Can a non-cheating AI be made?

I've been playing games since before Windows was even a concept.

The first game I ever played was 'Warmonger', an RTS-type game whose goal was to conquer a series of islands and thus control the world. (Good luck finding ANY info on it, though.) Next was Civilization.

In the 17 years since Warmonger and Civilization (DOS) were released, AI programming has come a LONG way. However, one fundamental element of difficulty settings remains the same:

More difficult AI == gets to cheat + more freebies

We all know what I'm talking about. On difficulty settings above Noble, the AI doesn't get any smarter. Instead, it gets free units & techs, building & research discounts, health & happiness bonuses, etc.

It's not better ... it's just bigger.

I can compare this to a number of IRL scenarios where this philosophy makes absolutely no sense and even seems outright stupid:

• What if at the end of the 2004 Summer Olympics in Athens, they would've told Yan Bhartelemy Varela he needed a more difficult opponent than Atagün Yalçınkaya and so replaced Yalçınkaya with Alexander Povetkin -- a super heavyweight boxer nearly twice his size?
• Or maybe in that same year we give David Williams an extra Ace to best the two 8's Greg Raymer, winner of the 2004 World Series of Poker, was holding in that last hand to 'challenge him more'.
• Even better ... let's see if Kurt Busch, 2004 Nascar Champion, can still win if his car is loaded down with Bricks.

I mean, seriously, isn't that basically what CivIV (and other games) are doing when they hamper your progression but stimulate the AI's?

Chess is about the only game a computer can accurately portray a 'better' opponent, because the best players are the ones who can see the most moves ahead in order to choose the one right now most likely to result in a victory -- not the ones who get two Queens or 4 Knights (which never happens).

So why can't Firaxis get it right?

I can't say I've ever seen the AI do anything 'better' in any of my CivIV games. They just do it faster and for cheaper.

There are strategies, tricks, all kinds of stuff we use to determine what to do and when to do it.

Does the AI know how to best use an Oracle or Liberalism slingshot? If the AI's taking Sid's suggestions (adviser recommendations), then I can guarantee they don't.

Likewise, based on the blue circle and highlight suggestions of Workers and Settlers and comparing those against what the AI will do when controlling an automated worker, I can also assure you they're clueless on how best to improve tiles. Why do they always seem to start with the worst tiles?

I would also argue the AI has no real concept of chop-rushing, because -- again -- automated workers are seemingly mindless and will chop down forests regardless of what the current build is.

This list goes on and on and on, and I think nearly everyone could probably add multiple examples of when the AI completely took a wrong turn.

How do we make it better?

BetterAI was a giant step in the right direction, but it still just doesn't feel like enough ... not until the cheating can be eliminated, imho.

I'm not a programmer, so I'm not too keen on what tricks there might be for making AI's more 'strategic'. Otherwise, I'd probably be working on BetterAI++ or something.

The only suggestion that seems to make sense to me is to teach the computer better strategies and how to think ahead more.

I definitely don't agree with giving the AI information they shouldn't have (info I wouldn't have in the same situation). But how many times have you been able to take all the puzzle pieces and figure out what most likely is going on?

Advanced players are able to deduce facts they cannot know just based on other information.

For example, if I'm isolated on an island without Astronomy, then I know anybody I can trade with has Astronomy, and those who can't don't. Thus, I can conclude a spike in the Power graph probably does not mean an attack on me.

In the same scenario, if I can only see my own island, all I have to do is check the Relations window to see who has Open Borders and border tension (our close borders ...) to see roughly who's next to whom in the blackness of the globe.

Similarly, if I'm founding a new city and know I'll be chop-rushing a Courthouse, I usually send in 2 or 3 workers to start pre-chopping before the Settler arrives. Then on the last turn of the chopping, I found the city and BAM ... all the hammers drop on my new city on turn 1 -- thus drastically reducing its immediate and lasting impact on my economy.

