will not be because the AI can beat a player fairly.  It will get better when Stardock gets rid of stupid diplomatic actions, fine tunes production/rushing bonuses, and incorporates strategies for specific conditions.

But mostly, AI will get better when you have a more powerful PC, and the AI has more complex algorithms.  What I do is just make up home rules, set up enemy teams, and avoid the most powerful synergies for my side.

Right now I most enjoy a 1-2-6 distribution of sides (player/aligned/opposites) with a custom AI difficulty (5x cash, 3 x production, and 2 x research and the rest Ridiculous) on a Ridiculous large map. I stay away from beast lord, wealthy, stealth, the quest loop, elusive troops, and mass killing magic.  The result works for me.... I win most, I lose some, and the games last about 150 turns, which is enough to keep me interested.

And seriously, you should not bother going to the Victory screen.  One seldom plays chess to a mate... drop a game when it's no longer fun.

Well actually, GalCiv 2 does like to throw a spanner in the works now and then.

But yeah, it is an intrinsic fault of the genre.  When you hit all the four Xs you cement your position at the top of the leaderboard and nothing can touch you.  It is true that the AI could play better, but realistically in order to do that we need to pattern more human-style strategies in the AI.  We have to let it take calculated risks, even if they fail.

I'll give you an example.  In my last game, I saw my opponent building up nearby.  I saw this is an opportunity for a decisive first strike.  I readied my fortress to crank out soldiers, built a road up to the closest enemy city, and sent my first army up to take it.

Now although I took that first city without any casualties, the next one was not so easy and I lost a unit.  But that was okay, I had reinforcements on the way.

The problem that I ran into was that the enemy had more cities cranking units than I could have anticipated.  Even though they were armed with clubs, there was no way I could fight against the sheer quantity of them and gain ground, especially not against the enemy sovereign who'd got himself a Sand Golem and an Ophidian.  I gave up, knowing that the deck was stacked against me, and started a new game.

I wasn't wrong to take that first strike opportunity.  I had no way of knowing why the enemy was so powerful based on the faction power rating.  But I could have built up a greater quantity of troops before doing so.  That might have bought me some time and given me the ability to fend off the enemy sovereign.

Similarly, the AI needs to ignore for the moment the player's faction power rating and consider that you've left a tasty morsel of an undefended settlement on their doorstep.  That they can go in, take your settlement, and destroy all your buildings before abandoning it.  So then it doesn't matter if you send your armies in to retake it, you have to build it up again and chase after their fleeing army to exact retribution.

Would it be too difficult/too "random" to have the game do a check which says "if human player x% higher faction rating (yes, I now faction rating is weird, but whatever metric is best), then have y% random event happen to weaken human/strengthen AI"....stuff like famine hits, or cities/units xzy revolt and join enemy, or  AI reinforcements arrive from overseas allies, or something...so game gets harder as you start to win, rather than easier?

Would this make game more fun for those who gte bored at the late game, or more annoyingly random?

Of course, to beat the ineviatable response...set it up as an option for the player to choose!!

I suppose the answer to that is, are these events preventable or do you have any warning that they're about happen?  Like global warming in Civ.  Or, do they just happen and you're like, "Damn it, now I have to fight my way back out of the Stone Age?  I'm just going to start a new game."

That's more of a rubberband style which has proven to be nice on paper but a failure in execution. It's arbitrary and not "permanent" since it's a bunch of band-aids on a gushing wound. 

I know that AI takes a lot of processing power, but can large AI factions, or "end game" AI factions delegate control? As in, split them up so each AI city or delegation can react to its particular circumstances? Seems like the AI has a one-size-fits-all brain that does not get any smarter or dumber, and thus, gets dumber as its body gets bigger.

I would be interested to hear from an experienced programmer, or a Stardock employee as to their thoughts on this inherent balance issue with 4X games, and if there's any resources put into addressing it. It is possible that they are simply satisfied with shorter games and replay value. 

I find your post to be quite accurate in describing a fundamental flaw with the genre itself. Any game that has 4X elements will fall victim to human ingenuity. Even more so if you have a save/load feature. Unless of course you ironman it, but even that doesn't always provide adequate challenge. Once I've reached a stable situation within a given game, I usually just play on defense. As I already know that if I wanted, I could stroll into any territory and destroy them at my leisure. However, GalCiv2 allowed me to do something far more entertaining.. Minor factions. I built powerful warships, of the likes of which no other race could even hope to compete against, and sold them to the minor factions, sat back and watched with glee.

