This report has been painful for me to write, and it's going to get worse before it gets better. Civ4 has solved many of the old problems, but the solutions generally have side effects. In some cases, the cure turns out to be worse than the original disease was.

I was part of developing this AI, and I'm extremely proud of what we achieved tactically. The AI handles its economy and tile improvements a lot better (and this with more complicated plot improvement options). Unit combat is also more complex and yet the AI does a decent job of producing a mix of units -- at least it doesn't get stuck on only one type of unit, though in most cases it only has two or three. There are so many little details that work well, especially compared to other AIs, including the Civ3 AI. Yet at the strategic level, I must conclude that we failed in most of our objectives. The one thing the AI does very well is play effectively at mid-range difficulty levels against relatively inexperienced or unskilled players, giving them a good and fun experience. We got it right for most of the customers. However, for the high end players, we missed the mark. In some ways, we missed it by a wide margin, too.

There is so much I would like to write on this topic. I'm constrained from speaking at all about the game's development, though. All I can discuss is where things ended up, and where they stand vs other games.

The trigger event here, of course, is that Gandhi declared on me at pretty much his first possible opportunity. He planted a city on my continent, then almost immediately gathered an attack force and came calling via the only way the Civ3 or Civ4 AI ever attacks: with a sneak attack aimed at the softest point it can find.

The AI did something clever here, right? ... No, not really. Tactically it's clever, but strategically it is vacant-minded. This game has set up the most extreme possible test case, with the least trustworthy aggressor and the most pacifistic builder, and both AIs fail the litmus test.

We have here the perfect metaphor for the strategic shortcomings of the Civ4 AI: a warmonger who won't attack an AI, and a peacenik who won't leave the human player in peace. It's a perfect storm, a total strategic disaster. The AI is hardwired to screw the human player -- not by design, but as emergent behavior from routines designed to do other things.

The Civ2 AI wasn't intentionally designed to screw the human player either, but it does. We've somehow fallen back in to that trap, as deeply as any game ever has, and it's going to undermine the legacy of Civ4 among the most hardcore single player fans unless the situation is improved before development comes to a halt.

The warring half of the game in single player is in bad shape. There's no POINT to capturing enemy cities in the early going in a standard situation. If you can take enemy capitals with an immediate Axe or Chariot or Warrior/Quechua rush, taking over a whole AI's territory and wiping out a rival, THAT will pay off, but almost anything else will only saddle you with territory and cities you can't afford to keep. You stagnate your economy and fail. Despite their economic advantages, the AIs are in the same boat. They CANNOT form huge early-game empires and thrive. They stagnate and die out, unable to continue to conquer rivals who have outteched them. The game is literally engineered to remove the total warfare option, leaving only late-game "mop up" style warfare vs technologically inferior opponents -- anything else and you run out of time, thanks to the AIs and their one strategic point of competence: the space race. Adding penalties to owning too many cities has taken away too much of the game. One monster has been replaced with another, while game balance remains painfully elusive.

The AIs don't have objectives. They have only behavior states. The only thing they do well is found cities and tech along, and even that is propped up (under the hood) with EXTREME handicaps. (As many have pointed out, even on Noble the AI gets huge economic discounts on unit and city maintenance, upgrade costs, etc). So that performance is a combination of real improvements to the AI and a huge band-aid meant to cover a host of sins and shortcomings.

I've concluded that the massive economic bonuses are a dead end in terms of design. The actual effect they have created is a dichotomy between AIs who cannot keep up in production and AIs who run away with the tech race. There is no sweet spot in the middle. Take alliances, for instance. You pretty much need tradeable techs to be able to bribe AIs to fight on your behalf -- even your so-called "friends", even when the fight is in their own interests. Yet unless you are locked out of the tech-trading picture and well behind your enemy in technology, you don't NEED any help. This AI only wins when it out-techs the target. It's like the cliche about banks loaning money: they are only willing to loan it to those who don't really need it. The Civ4 AI is only willing to help friends who aren't in any real trouble. "Fair weather friends" of the worst kind! At least with the Civ2 AI, there is no pretense of befriending you. They all openly and directly gang up on the leading civ, inevitably the player if they have even a modicum of skill, since the Civ2 AIs are so generaly incapable of managing their economies.