The AI also doesn't seem to understand the concept of city and economy specialization, either. Some of their cities have the most random improvements. In some games I've used the World Builder to duplicate their city sites just so I can see what's going on inside ... sometimes I realize it's impossible to run the size city they're running at all. They're just getting free stuff from somewhere to keep the city big.

In closing...

This has been one of my longest posts ever about one of my biggest peeves ever.

I hope I'm not alone.

Hopefully, I'll hear some constructive criticism as to why this isn't possible or how to make it possible.

Who knows? Maybe Firaxis will include some of our ideas in CivV ... no much telling, eh?

No you are not, when i play a higher difficulty i want a challenge but i also want a fair challenge where the Ai is just cleverer

I think you're bang-on. The ideal has to be an AI that on the highest skill level is - to all intents and purposes - impossible for a human player to beat because it knows every trick there is to know and will exploit every game mechanic for the absolute maxmimum efficiency. Lower skill levels then use progressively 'hobbled' versions of the AI. (Maybe this could be done by reducing it's thinking time for shallower analysis or selectively switching off certain algorithms or both).
I'm sure that's what the programmers would like to produce as well, and we'll probably see it 'soon'. It's probably not practical on todays processors though, and it would take a very big programming effort, maybe even using techniques that don't exist yet.
Remember it's only relatively recently that a computer beat the world's best chess player (and he disputed the circumstances), and the number of valid 'moves' in any given situation is much smaller than in Civ. Civ is a broader game on a somewhat bigger map...
Knowing the AI is using performance-enhancing drugs is annoying but it's the only way to get a challenging game onto the shelves in a reasonable (i.e. financially viable) time-frame at the moment.

The BetterAI mod does a good bit of what you're talking about. In fact, they had to make a BetterHandicaps to go along with it, which scales waaay back the AI bonuses.

The current edition of the mod is quite good.

The mod has currently not been updated in about 2-3 months. They haven't said why, but there is speculation their work has been co-opted into RTS. Regardless, I heartily recommend installing the mod and handicaps NOW.

Wodan

Wodan said:

The BetterAI mod does a good bit of what you're talking about. In fact, they had to make a BetterHandicaps to go along with it, which scales waaay back the AI bonuses.

The current edition of the mod is quite good.

The mod has currently not been updated in about 2-3 months. They haven't said why, but there is speculation their work has been co-opted into RTS. Regardless, I heartily recommend installing the mod and handicaps NOW.

Wodan

Click to expand...

I have, but it still feels like it should be better.

I didn't know about the BetterHandicaps though ... that might satiate me for a while .

You've mentioned empire and city management in your post. I too would be very impressed by an AI opponenent who can actually maximise (or at least reach over 50&#37 their efficiency in running their empires. This would be only the first step, in my opinion, to creating a stronger AI opponenet. A healthy empire with a healthy economy would provide the AI with a firm base for other areas.

Those areas are diplomacy and warfare. I want an AI that can actually pull off a diversionary attack. I'd prefer an AI that doesn't attack my lone pillaging unit, while ignoring the Stack of Doom looming next to their capital. I'd like an AI that doesn't send only an archer and a warrior as its 'backdoor' offensive. Diplomacy-wise, I don't want to be able to completely trash the world's economies by getting my rivals to fight amongst each other. Don't they find it curious that while they're fighting the Fourth World War (of which I am, once again, either neutral or a minor participant) I race ahead in the Tech Race? Diplomacy and Warfare are two massive weaknesses of the AI - their awful economies just further enhance that.

I want to play at a difficulty with no advantages for either the AI or myself, and actually struggle to dominate the world. I'm glad to see BetterAI and I would support it financially if I weren't broke.

Non-cheating good AI can be done. It depends if the dev team wants to allocate resources from graph glittering to AI coding. And there would be demands from fans about the subject, however we are in minority on this issue.

I haven't checked any AI mods so I can't say how motivated and so on the people are there.

