Can a non-cheating AI be made?
- Thread starter OTAKUjbski
- Start date
winddbourne
Prince
I like to play peacefully myself, and usually beat all my friends in the early game where religion and early wonders give a huge boost. However in the later game (gunpowder and beyond) war becomes more and more rewarding and more and more necessary. Especially since most of the land you want to move into is already taken, even on huge maps.
Beyond the sword needs to not only make the AI better, but give the better "Ghandi" AI tools for peacefully kicking butt throughout the game. Corporations and espionage sound like winning avenues for just that. Providing they are done right.
I definitely want each AI to be different. My ideal would be a modular AI with a ton of different strategies and decision making algorithms, and maybe even a few random data-bases that store information from previous games and thus "learn". That would not only allow for smarter individualized computer opponents, but also random AI's.
It would be nice later in the year, after I knew what each opponent was up to, to hit "randomized AI" and get an opponent with Ghandi's tech tree, Stalin's espionage strategies, and Catherines prefered ways to improve terrain file. Not to mention a randomized aggression rating.
If you really wanted to go gaga give me the option to randomize the special units/buildings each civ gets. How would ghandi do with the podoga and his usual special unit? How would catherine do with the roman special unit? Washington with the samurai? It's not an actual "AI change", but it changes the opponent quite a bit. And when it comes to surprising and challenging a jaded human player that's might just be all that matters.
Sure a random AI with a random special unit and random special building could be a total putz too. A pacifist tech tree, going for a war strategy. and not improving any terrain at all. But with a high number of computer players in the game somebody should give me a challenge.
Plus you could select some civilizations to randomize, and some to keep playing as they normally would. Eliminating the chance of having 15 cheiftains playing against you on noble.
The ability to save that "random AI" that really smacked you down and load it up again later would also be a HUGE bonus. After a few games with random idiots you could stack the game with superior opponents. Or at least fun ones.
Some marginal programming restrictions would stop the WORST combinations from happening. Though I'd like those restrictions to be able to be turned on and off. After all something that sounds totally dumb might just work due to other random events in the game, or due to some programming loophole the developers never noticed. Letting the AI exploit the same kind of "dumb" strategies human players use all the time.
Beyond the sword needs to not only make the AI better, but give the better "Ghandi" AI tools for peacefully kicking butt throughout the game. Corporations and espionage sound like winning avenues for just that. Providing they are done right.
I definitely want each AI to be different. My ideal would be a modular AI with a ton of different strategies and decision making algorithms, and maybe even a few random data-bases that store information from previous games and thus "learn". That would not only allow for smarter individualized computer opponents, but also random AI's.
It would be nice later in the year, after I knew what each opponent was up to, to hit "randomized AI" and get an opponent with Ghandi's tech tree, Stalin's espionage strategies, and Catherines prefered ways to improve terrain file. Not to mention a randomized aggression rating.
If you really wanted to go gaga give me the option to randomize the special units/buildings each civ gets. How would ghandi do with the podoga and his usual special unit? How would catherine do with the roman special unit? Washington with the samurai? It's not an actual "AI change", but it changes the opponent quite a bit. And when it comes to surprising and challenging a jaded human player that's might just be all that matters.
Sure a random AI with a random special unit and random special building could be a total putz too. A pacifist tech tree, going for a war strategy. and not improving any terrain at all. But with a high number of computer players in the game somebody should give me a challenge.
Plus you could select some civilizations to randomize, and some to keep playing as they normally would. Eliminating the chance of having 15 cheiftains playing against you on noble.
The ability to save that "random AI" that really smacked you down and load it up again later would also be a HUGE bonus. After a few games with random idiots you could stack the game with superior opponents. Or at least fun ones.
Some marginal programming restrictions would stop the WORST combinations from happening. Though I'd like those restrictions to be able to be turned on and off. After all something that sounds totally dumb might just work due to other random events in the game, or due to some programming loophole the developers never noticed. Letting the AI exploit the same kind of "dumb" strategies human players use all the time.
Ok, as a programmer, I regret to inform you that most of that was completely useless. The problem is not what the AI knows. If you want the AI to know something, it has full access to all the game data, you can make it omniscient if you want. The problem is making the AI make correct choices. To do that, you need clearly defined rules. For example:
I could write code to do that. The more explicit you get in those kinds of instructions, the easier it is to write the code because I know very clearly what to do.
magicalsushi
Prince
Perhaps I shouldn't be speaking for OTAKUjbski, but it sounds to me like you missed the point. We don't want a cheating AI which knows stuff that a human couldn't know (or deduce) in the same situation, but we do want the AI to be able to make deductions in the same way that a human can. I'm a programmer too, and it's pretty obvious to me how one could implement code that gave the AI the ability to make the interesting deductions he mentioned.