I like taking a Vorlon/Shadow approach to things. Be the most technologically advanced, sit back and just enjoy the show of the little-ones. At least until they've consumed one another to leave behind only a giant empire that may be worthy of a fight. But considering I'm the one who most likely supplied them with their weapons, technology, and economic might.. I've not lost. GalCiv2 is by far my favorite Stardock game. One of my favorites period.

I have actually. In an older 4X game, another great favorite of mine.. Star Trek: Birth of the Federation. Now, for full disclosure, had I been playing as the Romulans, this wouldn't have even been a remote issue.. But, I was playing as the Federation at the time. I had an armada of 150 ships, completely dominating the other races, enforcing peace and all that. Then during a late phase in the game, it happened... They came.

To my horror.. They wiped out the only other faction that was even remotely able to fight me, the Klingons. They were almost the same size as me in terms of territory, but had vastly more ships than I. Though fewer in number, I had superior firepower. But anyway, I got a message saying the Borg had arrived. I figured I'd have no problems, as I had Sovereigns, Galaxy class warships, you name it, I had it. Well.. It wasn't long before the Klingons stopped sending in fleets, then I went looking for a fight. What I found scarred me. Klingon world after world.. All Borg now. Then I found it. Some scout ships discovered it approaching my own space. So I sent my entire fleet, all 158 ships. Battle phase! The opening salvo from the Borg cube destroyed 90-100 ships. Instantly. But it also took heavy damage from quantum/photon torpedoes. Battle phase 2.. The Borg were destroyed. I may have won the battle, but I as good as lost the war. I had 6 ships remaining. And some very eager Romulan and Cardassian "friends" on my boarders that now knew I was weak. I was able to win the game after a MUCH longer length than would have required before hand. But it was hell. It was my own Wolf 359.

Thanks for sharing your story. GalCiv2 attempted that with the Dread Lords, but they never once posed any challege. The only random event that even made me flinch (after reaching critical mass) was the jagged knife, boy do those guys suck.

Here is my attempted restatement of some of the points raised here:

I see two fundamental flaws with AI: Geometry and choice of goals.

Geometry is a computational problem - there are many possibilities and permutations involved in dealing with geometry and to deal with those issues using brute force calculations becomes expensive and inefficient -- you get into "consider all combinations" issues.

The ability to choose a relevant goal and then assign priorities is another take on this kind of difficulty -- an AI's decisions have to be made using a mixture of pre-determination (long before the board is populated) and approximations.  Anyways, once you understand how the AI makes its decisions you can "game" it, deceiving it into picking a bad goal.

Of course, if the computer does not have to have good goals (for example, when it can cheaply churn out juggernauts with 700-ish hp and 100-ish attack each, faster than you can build scouts) -- sometimes, some of these issues do not matter.

All the discussion so far has been about the limitations of AI, but don't forget that this problem exists in all-human games as well.  When playing Risk, Shogun, Axis & Allies, etc. with my friends back in the day, it was invariably the case that the game started out difficult as you struggled to consolidate your power, often in the face of multiple nearby opponents, but, once one player or team reached the point of being able to effectively defend their position against all immediate threats, then the game was pretty much over unless everyone else ganged up on them.  Otherwise, they could build up for as long as they like in that defensible position before going out to roflstomp the rest of the players.

Which is not to say that I don't see a significant problem with 4X AI, but it's a more specific problem:  When the time comes to wage war, the AI simply sucks at it.  I'm fairly new to FE and just finished my first game of it last night, but it went pretty much like every 4X does in my experience.  I started out kind of muddling around (like I said, it was my first game of FE and I didn't really know the game's details) and, when I finally made contact with the sole AI player, he had about a 50% lead in faction power.  I continued exploring, building, and researching until he finally declared war on me, by which point his faction power was a little less than twice mine and he started throwing armies of juggernauts at me.  Within five turns or so, he'd lost three cities and 75% of his faction power to my single army of three heroes and some lesser troops (I didn't know about the XP division at that point).  I, on the other hand, hadn't lost so much as a single unit.  The AI later retook one of those cities when I decided to ignore him and take my hero stack off to clear wildlands instead, but that was my only loss to him.