Better AI would be sweet, but then they'll have to take resources away from making the graphics better (the number one selling point of most games).

I'd say Civ-fans would understand if Firaxis used some time to create a smarter AI instead of more, nifty graphic details.

Diamondeye said:

I'd say Civ-fans would understand if Firaxis used some time to create a smarter AI instead of more, nifty graphic details.

Click to expand...

I would even go so far as to spend more money on it, too.

Of course it's possible to make a better AI for the game. But it would take more time, more people and more money. PC games have been and will be $50 for quite some time. It's not feasible to make a Civ game with a much better AI without cutting something else out. And I know there are plenty of people here who would have no problem cutting some of the things from Civ. But the reality of the situation is that the die-hards here like us are only a small fraction of the total number of people who buy the game. So Firaxis can't simply go back to 2D and hire 5 AI Programmers instead of 10 Artists.

I think the best solution here is exactly what Civ 4 did: make the AI open-source and allow the creative and talented folks here in the community tinker with the game until it's perfect (or as close as you can get to it). As many people know, Blake's done some amazing things with Civ's AI - things that wouldn't normally be possible to ship with a game. And it will only get better from here I'm sure.

Nephrite said:

Better AI would be sweet, but then they'll have to take resources away from making the graphics better (the number one selling point of most games).

Click to expand...

It's not the number one selling point of Civilization. If I see a forest, I don't care how detailed the trees are.

This ("the AI should be smarter on higher difficulty levels") issue gets raised often. I can understand why; people who routinely beat the high levels get bored running circles round the AI and wish it was smarter. That's fair. But...

You're asking the wrong question. The AI *can't* get smarter unless you install updates (either the latest Firaxis patch, or the latest BetterAI release). Please don't be selfish: if an update makes the AI smarter against Deity players, why shouldn't people playing Chieftain (or Prince, or whatever) get to play against the same competant AI, but with an appropriate handicap in the human's favour (or against them, if they're good enough to beat Prince)? You shouldn't be saying "make the AI get smarter on higher difficulties"; what you really mean is just "please make the AI smarter". Which, as you know, Firaxis and modders are doing anyway.

There isn't a team of microscopic programmers sitting in your computer changing the AI when you select a higher difficulty level. There's no reason for multiple levels of AI; the bonuses and penalties system works fine. Here is the basic issue: THE AI IS A NOBLE-LEVEL PLAYER. In fact, the skill level of the AI is what *defines* a Noble-level player (ie. someone who can usually beat the AI in a fair fight). If you improve the AI, Noble level becomes harder (and Noble players become Warlord players). *All* of the difficulty levels become harder. That's how it should be; the benchmark of Noble will rise, and players will find they have to drop a skill level or so to play an equivalent game. Deity players will still get their challenge, but no-one else will have to suffer.

So, don't ask Firaxis to make the AI stupider on lower levels. Let them carry on doing what they already do - using the same AI on all levels, so everyone benefits from the improvements. If you still find Deity too easy, that's simply because there isn't a good enough AI out there for you, period. It's nothing to do with the difficulty levels, they are a red herring - please stop dragging them into the debate, and just wait for (or better yet, contribute to) a better AI.

---------------------------------------------------------------------------------------

As an aside...sorry for the slightly patronising "there isn't a team of microscopic programmers in your computer", but I thought I'd mention it because...there could be. Figuratively speaking, that is. AI doesn't *have* to be something static that only changes (improves) when you install an update. If Firaxis implemented machine learning as part of the AI, it could adapt itself to learn to beat your playstyle. This would be difficult and costly to implement, but it should be possible to some degree.

A related possibility (a specific type of machine learning) would be to use evolutionary programming - this wouldn't really help the AI learn to beat a particular player (you'd need to play thousands or millions of games against it in person for that method to work), but if you left your computer evolving new Civ IV AI strategies while you were at work, the AI would (in theory) be slightly better at the game when you got home. It would only be competing against itself, so it wouldn't have learnt anything about how *you* play, but it should still have learnt some useful stuff about the game in general.