He didn't specify in detail how the information should be used, but, his post was already extremely long. As it doesn't require much imagination to see the value of the information he was talking about (paraphrase: "I've worked out that Elizabeth just researched Astronomy, so that huge spike on her power graph doen't mean she's just built lots of units" - it seems obvious to me how this information might be useful), there wasn't any need to go into further detail. Sure, the AI needs to actually make choices, but gathering good information on which to base those choices is a non-trivial matter. As a (fairly dismal) human player, I might occasionally have made the same astronomy deduction he mentioned (for one example), but I've never habitually looked out for it, and if I'd tried programming AI myself before reading this thread, it probably wouldn't have occured to me to make the computer try to deduce that piece of information. There are many other examples of the same kind of thinking (and the fact that he didn't enumerate them exhuastively doesn't mean no-one is thinking about them), and an AI which made use of this kind of this information would be superior to one which did not. In order to use it, it first has to deduce it. It's an important issue.
Sure, the computer doesn't "need" to deduce stuff like this, because it's all sitting in its memory somewhere, but if AI leaders just get to peek at information they shouldn't know, this is cheating. It ought to be obvious (perhaps it's not) that each AI leader has (or should have) an internal model of what information he/she currently knows. Logicial deductions of the type described by OTAKUjbski would be valid (and valuable) additions to this model. Arbitrary access to all the hidden information in the game would not be.
At the risk of stating the obvious, the AI ought to distinguish between deductions that it's certain about, and deductions about which it's confident but not completely certain. Information has value, and that value is proportional to its certainty. Also, it's probably worthwhile for each AI leader to maintain a model of information about what the other leaders know...for example, it might be useful to Catherine to consider that her enemy, Shaka, is unable to see that she has researched Military Tradition because he has not yet researched the prerequisites for it. Anything about knowing that someone knows that you know that someone else knows that blah blah blah can get a bit complicated though.
Daedal
Prince
I'd love to see a non-cheating AI that can scale in difficulty (like the chess program mentioned a while back). However, chess algorithms are much easier to program than Civ4 algorithms because of all the variables involved in Civ4. Not only do you need to find a way to take into account all of the variables in the game, you need to make it scalable so you can have difficulty levels. It's not only a matter of scaling how far in the future the computer thinks. The idea of programming something this complex is a daunting task. You basically have to design an AI that can think like a human (at least as far as Civ4 play goes).
In a practical sense, what's the incentive for Firaxis to do this? So a couple of players quit or threaten to quit. You still bought the game and you'll still buy the next game.
You're hooked so you're no longer important. A great AI might attract a DoD war games contract, but it won't attract nearly as many new players as flashy graphics and laundry lists of underdeveloped new features (vassals anyone?) that can be listed in reviews and press releases.
In a practical sense, what's the incentive for Firaxis to do this? So a couple of players quit or threaten to quit. You still bought the game and you'll still buy the next game.
You're hooked so you're no longer important. A great AI might attract a DoD war games contract, but it won't attract nearly as many new players as flashy graphics and laundry lists of underdeveloped new features (vassals anyone?) that can be listed in reviews and press releases.I wasn't suggesting to base difficulty based primarily on the number of featres. We disagree on some things, but I actually think we agree on more then it might seem.I agree that it would be a bad idea to take features out of the easier game settings. I also think the AI needs to be smarter but the thing when I play on a higher difficulty level I don't want JUST a smarter AI I also want new challenges.
As it is right now, the only challenge for the higher diifficulty levels is trying to win a diplomatic or cultural victory. I find that in higher difficulty levels I am forced to go with conquests as eliminating the AI is the only way to deal with the AI. Especially during the early game when you need to establish your power. A good game does not make it so that one path is the clear choice for winning. In fact that is how it should be for the EASIER levels. For a higher difficulty it should NOT be clear what path is the best to take.
The thing about features is that they are too easy to master. A smarter AI only means that an AI can more easily master these features, but it doesn't make it any more difficult for ME to master them. A good game is a game where higher difficulties challenge the player in new ways. For example, and I stress pollution and city riots/revolts because they WERE important in managing your empire and it became more difficult to manage it. Maybe people thought they were annoying but that is why I say they should be on the more difficult settings.
Just curious, do you not think that people would complain about the new Events feature in BTS? I could imagine many people complaining about an earthquake all of a sudden ruining their plans. Or that a marriage now makes it more difficult to decide to declare war on somone. There will always be complainers about "annoying features", but they need to realize that these things are what we call challenges.