4X AIs consistently base their diplomatic decisions on power ratings which highly overrate the quantity of military forces, while undervaluing the quality of those forces and undervaluing or completely ignoring force mobility (my hero stack is on the wrong side of the world, but I have Cloud Walk and a ton of mana, so it doesn't matter that they're not hanging out on your border) and production capacity (your obsolete troops may have me outnumbered 10-to-1 now, but I just need to stall you for a couple turns and then I'll be producing several new, superior-quality units per turn).  And then, when war does break out, they're typically too timid strategically (take one city, then sit in it for another 20 turns instead of grabbing the two undefended cities right next to it) and utterly incompetent tactically ("Oh, your high-initiative mages and archers just shot down my really-expensive-but-slow juggernaut stack without taking any losses?  I think I'll attack them with another juggernaut stack!") to have a chance against even a moderately-established human player.

I've actually wondered why everyone keeps the AI bonuses linear. Why not make them get larger for the late game etc. with a "roof" to prevent it from breaking the game, you can even let some player triggers make them bigger. I don't want ridicilously powerful AI early, I want it late when my empire is in shape. 

Seems to me the major problem with this would be solved with only going away from linear bonuses. (and some tweaking to the AI sovereigns and armies)

I don't remember Star Trek: Birth of the Federation being that good. 

As someone that ignores about AIs, coding and all that, I sometimes wonderered about exactly that. To provide a constant challenge, without breaking the balance, I guess it must be hard to program. But with experience, testers, and many feedbacks, I think that could be a way to provide the needed challenge for the late game (or post-game, if you wish).

It is very clear for me that the most difficulties are in the early game, because no techs, no army, no nothing. And the most fun is the mid game, because with a bit of them, I begin to say: Yes, I'm here.  Poor AIs do not provide challenge, and strong AIs steamroll at this point. Balance here is difficult. I normally try to stay good with many AIs to avoid early wars.

But once in the late game, the player is powerfull and the AI just can't use the hundreds of gildar and techs, so it offers no challenge. This is the point where the AIs need more help (cheats or whatever), but mainly, a way to channel that, a target, a strategy to deploy armies while defending their lands.

Diplomacy should play here an important role, trying to reinforce alliances against the strongest faction and things like that...

You don't have to remember it being good, as I was telling a story from my experience.

I do not think it is merely a limitation on programming, but more an issue of processing power. To have more complex AI means you need more powerful CPUs to do that. I'm quite certain that if Brad wanted to spend a few hundred man-hours to create a beast of an AI that would take a top rig computer today a few minutes to run simulations on possible outcomes of this and that and then make the best choice based on the human player's situational actions, he could. But I think that'd not be so fun having a game that takes 10 minutes between turns. As a note, all numbers and estimates are conjecture. He may be able to do it faster.. Or slower.

With current technological limitations and lack of pure processing power for situational events (and accounting for potential random occurrences) it is just not practical to create an AI capable of giving humans a real challenge in 4X, aside from bonus stats to their base. While it may be out of the question for the genre right now, you can however get plenty of challenge in chess. If you want to play the AI at it's own game. Deep Blue has nothing on what is out there now.

You know what is fun about this discussion? Derek Paxton aka Kael when he did his FFH supermod for Civ4 did both mechanisms we are talking about. Now, perhaps they were not fine tuned correctly, but they did exist.

The Armaggedon counter when it reached 90 or more unleashed really bad things upon the world. This was in essence a mega event to challenge your late game. The Demonic faction too was also a kind of mega event. I think this was the right direction to maintain interest, even if the balance was not subtle enough to challenge the player most of the time in a proper way.

Also, there was a game option called 'flexible difficulty' that added one notch in the AI if you were in the top 3 I believe, and this test triggered every 50 turns. Basically, if you did good, the AI changed from 'emperor' to 'immortal' for example, then to Deity (and I created further difficulty levels because it was not enough, like Greater Deity and then Pancreator, haha).

So mechanisms existed,  they could have been developed further in FE. Perhaps in a patch?

No you don't understand That sounded good. Did they have some sort of major patch or expansion. I only gotten the vanilla game and it was horrible. If they did put out a hugh patch or expansion I wanted to try to track it down and give it another try.

Not all AI's are equal but AI can never be stronger than the weakest link between the maker's programming skill, playing skill, and motivation focused on AI development.  It's rare to find those three attributes in one person, and it has to be within one person to work.

--An awesome programmer can't make algorithims at a world class playing level if he doesn't understand how to play world class at an instinctive level, not just hear someone talk about it. Even with a world class player feeding them info too much is lost in translation.  In all things, it takes very little change to reduce great to average.  Your typical great programmer is an average player. He went into programming because it's his calling, not gaming. 