I really can't imagine this being implemented in a mainstream game, and it would be extremely difficult to do it effectively for something as complicated as Civ IV. But maybe it could be done.

Well, non cheating AI isn't gonna get you real far in difficulty levels. Giving AI the ability to cheat add more diffulculty levels to the game. You wouldn't go up to Sid mode without cheating. There wouldn't be one.

Another reason why they should take my idea and make the actual GAME more difficult on higher settings rather than just AI boost. There SHOULD still BE AI boost but they could retool the game to make it that each difficult setting brings a new level of complex empire managing. Pollution, revolting cities, more complex economics, complex espionage, overall more challenges on the higher difficulties. There are plenty of games out there where the the harder levels add new challenges to the game rather than just relying on giving super boost to the AI. For example the new feature Corporations. For each difficulty setting, you can make the feature more complex. The problem with the Civ Team is that they are so focused on balancing the game for people of all skills because they want to stuff every single feature in the game for everyone to always play that they need to resort to the lazy way and just add bonuses to the AI.

AznWarlord said:

Well, non cheating AI isn't gonna get you real far in difficulty levels. Giving AI the ability to cheat add more diffulculty levels to the game. You wouldn't go up to Sid mode without cheating. There wouldn't be one.

Click to expand...

Yes! This is exactly the point I tried to make earlier, but you've managed to do it without rambling for hundreds of words!

NYHunter said:

they could retool the game to make it that each difficult setting brings a new level of complex empire managing.

Click to expand...

Nice idea, but as you say, firaxis have already gone down the route of putting *every* feature in every difficulty level, and yet people still don't find the game challenging enough. But suppose they followed your suggestion and added extra features at the higher difficulty levels. Firstly, they'd need to add a heck of a lot if they were hoping to make the game significantly more difficult (imagine how many new features you'd need to add to justify the Sid difficulty level). Alternatively, they could take stuff out of the lower difficulty levels. I don't think people would be happy with either of those plans, and the latter option wouldn't even make the higher difficulties any harder.

Moreover, if we're basing difficulty levels primarily on the number of game features, I don't think this will actually make the game significantly more challenging at higher levels. It doesn't take long to learn how to use a new feature, and then the game's easy again - albeit more cluttered and time-consuming.

NYHunter said:

Pollution, revolting cities, more complex economics, complex espionage, overall more challenges on the higher difficulties.

Click to expand...

Firaxis deliberately removed pollution and revolts from the Civ series because they're not fun. Preventing revolts required tedious micromangement - not a challenge, just a pain. I quite liked the principle behind the pollution mechanic in Civ 1 and 2 (not played Civ 3), but it wasn't implemented well. The basic idea was good though, creating a kind of game-theoretic dilemma. With the new diplomacy style of Civ 4, pollution could have been a viable game feature (in addition to environmental damage, pollution could cause diplomatic penalties, and there could be some UN sanctions against nations breaking some kind of pollution limit). All the same, it's one that most people didn't like. I don't think we should be adding stuff that people hate to the higher difficulty levels; Deity level players are likely to be the ones who find time-consuming micromanagement the most frustrating in the first place.

On the other hand, stuff like complex espionage ought to be in the game at all difficulty levels, except possibly Settler. If it's difficult enough to learn that we need to hide it from inexperienced human players, how do you think the computer-controlled civs are going to cope with it? If anything, adding over-complicated features to higher difficulty levels could make the game easier, as the human players are bound to find them easier to cope with than the people who have to write AI for them. Writing AI is tough. Making the game more complicated means the AI will be weakened, or lots of effort will be needed to bring it up to its former level. Effort which might as well have just gone into improving it in the first place, instead of training it to handle unnecessary new game features.