As it is right now, the only challenge for the higher diifficulty levels is trying to win a diplomatic or cultural victory. I find that in higher difficulty levels I am forced to go with conquests as eliminating the AI is the only way to deal with the AI. Especially during the early game when you need to establish your power. A good game does not make it so that one path is the clear choice for winning. In fact that is how it should be for the EASIER levels. For a higher difficulty it should NOT be clear what path is the best to take.
The thing about features is that they are too easy to master. A smarter AI only means that an AI can more easily master these features, but it doesn't make it any more difficult for ME to master them. A good game is a game where higher difficulties challenge the player in new ways. For example, and I stress pollution and city riots/revolts because they WERE important in managing your empire and it became more difficult to manage it. Maybe people thought they were annoying but that is why I say they should be on the more difficult settings.
Just curious, do you not think that people would complain about the new Events feature in BTS? I could imagine many people complaining about an earthquake all of a sudden ruining their plans. Or that a marriage now makes it more difficult to decide to declare war on somone. There will always be complainers about "annoying features", but they need to realize that these things are what we call challenges.
Daedal
Prince
Random events are not the way to fix the broken AI. They make each game more unique, but the AI will still act the same way. Adding more features like pollution and riots is fine, but again irrelevant. That just gives me more stuff to do without changing the AI at all. The path to victory is always the same: beeline to war tech, build army, attack neighbor, vassalize/destroy neighbor, lather rinse repeat. You can't beat the AI in tech so don't even bother trying. The details of geography add very little to the game. There's no variety at all. The AI is 100% predictable and the only thing you need to do on higher difficulty levels is overcome the level of cheating the AI does. There should be more intelligence and cunning in the AI's actions so that it doesn't need such extreme bonuses to do well against humans.
dh_epic
Cold War Veteran
You can't point to a game where the AI can beat you without handicaps, unless that game is as simple as tic tac toe, or the AI is *everything* like Deep Blue. They can't make an AI that doesn't cheat at all.
But can they make the AI better? Probably. But game development is a zero sum game: you only have so much time and resources to make what it is that you want to make. Are you willing to give up Civics? Are you willing to give up everything from the industrial era onward?
I wish they'd drop all the 3D artists and go back to 2D, so they can focus on game play... but alas, it's not gonna happen.
That said, the Civ 4 AI is way better than Civ 3. I'd like to think that this wasn't just because they spent more time on it, but because they learned something from the previous game. If that's true, then Civ 5 can learn from a lot of the Civ 4 mistakes.
But can they make the AI better? Probably. But game development is a zero sum game: you only have so much time and resources to make what it is that you want to make. Are you willing to give up Civics? Are you willing to give up everything from the industrial era onward?
I wish they'd drop all the 3D artists and go back to 2D, so they can focus on game play... but alas, it's not gonna happen.
That said, the Civ 4 AI is way better than Civ 3. I'd like to think that this wasn't just because they spent more time on it, but because they learned something from the previous game. If that's true, then Civ 5 can learn from a lot of the Civ 4 mistakes.
AI of kind you want would not be posible untill REAL AI come to play.
Even if AI know how to use every trick in a book, human is adaptable. Early or late it will figure AI out and AI will become easy again.
Human knows every trick in a book, but when to use a particular trick?
Lots of time agains a human opponent it is actially a risk assesment.
If I invest this in to economic, I might get to better units faster. But if he hit me right now, I might be in trouble.
Chess Kind of AI (which simple use raw force to check all posiblemoves) is imposible in civ, as there way to many posible actions.
IN addition, would you really will want to play again AI which need 1/2 of an hour for a move?
Even if AI know how to use every trick in a book, human is adaptable. Early or late it will figure AI out and AI will become easy again.
Human knows every trick in a book, but when to use a particular trick?
Lots of time agains a human opponent it is actially a risk assesment.
If I invest this in to economic, I might get to better units faster. But if he hit me right now, I might be in trouble.
Chess Kind of AI (which simple use raw force to check all posiblemoves) is imposible in civ, as there way to many posible actions.
IN addition, would you really will want to play again AI which need 1/2 of an hour for a move?
I'm all for anyone making the AI better than it is. I'd hate to see different AI strategies at different difficulty levels though. The civilization series is one where handicaps are the best way to separate difficulty levels. Making comparisons with chess is mostly useless because in chess you can't make an AI a better challenge by giving it a handicap like 2 queens or 3 knights (well you could... but I don't think I know anyone who'd play such a game). In other words, to make an AI play chess more challengingly you must make it smarter.
If I were a Monarch player in Civ, I'd like to see the AI use all the tricks it has in its book - not just most of them.