--An awesome player can't make algorithims at a world class level. They might understand how to play great but if they can't program at that same level their thinking doesn't translate to the algorithim.  Once again great gets reduced to average.  Frankly, most people talented in anything rarely understand themselves why they can do what they do. That makes it all the harder to put it to paper. In sports, very few great players make great coaches, and they all are focused on the same thing! (focused on their sport).

--The final component kills off those rare, very rare few that could get past the first two hurdles.  To play at an extreme level AND program just as skillfully are two talents that could easily take up all of one's time individually.  Someone that good at programming AND playing (and manages to not have one talent comsume them) would have a skill set that could be used in so many directions it's unlikely they are inspired to put those talents to work in game AI (more likely to be in real life AI research and aplications).

There certainly are some people out there with a decent amount of both skills required. This gets proven each time a player community improves upon a developer's initial product.  "Decent" isn't "great" though so we just get improvements on "average" to "slightly above average."  There won't be great AI unless someone with the right combination decides to take their talents to "slum" at a game programmer wage.  It could happen at some hungry start up Indie but more likely that awesome AI will come from outside gaming and get ported in at the point when the average programmer can copy it.

Well you're not wrong about the combination of those talents being rare, or that someone with those talents could go in so many directions.  But circumstances often dictate where people end up, or don't end up as the case may be.  It's not always about how much money is available for doing some other job.

I would actually love to design game AIs but I'm not qualified to do so.  Besides, nothing is worse for game sales than lots of customers complaining that the AI is 'too evil.'

I have just completed my first graduate course in Intelligent Systems, and this is my take on how hard it would be to make the FE AI good in the strong human player sense of the word. This doesn't make me an expert, but I have some perspective on what the problem looks like from a machine learning perspective.

First, if you could make the FE AI as good as a strong human player in a game development cycle timeframe, I firmly believe you would be doing so by passing up the opportunity to earn at least the gross sales of Fallen Enchantress for the same timeframe elsewhere. The problem is that hard from a logical standpoint, and it doesn't matter if you have 10 orders of magnitude more processing power. 

This is because the game has far, far more scope than Go. In Go you can do one thing, place a stone on a board (albeit a large one), and the best humans still do it better. Here is a short list of elements that FE has that Go does not: hidden information, stochastic setup of the world and random events, the scope to make dozens of moves within a single turn, multiple opponents, different kinds of "pieces" (cities, units, spells, monsters), and the complexity grows as the game goes on instead of diminishing. This means any solution would, unlike any Go solution, have to be learning as it playing. You can't design a static AI solution for something with so many dynamic factors. This is why there is always a strategy in these types of games (usually dubbed an exploit) that the AI cannot handle that works every time. I don't know this for sure, but I highly doubt any strategy game has ever implemented any online learning. This is a dynamic element that is computationally expensive (which is why I doubt anyone has done it), but even more to the point, even harder to do well. For perspective, at my university they had a project that did diistributed document filtering that learned as it operated (with a cool multi-agent economic model) and it took ten years to develop. I don't think that problem was nearly as hard as FE AI (though far more important).

I find it so much of a harder problem that I believe if it was widely studied as a problem academically for a couple of decades (though it is difficult to account for the evolution of intelligent systems methodology), then I find it conceivable it would have Go's current status, an AI that could beat everyone who isn't a serious, committed strategy player. This wouldn't happen, but I think it is hard to appreciate just how difficult the problem is. It is why so many other games turn to just making an AI that looks like it is a good human player, but isn't playing the game at all.

@Tiokon, while I think you underestimate the challenge in designing and executing an AI in a game as complex and dynamic as FE, you have a general point that is worth discussing.

The person designing the AI needs to be (or have access to) a masterful player of the game.  Otherwise, no matter how keen their AI skills, they face a real struggle to design an AI that plays masterfully.  That however, is not a problem, thanks to player feedback on these forums and things like the "Let's Play" videos, Brad and co. can see top players employing the most successful strategies rather than trying to figure those strategies out for themselves.

On the other hand, since much of the game of FE is focused on handling and defeating the AI's tactics, you could say that no amount of mastery would allow you to make a masterful AI.  Once you release it to the wild and hoards of players have at your AI, strategies will emerge that are specific to defeating your AI no matter how good it is at the game.

Fortunately for everyone involved (except the family's of the programmers) is that the game's AI is not a static thing, but rather improves iteratively with each release.

Right now, I feel like the we're just emerging from the initial post-release bug fixing stage (getting rid of crashes and freezes and "the AI can't see that kind of improvement" or "the AI doesn't understand the significance of the defense stat" bugs).  We're entering the stage where they are balancing the AI so once it has the correct information it can make semi-intelligent decisions.  This would be the part where the mid-game and later AI gets a little more wooly.