That's just my opinion though; it sounds like we disagree about how complicated the game should be (in terms of actual game features, rather than the AI itself). I think the game's mostly fine as it is (I'm a little apprehensive about some elements of BtS); you may think it ought to be much more complex. It's a perfectly valid view, but not one that I agree with.

AznWarlord said:

Well, non cheating AI isn't gonna get you real far in difficulty levels. Giving AI the ability to cheat add more diffulculty levels to the game. You wouldn't go up to Sid mode without cheating. There wouldn't be one.

Click to expand...

Indeed, AznWarlord did sum up your post quite nicely.

Unfortunately, I know this to be true ... it's this very truth I would like to see eliminated.

I guess what I'm really asking for is not to make the Noble player better but to make the Monarch+ players better. SAY WHA?

Example: Halo and other FPS's

If I want to annihilate somebody, I'll probably call my brother and see if he wants to play. If I want a fair fight, I'll call Jordan. If I want a challenge, I'll call my friend Q. And if I want to be ground into a bloody pulp, there are plenty of people on X-Box live who can run circles around me.

What makes these 'difficulty levels' so different?

My brother doesn't play and knows nothing about the levels. (Settler)
Jordan is good but, like me, only knows a certain amount of 'tricks'. (Noble)
Q has fast reflexes and can counter most 'tricks'. (Monarch)
XBox Live guys know everything and have very fast reflexes to boot. (Deity)

--------

This is kinda what I'd like to see in Civ.

I guess in a sense it's making a smarter AI only to 'dumb it down' for the Noble difficulty level, but that's much more like the real world when you think about it.

A Noble player might put Horses / Copper in the 2nd ring to wait for a border pop, but the Monarch player is much more likely to put the strategic resource in the first ring so they can spank their neighbor that much sooner.

Similarly, the Warlord AI might try to over-expand or under-defend his cities.

The Noble AI (similar to a human) might not specialize their cities to 100% or might focus on building too many wonders.

Prince AI would definitely specialize their cities and might run a CE or SE across the whole empire instead of the Hybrid Economy the AI seems to enjoy.

Monarch AI would value early wars and bee-lining techs. They might even begin using gambits.

Etc. Etc. We all know the tricks and the strategies, and what really sets apart the Deity players is not only knowing them but knowing when to and implementing them at the right moment.

Hopefully AznWarlord can sum that up for me.

magicalsushi said:

A related possibility (a specific type of machine learning) would be to use evolutionary programming - this wouldn't really help the AI learn to beat a particular player (you'd need to play thousands or millions of games against it in person for that method to work), but if you left your computer evolving new Civ IV AI strategies while you were at work, the AI would (in theory) be slightly better at the game when you got home. It would only be competing against itself, so it wouldn't have learnt anything about how *you* play, but it should still have learnt some useful stuff about the game in general.

I really can't imagine this being implemented in a mainstream game, and it would be extremely difficult to do it effectively for something as complicated as Civ IV. But maybe it could be done.

Click to expand...

That would actually be easier than you think ... as long as implementing the code isn't too big.

The kicker here would be that Civ would upload what it learns from your games to a server where it would then be combined with what Civ learned from every other player around the world. At the end of every month, Firaxis would compile all this data into one AI update.

Every month after the new AI has been compiled, you would download and upgrade your AI. Presumably, the later build dates would be more difficult.

Eventually, you might even have an invincible AI on your hands.

With this model, handicaps would remain unchanged. Except instead of just choosing your difficulty, you would also choose which AI build to use. So if you want to play a smarter AI on a level playing field, choose a later date. Otherwise choose an earlier AI and play it on whatever handicap you want.