I don't think Civilization will ever have different core features at different difficulty levels. Besides, if more empire management features were added at higher difficulties, you'd more likely make the AI less of a challenge because in general the AIs tend to be more effective with simpler game mechanics. Civ4's AI playing Civ1 (use your imagination) would be a sight to see!
By the way, at Noble the AI still gets a few bonuses over the player - significanly lower upgrade costs is one example - so it is not really the level playing field. There is no difficulty where the player and AI are handicapped equally.
If I were a Monarch player in Civ, I'd like to see the AI use all the tricks it has in its book - not just most of them.
I don't think Civilization will ever have different core features at different difficulty levels. Besides, if more empire management features were added at higher difficulties, you'd more likely make the AI less of a challenge because in general the AIs tend to be more effective with simpler game mechanics. Civ4's AI playing Civ1 (use your imagination) would be a sight to see!
By the way, at Noble the AI still gets a few bonuses over the player - significanly lower upgrade costs is one example - so it is not really the level playing field. There is no difficulty where the player and AI are handicapped equally.
Daedal
Prince
Here's what I've come up with in terms of a non-cheating AI where difficulty increases through strategy, not handicaps. I'll assume 7 levels of difficulty, I forget exactly how many there are and I'm not at home to check. For all of these assume an even playing field, including diplomacy bonuses.
1 (lowest): AI builds random buildings, units, and improvements. It researches random techs. Diplomacy checks are still made and the AI will still go to war if provoked. This is like a complete newbie messing with the game without really understanding what anything does.
2: AI understands what the buildings, units, and improvements do more or less, and has some idea of what techs are generally useful. It will build whatever buildings/units/techs are available to it that are flagged as useful without any overall strategy and without specializing any cities.
3: Same as before with two big changes. Once the AI is close to some type of victory (has 70% of techs, has large culture in 3 cities, whatever checks are built in to the AI) it begins to beeline toward that victory with a secondary focus on defense. The other change is specialized cities. It will now build science centers, military centers, etc. starting early in the game. There is still no overall strategy until near the end.
4: The middle level.. AI has a clear idea from the beginning of the game what strategy it will focus on to win. The strategy is randomly chosen, but the focus carries on throughout the game. Secondary focus is on blocking other civs from winning through pillaging wars and taking important cities. Tertiary focus is on defense. The AI will start making tactical diplomatic moves, such as defending a weaker nation against a larger rival to prevent runaway success of a civ.
5: The AI will now choose its strategy based on the leader characteristics available to it. The AI understands great people better and will use these to optimise lightbulbs vs settling. The AI employs a perfect blend of CE/FE throughout the game based on its chosen strategy. The AI will further refine its diplomacy to include setting up temporary pacts to gain power quickly, and then backstabbing its ally due to that ally being the highest rival. It would have done this anyway, but now the AI will anticipate this and build up forces next to allies for when that situation occurs. At this point the AI should also use diversionary military tactics to trick other AIs and to trick you.
6: The AI now sees the entire map. Compared to the cheating it does nowadays, this is a tiny concession. This way the AI will know where to build cities for optimal performance in terms of cost to maintain the city, function of that city, etc. It will be able to calculate all these things ahead of time because it will know exactly what buildings it plans on putting in those cities given ideal conditions.
7: The AI is omnipotent. It knows who is in the lead and what strategy they are employing. It uses that knowledge to block that avenue of success and then goes down the list to the next one. This may not seem like a big advantage, but consider that you can't use a diversion against this AI, it knows exactly which cities to strike and which tiles to pillage, it knows how to block a win and it will band together to achieve that goal. I can see the timer running out for many of these games.
Anyway, those last two levels employ some level of cheating, but it's what I consider passive cheating because it gives the AI knowledge. It's not active cheating like giving him 20 extra hammers per city, 50% chance to win any battle regardless of the odds, etc.
I'm not even a very good Civ4 player, I'm sure many of you could come up with better ways to give the AI a fighting chance with strategies that can be reduced down to a fairly simple algorithm that can be programmed. If you think a game like this would be too easy compared to what you play now with the odds stacked ridiculously against you, just remember that at the moment the AI is about as smart as a sack of doorknobs and you can easily manipulate it and defeat it because you know exactly what it's going to do every single game.
1 (lowest): AI builds random buildings, units, and improvements. It researches random techs. Diplomacy checks are still made and the AI will still go to war if provoked. This is like a complete newbie messing with the game without really understanding what anything does.
2: AI understands what the buildings, units, and improvements do more or less, and has some idea of what techs are generally useful. It will build whatever buildings/units/techs are available to it that are flagged as useful without any overall strategy and without specializing any cities.