I don't think we'll ever get to the point where the AI is a mind-bending challenge for Tuidjy or the other top players here.  They'll always have to handicap themselves in some way to keep themselves interested. 

The challenge really is that the complexity of the game is so great by the mid-to-end game that you're asking the AI to perform some fairly sophisticated deductive reasoning just to get a clear understanding of the world around it.

Putting aside the incredible complexity of getting a machine to understand (especially with in a limited number cycles) something.  There's the added problem of trying to get a machine to anticipate something.  Creativity and imagination are the elements that are most important to the mid and late game, when you're no longer playing against the static map, but now competing directly with dynamic opponents (at least the player is dynamic). 

If deduction is hard for an AI, creativity and imagination are essentially impossible.  And this is usually the phase where the developer substitutes his own creativity instead, (which has its own pitfalls -- like how do you describe a dynamic situation so that the AI will recognize when a specific strategy is called for).  This inevitably falls prey to the above issue that the players quickly learn the hardcoded strategies and adapt their own strategies to work against them. At best it's a losing battle for the developer. 

And it should be.  I don't expect Stardock to be at the bleeding edge of artificial intelligence research and I wouldn't expect them to produce a miraculous breakthrough in intelligence while running a background thread on my commodity desktop machine.

There are ways to tackle the problem of making the end game more interesting without programming consciousness into the AI.  Random events that throw the world into disarray is one.  But giving the AI a simple understanding of diplomacy would add to the game in boundless ways. 

(TL;DR -- here's my suggestions:)

1. Teach the AI to gang up against the top players would be nice.  Take the case where two AI factions both border a faction that is more powerful than either of them, but not as powerful as both of them.  If the two border AIs both declare war, they would be able to crush the third faction and build their own strength accordingly.

2. Teach the AI to gang up against aggressive factions.  A mid-strength empire that has gained its strength through conquest of its neighbors is not to be trusted.  Its in the more powerful factions interest to ensure that the upstart stays in its place.

To implement these ideas (and others) properly you'd need to move beyond the simple +1/-1 relation calculators and allow the AI to formulate diplomatic strategy the way it formulates military strategy and expansion.

I tend to disagree that the person working on the AI needs to be especially good at playing the game.  They should have played a variety of games and know all the various cheese tactics, but not necessarily play that way all the time.  Rather, they should play in a way that can be mimicked by an AI.  It's a lot easier to implement an algorithm that plays a particular strategy when you've been inside it, walked through each step.

Once the AI has these strategies at its disposal it will be a lot harder for a human opponent to get one-up on them using the same old tricks, More to the point, when the AI leaves that city of yours that it captured unguarded, you won't be rubbing your hands with glee, you'll be paranoid that his army is just over the hill out of sight, waiting for you to take the city back.

That depends on my sovereign!

The next sovereign I play will have several goals in play when taking a city, one of those goals will involve turning certain cities into crippled, abandoned husks.  (The game mechanics have been in place to make this viable, and I need to see how well it works.)

I know you say the AI is bad at the late game, but I think you're over-estimating how good the AI's early game is. From my experience, AIs in 4X games are good at one thing and one thing only: fast expansion. Fast expansion gives them an immediate advantage right out of the starting gate, as having more cities and more territory usually means better economy and better research. But for some reason, the AI is never able to take advantage of that. What invariably happens (assuming you're any good at the game) is that you start to catch up with the AI, and then you pull ahead. And because 4X games operate on a massive slippery slope, once you've pulled ahead, it's really difficult for anyone to catch up.

And I think that's pretty much the problem. In slippery slope games, once you pull ahead (barring some clever strategizing or alternate win conditions), the game is already decided. It's inherent in slippery slope games.

You can try to make the game challenging again by sticking a rubberband on it, but if you take it too far, you anger players for the opposite reason: early game feels like it doesn't matter anymore. (Interestingly, most 4X games do have rubberband effects that kick in once your empire grows too large, but no on one ever complains about them, so...) You can try making the power curve as flat as possible, so the slippery slope effect doesn't matter as much, but then players who play 4X games specifically to tech their way up the power curve feel cheated.

I remember playing that a decade ago. It was prolly the first 4X game I ever played, before I knew what a 4X game was. Has it aged well?

Okay here's a theoretical situation, let's say you had two equally-skilled human players playing this game, do you think that the player who suffers the first defeat at the hands of the other could pull it back and win?