My intended scenario design for Epic Four is a total failure. I lined up all the pieces correctly, but the AI is simply not capable of the level of sophistication needed to cement them together. Since this turned out to be the perfect illustration of the AI's overall strategic flaws, I simply found the event too disheartening to play through.

I'm sure that if someone else had designed the scenario and I lacked any spoiler info, I could have accepted the situation "as is". I would have befriended Gandhi right away, not knowing how many other AIs were out there, and Gandhi at least would have left me alone. Qin, then deciding to attack me (if he did) would have made sense and fit with his personality, and I could have taken any beating he could manage to dish out. I would almost have expected it, and then at least tried to prepare. Instead I got cold-cocked by a two-faced Gandhi, and I just didn't care to continue. An irony, since the scenario description urges players to press on "no matter what". I can't even play out my own scenario!

Still, it's the right choice for me to have made. My original concept for the scenario was to lock all four AIs behind the peaks and shut off tech trading, but leave the space race enabled. Well, I forgot to shut off the tech trading totally by accident, the first time I produced the game, and then when I remade the map, I decided instead to go with locking away only two AIs. So instead of being on clock, trying to fend off the barbs before the OCC and 2CC AIs could launch, the player would have to tackle the AIs themselves, but have unlimited time to do it.

Depending on the reports of others, I may go back at some point and try my original idea. I certainly had a blast fighting the barbs. It was indeed a flashback to Civ3-style Always War. Perhaps this concept will be worth another game, or perhaps even several games. I'll explore that later.

I need to give up on the AI, though. It is not, in the end, an AI with several personalities. There are some tactical differences, most of which emerge from having different traits and UUs and different starting techs. Strategically, there is only a single AI in there, and there is very little the player can do at the diplomatic table to sway things. One of the biggest is to avoid having a state religion, or in rare cases, to choose your state religion carefully. One can try to avoid ticking off AIs who have a "worst enemy", but as seen here, that approach doesn't get you anywhere. AIs who are neutral toward you will treat you like an enemy, yet even enemies amongst AIs won't fight each other. As painful as it is, I need to give up on any scenarios that try to balance the AIs. It literally doesn't matter which AIs are on the board. In the end, they all play the same at the things that count.

I believe I have some of the answers for solving the problems, but they are all fairly work intensive. Unless gaming companies dramatically increase what they budget toward their game AI, then even people who have the right answers and are capable of implementing them won't have enough time to craft the solutions. There are also some paths down which most game AIs travel that are dead ends. More ambitious methods would have to be employed -- and of course, when breaking new ground, there are always side effects and unintended consequences. Things might have to get worse before they CAN get better, and that's not a formula that companies who live title to title can afford to engage. Thus there is practically ZERO "Game AI R&D" taking place. Civ4 is as close as anybody has come, as far as I can tell, and look how many things are still out of place.

For now, players are going to have to play with what they've got. For me, that's not a particularly appealing option. There are only so many times a Civ AI can backstab me for no apparent reason and in defiance of any logical in-game objective or motivation -- but instead merely the inevitable result of AI hardwiring -- before I roll my eyes and start looking for something else to play. There is still some life left in Civ4 for me, but it's isolated to certain pockets. The AI itself has worn threadbare to me already, because of certain strategic inevitabilities that occur without pause in every game -- and which are not particularly fun to try to play with. (Repetition is OK if the fun is the kind that doesn't wear off after a few uses, but few games manage that these days, and everybody's breaking point differs).

The Civ4 AI is well more advanced than any other in the genre. Where it goes from here, though, I don't know. If nothing useful happens before the expansion(s) run their course, then some day I may come back to it and try my own hand at independent development of an upgraded Civ4 AI. No promises, though, and even if that ever does happen, it wouldn't be for years yet, for a variety of reasons. And if I do it, I may not release it. I may simply use it as a stepping stone to getting the chance to work on AI again in a situation where it could have real impact. (Working on dead games rarely goes anywhere in and of itself). This is just hot air, for now, but I wanted to vent it anyway.

I've always used my game reports to open a window in to my experiences, good or bad. I tell it like I see it. What else is there worth doing when it comes to games?

Here's hoping that most of the other players of this event, who all got to play unspoiled games, had better outcomes than this one, win or lose.

Thanks for reading.

- Sirian

| You have reached the end of This Report |

| Table of Contents | The Epics | Main Menu |