3: Same as before with two big changes. Once the AI is close to some type of victory (has 70% of techs, has large culture in 3 cities, whatever checks are built in to the AI) it begins to beeline toward that victory with a secondary focus on defense. The other change is specialized cities. It will now build science centers, military centers, etc. starting early in the game. There is still no overall strategy until near the end.
4: The middle level.. AI has a clear idea from the beginning of the game what strategy it will focus on to win. The strategy is randomly chosen, but the focus carries on throughout the game. Secondary focus is on blocking other civs from winning through pillaging wars and taking important cities. Tertiary focus is on defense. The AI will start making tactical diplomatic moves, such as defending a weaker nation against a larger rival to prevent runaway success of a civ.
5: The AI will now choose its strategy based on the leader characteristics available to it. The AI understands great people better and will use these to optimise lightbulbs vs settling. The AI employs a perfect blend of CE/FE throughout the game based on its chosen strategy. The AI will further refine its diplomacy to include setting up temporary pacts to gain power quickly, and then backstabbing its ally due to that ally being the highest rival. It would have done this anyway, but now the AI will anticipate this and build up forces next to allies for when that situation occurs. At this point the AI should also use diversionary military tactics to trick other AIs and to trick you.
6: The AI now sees the entire map. Compared to the cheating it does nowadays, this is a tiny concession. This way the AI will know where to build cities for optimal performance in terms of cost to maintain the city, function of that city, etc. It will be able to calculate all these things ahead of time because it will know exactly what buildings it plans on putting in those cities given ideal conditions.
7: The AI is omnipotent. It knows who is in the lead and what strategy they are employing. It uses that knowledge to block that avenue of success and then goes down the list to the next one. This may not seem like a big advantage, but consider that you can't use a diversion against this AI, it knows exactly which cities to strike and which tiles to pillage, it knows how to block a win and it will band together to achieve that goal. I can see the timer running out for many of these games.
Anyway, those last two levels employ some level of cheating, but it's what I consider passive cheating because it gives the AI knowledge. It's not active cheating like giving him 20 extra hammers per city, 50% chance to win any battle regardless of the odds, etc.
I'm not even a very good Civ4 player, I'm sure many of you could come up with better ways to give the AI a fighting chance with strategies that can be reduced down to a fairly simple algorithm that can be programmed. If you think a game like this would be too easy compared to what you play now with the odds stacked ridiculously against you, just remember that at the moment the AI is about as smart as a sack of doorknobs and you can easily manipulate it and defeat it because you know exactly what it's going to do every single game.
dh_epic
Cold War Veteran
I think it's very realistic to say that the highest levels will need some cheating. The real question we should ask is what kind of cheating would be acceptable.
I would agree that I'd rather have the AI get some knowledge benefits than to whip them into an economic frenzy. The problem isn't that the economic frenzy is too hard or annoying, just that it actually changes the way the whole game is played. If the AI just had more map or combat knowledge, they would probably play a lot more like a compelling warmonger. Or maybe I'm overestimating how much "knowledge" would help the AI.
Something to keep in mind, though... if the AI is cheating with map knowledge, the player can often follow the AI around to figure out the same knowledge. At least, that's how it was in Civ 3.
Another idea: at the higher levels, the AI should collude more. As the player starts conquering the world, another AI on the other side of the world should not only conquer their neighbor -- but that neighbor should let themselves be conquered. That would make for a compelling face off in the mid to late game.
I wasn't arguing that Random events would fix the AI. Who on earth would ever claim that?
I think everyone misunderstood me. I am not claiming that the AI should remain predictable and dumb. But I was simply arguing that while an improved AI means that the AI will be smarter and rely less on bonuses, it still does nothing to make the game more difficult for me.
The whole reason why I brought in random events was because someone said pollution and city riots were not fun but I was suggesting that random events would have a similar amount of complaints for example: "I had a large wealth producing city and than that darn hurricane came." It really was just to show that every feature is going to have a group of people who complain about it.
I think everyone misunderstood me. I am not claiming that the AI should remain predictable and dumb. But I was simply arguing that while an improved AI means that the AI will be smarter and rely less on bonuses, it still does nothing to make the game more difficult for me.
The whole reason why I brought in random events was because someone said pollution and city riots were not fun but I was suggesting that random events would have a similar amount of complaints for example: "I had a large wealth producing city and than that darn hurricane came." It really was just to show that every feature is going to have a group of people who complain about it.
OTAKUjbski
TK421
Semantics.
I understood where you were going with your comparison, but I personally feel like it's the proverbial comparison of an "apple to an orange": AI intelligence is not merely a feature of Civilization -- connoting it is just an addition or bonus to the game. It is so much more than that.
What tricks?
The AI is already programmed this way. And, therein lies the problem: it doesn't know any tricks!!! At every level, the AI is a blundering imbecile incapable of implementing any modicum of strategy.