Sweatyboatman I think we are in agreement. I may have miscommunicated my point if it sounds like I think creating a great player level AI is going to happen.  I think it would be well nigh impossible short of artifical sentience.  I agree with all that you said except that I think it is far harder for a programmer to program something they see on Utube if they don't play at that level themselves.  It's one thing to see someone use a strategy but it's another thing to know when to use it and how to bring about the same results.  If it were that easy all average players would play at extremely high levels after a few videos.  For that matter all sports teams would be champions instead of the same few usually being elite and the rest being talented but never seeming to get a clue.  This is true at all levels of competition whether money is involved or not.

"Seeing and copying strategies" CAN work to simulate a good AI to a degree in hidden games when MANY of those strategies are built into the AI and it could be using any of those strategies. Until the human opponent knows what "personality" it's facing it will keep them guessing.  Such "personalities" are still a glorified script but enough variation masks the lie.  Games with better AI's often do this.  Games with weak AI have more limited personalities, making it easy for all but sub par players to get bored with them.  Creating many personalities may take more time than developers often have but it really comes down to how strong of a game they are ok with making.  It is an area where player communities can, and sometimes do, help add a lot to a game when allowed to.

A really sophisticated AI can have stronger mid and/or late game evaluation processes to help it choose and switch to a sub personalitiy that once again can keep the human player guessing.  Current AI's (in some games more than others) do attempt to do this at a very basic level but most developers just don't have the time, ambition, and/or right set of skills to take this to the level any player slightly above average finds satisfying for very long.  As some have mentioned, a few games have gone the extra mile with AI's, just not many.

The true problem is that we are making excuses for something that shouldn't be excused.  "Dev's don't have enough time" is making an excuse.  Solo play is the industry standard.  Solo play requires AI.  The 4x genre isn't suited to multi player on a regular basis.  Perhaps some sort of organized and regulated ladder could help multi player grow but it will never be the main way 4x games are played.

A game won't have any degree of replayability beyond the strength of it's AI no matter how cool the theme or game mechanics are.  If a developer wants to make a game above "average" and become a game industry leader they need a strong AI to help them get there.  Like the teams that don't get into the championship, everything else is just excuses.  It comes down to whether the goal is to be excellent, or average.

I grew up with some guys that develop games for a living at a well known game company.  They have actually moved around to various companies since then as projects finish and as is often the case with an industry in a constant state of growing and dying at the same time.  I used to play games online with them WHILE they were at work.  When I asked if they would get in trouble not not working they replied that actually they needed to go because the company Space Empires games was about to start.  In this particular example two entire games of Space Empires were played by all company members during "work."  If you know that game you know each game can take weeks in a large multi player.  They didn't play a thrid time because a deadline was approaching in a few weeks. (NOTE: The game play wasn't "research." They weren't developing a game related to Space Empires in any way).

They've described the same "development process" at all these companies.  Months of screwing around with little effort put toward the game followed by a short period of "frantic" to make an artsy demo that has little to do with the real game but promises the world.  Then months of relaxed screwing around again with everyone saying they were waiting on someone else to get them something or other they needed, followed by a short period of frantic as final deadlines and release approach.  It used to be that a third period of screwing around would often take place when the developer convinced the publisher/backer that more time was needed to "do it right" and then a final round of frantic once most of the grace period was wasted and the final deadline truely approached.

Some would call it "the artistic process."  Definitely true that creativity can't just be clocked in when someone wants it to be but these guys say that very little effort was put to actual game making as compared to playing other games and doing just about anything other than working on the games during the screw around periods.  This process repeats over and over again and secrets get out.  It's no wonder the big publishers keep shortening the deadlines for their developers to put out a product.  Indie companies are often hungier and harder working, putting out much more creative products but often don't have the resources, manpower, management or complete set of skills needed to make the product they envision.  Brad, Derick, and company fall into the second category no doubt.   I don't think they are anything like these companies I described.  Those companies might have been large but they were barely aiming for average and rarely hit that high.  Brad and company have made some pretty darn good games that show a lot of creativity, effort, and skill.  Better than anything I saw come out of those larger companies.  If they want to get to "great"  though the weak spots would have to be shored up one way or another.  AI appears to be a weak spot that any company that takes it up a notch could make a name for themselves through.  It's not going to happen through the big companies.  Their publisher's only goals are the stock reports and their main investments now are on marketing that promises more than a presidential candidate and delivers less.  My money is still on Stardock and the Indies.