As others have said, all this does is limit the number of viable strategies at higher levels. Conquest becomes vital and often leads to what seems to be the most widely achieved high-level victory, Domination (though by limiting victory conditions [a form of manipulation akin to cheating, imho], others can be achieved).
Sound logic.
Though I don't necessarily agree with all of Daedal's specific changes, I commend and support the logic behind them whole-heartedly.
That logic is to create an AI which at least to some degree 'thinks' and performs similar to how a human player would. Of course, the level of programming necessary to mimic an experienced player would eventually reach a plausible limit, and cheating would need to be implemented.
I simply feel the CivIV programmers gave up a little too early is all.
Human levels made into AI levels.
There are numerous strategies we've developed while playing CivIV (thanks mostly in part to these boards). Sadly, even some of what we refer to as "basics" appear to be completely lost on the AI at all levels of play.
Many people seem to be against 'dumbing down' the AI at various difficulty levels. I'm still trying to figure out how this is any different than the current method of penalizing the AI. Additionally, this is already how the 'human levels' are determined -- based on your 'dumbness' to the ways of CivIV!
However, Prince and Monarch are virtually guaranteed victories, so I feel like there is no more "challenge" in the game for me -- except maybe to beat personal bests.
I think implementing changes to improve the AI's competitiveness at higher levels without restricting gameplay will go a long way towards improving the long-term playability and appeal of current and future Civilization titles.
My plight.
Perhaps I am alone, but I will not be rushing out in 41 days to pick up BTS. Instead, I will be waiting for report on how well the AI plays and if there really has been an improvement to the AI (and not just the implementation of a BetterAI we've already been using).
If little / nothing has been improved, it's likely Warlords will be the last Civilization or Firaxis title I purchase. I'll take my $50 over to Nintendo or something.
I understood where you were going with your comparison, but I personally feel like it's the proverbial comparison of an "apple to an orange": AI intelligence is not merely a feature of Civilization -- connoting it is just an addition or bonus to the game. It is so much more than that.
What tricks?
The AI is already programmed this way. And, therein lies the problem: it doesn't know any tricks!!! At every level, the AI is a blundering imbecile incapable of implementing any modicum of strategy.
As others have said, all this does is limit the number of viable strategies at higher levels. Conquest becomes vital and often leads to what seems to be the most widely achieved high-level victory, Domination (though by limiting victory conditions [a form of manipulation akin to cheating, imho], others can be achieved).
Sound logic.
Though I don't necessarily agree with all of Daedal's specific changes, I commend and support the logic behind them whole-heartedly.
That logic is to create an AI which at least to some degree 'thinks' and performs similar to how a human player would. Of course, the level of programming necessary to mimic an experienced player would eventually reach a plausible limit, and cheating would need to be implemented.
I simply feel the CivIV programmers gave up a little too early is all.
Human levels made into AI levels.
There are numerous strategies we've developed while playing CivIV (thanks mostly in part to these boards). Sadly, even some of what we refer to as "basics" appear to be completely lost on the AI at all levels of play.
Many people seem to be against 'dumbing down' the AI at various difficulty levels. I'm still trying to figure out how this is any different than the current method of penalizing the AI. Additionally, this is already how the 'human levels' are determined -- based on your 'dumbness' to the ways of CivIV!
- Settler. A Settler-level human is basically ignorant to all ways of CivIV and so blindly clicks buttons with little understanding or regard for the consequences.
Similarly (as Daedal put forth), the Settler-level AI should also act quite random -- teching and building with very little direction.
- Noble. In addition to Daedal's 'middle level' (4), the AI would also begin introducing various gambits -- most of which can be found in one of Sisiutil's War Academy Articles. This could also include technology beelines -- such as Alphabet, Music, Liberalism, MT, Democracy, Biology, Mass Media, etc.
It is my opinion that ALL of these gambits and strategies should be made available to the AI at every level. The difference between the various levels is determined by how many gambits the AI will attempt in any given game and even why they are sought.
At Noble level, perhaps 3 gambits will be randomly selected.
- Prince. In addition to implementing more gambits, the Prince AI would be set apart by its utilization of specialized economic models (what we call "CS/SE/FE") and city specialization (such as
farms, 
/
cities, 
cities, etc.)
The Prince AI might implement 6 or 7 gambits (whatever is determined to be ideal), and has a decent understanding of the ideal use of Great People with an emphasis still on settling.
- Monarch. I envision this as the "new middle level". This is when the AI not only knows every gambit, but also knows which ones are most likely to succeed and which ones are most crucial to their winning strategy (i.e., the Monarch AI typically won't try to build the Pyramids if it has no Stone, is Financial and/or is running a CE.)
This is also the level when the AI plays to its leader traits, UU/UB and immediate resources very well. (Such as bypassing unnecessary or redundant technologies and/or beelining the UU/UB technology.)
Finally, the AI fully understands pools and how to make the most of them -- especially where 'timing' specific Great People is concerned.
- Emperor. At least as far as my imagination will let me go, this is when cheating comes into play -- such as extra units and technologies, discounts, etc. (Much of what we see now.)
- Deity. I would want to see this level include a 'twist'. Basically, every AI would perform at the Immortal level except one -- which would be the Deity. After the map and starting locations have been determined, the AI checks every starting civ's location and calculates which starting location is the "best" (based on the available
and resources, proximity to other civilizations, etc.) and give that civ Deity status.
That Deity civ would have complete knowledge of everything in the game -- including all resource locations, other civs' building and research choices, unit locations, etc. With this knowledge, the Deity would take whatever measures are plausible and necessary to prevent other civs from succeeding (especially the human) and to making the most of gambits (such as delaying Liberalism until 1 turn prior to another civ's discovery of it).
Additionally (thank you dh_epic for the idea), the other Immortal AI civs would be more-or-less autonomous but with an ultimate collusion with the Deity AI civ against the human civ -- even to the extent of gifting cities and entire armies!!!
However, Prince and Monarch are virtually guaranteed victories, so I feel like there is no more "challenge" in the game for me -- except maybe to beat personal bests.
I think implementing changes to improve the AI's competitiveness at higher levels without restricting gameplay will go a long way towards improving the long-term playability and appeal of current and future Civilization titles.
My plight.
Perhaps I am alone, but I will not be rushing out in 41 days to pick up BTS. Instead, I will be waiting for report on how well the AI plays and if there really has been an improvement to the AI (and not just the implementation of a BetterAI we've already been using).
If little / nothing has been improved, it's likely Warlords will be the last Civilization or Firaxis title I purchase. I'll take my $50 over to Nintendo or something.
I agree - it is a blundering imbecile and that's a shame.
The AI doesn't use any tricks at the moment but that's irrelevant. I thought we were talking about making improvements to the AI. i.e. If we get a new AI, whether it be from Firaxis, Blake/Iustus or any other modder/s, which does happen to know some more tricks, then I'd prefer it to use those tricks at every difficulty level.
In all, there are two main problems with your idea. These are debatable but they are my opinion.
1) If the game developers were to design and implement several different AIs for several different difficulty levels, it'd take much longer. Perhaps if the differences between each "AI" were small it would be feasible, but the changes you suggest are dramatic between the difficulties. Rather than having more time spent making several AIs, I'd rather see lots of time spent making only one AI which plays impressively at all difficultly levels.
2) You risk discriminating against the low difficulty players. Suppose for example you made sea invasions a tactic that only Monarch and above AIs would use. Wouldn't that be unfair on the players who play Prince and below? They want to experience a challenging game too.
I also happen to be against any cheating on the AI's part and mostly against discrimination toward the human player (e.g. I'd make concessions for the AI to be more wary of humans and to expect them to be more cunning, but not to gang up so one AI can win instead of the human). Note handicapping is not cheating.
Anyway, I think you're going to find there are two camps of people on this issue - those who want
difficulty levels = handicaps
and those who want
difficulty levels = AI changes
Due to my personal opinion, I fall into the first category. Remember though, both camps want to see AI improvements in general - that's not what the dispute is over. Don't confuse me for someone who is contented with an AI that is a blundering imbecile at every level.
dh_epic
Cold War Veteran
All things considered, though... I doubt Firaxis just "gave up" on the AI. Having worked in software, division of resources is one of the toughest things you'll have to do. I'm sure they said "we have X amount of time for this feature, and Y amount of time for that feature... and that means Z amount of time to deal with the AI". I'm sure they had to make a tough decision like "will we have religion, or will we cut that out and work on the AI?" I guess they thought religion would be more marketable.
A non-cheating AI isn't really an issue of intelligence or skill. It's an issue of resources.
A non-cheating AI isn't really an issue of intelligence or skill. It's an issue of resources.
OTAKUjbski
TK421
@PieceOfMind:
Perhaps I am unjustly discriminating against the low difficulty players. However, I contend the majority of the CivIV populace (60-70%) plays at Noble, Prince or Monarch and that only about 10% play at the lower levels.
I also presume the majority of players who want to see a "complete AI" at lower levels are very unlikely to remain at those lower levels for more than a couple games as they learn the ins and outs of the game mechanics.
However, your point is more valid than mine, so I find myself wanting to agree more than disagree. A Complete yet handicapped AI would present more "monkey see; monkey do" learning opportunities.
I think I'm starting to fall into both categories but still with a major emphasis on AI changes.
I guess what I really want to see is an upgraded AI capable of challenging current Monarch-level players without any handicaps. Essentially, in the same way BetterAI takes most players back one level, I'd like to see another level of AI capable of doing this again.
Perhaps I am unjustly discriminating against the low difficulty players. However, I contend the majority of the CivIV populace (60-70%) plays at Noble, Prince or Monarch and that only about 10% play at the lower levels.
I also presume the majority of players who want to see a "complete AI" at lower levels are very unlikely to remain at those lower levels for more than a couple games as they learn the ins and outs of the game mechanics.
However, your point is more valid than mine, so I find myself wanting to agree more than disagree. A Complete yet handicapped AI would present more "monkey see; monkey do" learning opportunities.
I think I'm starting to fall into both categories but still with a major emphasis on AI changes.
I guess what I really want to see is an upgraded AI capable of challenging current Monarch-level players without any handicaps. Essentially, in the same way BetterAI takes most players back one level, I'd like to see another level of AI capable of doing this again.
I think everyone should recognize that 'bad AI' is the number one lament for probably just about every complex strategy game ever made. Across the entire business, all grognards complain about better aI.
Even with chess, the current AI is a result of programming that started in 1969, took thousands of people making improvements, had IBM devote substantial resources (meaning $ and lots of it) even though they knew it wasn't directly cost effective. And chess is a very simple game relative to CIV.
So, of course, EVERY gamer would rather face a curve of a smarter opponent as he/she gets better without the computer cheater, as we have if we play the Chessmaster where even a GM can have a good game.
What we as gamers have to face is what level of 'smartness' can be effectively programmed and being reasonable. I bet the overwhelming majority of CIV players would answer affirmative to the following question:
'Do you want the CIV AI to be better across the board and therefore we give it fewer advantages over the human at higher levels to balance'?
Now, let me ask the question another way -- 'Do you want to pay a lot more for the company to hire programmers and testers to do this?' Not so easy. Some will say yes and some will say no.
I doubt there will be too many complex games like CIV in which the computer can forego advantages at high level to be competitive with the best players. Besides grognards, the average computer game player who plays a game like this 2 or three times will have enough trouble winning on Noble.
Constructively, the best you can usually do is point out WHAT AREAS the AI needs to improve and where simple improvements may be cost effective. For instance, worker management comes to mind here.
Best wishes,
Breunor
Even with chess, the current AI is a result of programming that started in 1969, took thousands of people making improvements, had IBM devote substantial resources (meaning $ and lots of it) even though they knew it wasn't directly cost effective. And chess is a very simple game relative to CIV.
So, of course, EVERY gamer would rather face a curve of a smarter opponent as he/she gets better without the computer cheater, as we have if we play the Chessmaster where even a GM can have a good game.
What we as gamers have to face is what level of 'smartness' can be effectively programmed and being reasonable. I bet the overwhelming majority of CIV players would answer affirmative to the following question:
'Do you want the CIV AI to be better across the board and therefore we give it fewer advantages over the human at higher levels to balance'?
Now, let me ask the question another way -- 'Do you want to pay a lot more for the company to hire programmers and testers to do this?' Not so easy. Some will say yes and some will say no.
I doubt there will be too many complex games like CIV in which the computer can forego advantages at high level to be competitive with the best players. Besides grognards, the average computer game player who plays a game like this 2 or three times will have enough trouble winning on Noble.
Constructively, the best you can usually do is point out WHAT AREAS the AI needs to improve and where simple improvements may be cost effective. For instance, worker management comes to mind here.
Best wishes,
Breunor
dh_epic
Cold War Veteran
Very well put, Breunor. As someone who has worked in software before and studied AI in college, I'd say you summed up the difficulties very well.
Daedal
Prince
I agree that it is difficult to come up with an AI, but isn't it a shame that a chess AI has been in the works for almost 40 years, if that did start in 1969, and in that time this is the best there is? The applications for a real AI reach far outside of gaming yet so little progress is made. Is the problem a lack of funding? The issue of national security surrounding an AI that can plan global domination? The social turmoil that would occur given a real thinking machine? Maybe I'm over-thinking it, but at the very least some tiny baby steps toward a real AI would be nice. Maybe a more complicated script with lots of branches to take into account more of the variables in Civ, along the lines that Breunor suggested. Is that too much to ask for the money we sink into the Civ series? It seems like the collective knowledge of this forum is enough to provide all the branches needed and then it's just a matter of writing a whole lot of switch statements and throwing in a couple of random variables. I'm no expert programmer, but it certainly seems doable.