This would be very nice, I think. I'm currently a Noble difficulty player, although there are many times I notice that the AI just isn't very smart - and I definitely notice the penalties I receive and the bonuses the enemy receives the few times that I have played at higher difficulties. Which is sad, really, because every AI leaderhead is practically shouting out at prince and beyond, "Hey, look! I'm too stupid, so I need to triple my research rate to be any sort of a challenge!" It's sort of like if a pro bowler without bumpers was playing against a little kid who loaded his ball into a mechanical arm that shoots down a lane with bumpers, which is 99.9% guaranteed to hit a strike. The little kid doesn't know what the heck he's doing, and he doesn't deserve to win, but because of his handicap he likely will anyway.

The problem I have with that is that it might cripple any sense of AI individuality. I have noticed that some AI actually will beeline for certain things - but maybe not particularly intelligent things. For example, Gandhi beelines for Philosophy (for Pacifism), although this actually isn't a bad idea since it leads to Liberalism. There are other examples, however, in which the AI tends to beeline to the civic that provides their favored civic when it really isn't a very smart thing to do. Regardless, it gives them a sense of individuality. Gandhi is much less likely to create soldiers than, say, Catherine - which entirely makes sense. Hatshepsut, Ramesses II, and Loius XIV are also more likely to build wonders - again, this makes sense. Basically, every AI acts differently, but what you're suggesting makes it sound like every AI should act in a very specific, nearly unbeatable manner. I think most of us can agree that an early Bronze Age war is a nice way to kill off an opponent (even on higher difficulties) while they're still weak and small - all the while allowing for easy territorial expansion. And yet... Gandhi declaring war on you, just because it's a good strategy? Sorry, but no matter how good of a strategy that is, it simply shouldn't be. Facing reality, in a multiplayer game, the person who goes for Conquest is the one who is going to win (maybe because everybody goes for Conquest, so it's a 100% probability) - so that means AI will think the same way if they're "smart", which includes Gandhi. How can an AI like Asoka or Gandhi, who don't and shouldn't raze any cities, possibly go up against an AI that will strategically raze cities? They can't - and making Gandhi and Asoka raze cities just for the sake of strategy and "flawless AI" is illogical.

So while it would be nice, I think it would just completely rip apart any sense of AI individuality, which, to me, is a large part of the game. Meeting Alexander has a certain sense of, "Oh no," to it, whereas meeting Gandhi has a sense of, "He'll be an easy target / good trading partner." If Alexander plays exactly like Gandhi, because both Gandhi and Alexander are playing with an "unbeatable strategy," then every single AI is going to be identical and the game will be boring. Not to mention, you will only be as good as the AI, and if you're a pro that knows all the ins and outs, the AI can only be as good as you - meaning everything is going to depend on luck - your starting location. Luck? Yuck! The game should be about skill and strategy - not whether you start next to gold and bronze. Inevitably that does play a part even now, but with proper management you can make up for a loss if you didn't start off in an ideal location. In a game where the AI knows every little bit that you do, and is basically a mirror of yourself (assuming you both know everything), then at all boils down to nothing more than a giant pot of luck - and I don't know about you, but I don't want the game to be more about luck than it already is.

@Alsark

I totally feel you on the bowling analogy.

I'm also with you on the AI individuality thing ... that's not an aspect of the game I would want to do away with.

The reason death, destruction and early wars is so necessary at higher levels is because the AI cheats. Since they get an X% bonus here and a Y% bonus there, they can have Z% cities sooner than you that will eventually be Z% better than your city would be in its exact location.

So since the AI 'over-expanded', you MUST compensate and overcome the handicap by waging war to eventually have Z% more cities than them just to keep up. Even if you desire a peaceful UN or Cultural victory, you MUST go to war ... how effed up and backwards is that?

I think if the AI implemented better strategies, the only wars waged would be wars that needed to be waged -- wars over resources, against aggressive neighbours (AI individuality), religious crusades, etc.

Assuming equally verdant starting locations, every civilization would grow at roughly the same rate. However, this is never the case because of random starting locations (yes, luck ). This means all civs would need to play to their starting resources, leader traits, UU and UB -- not be handicapped with a mechanical arm. If all else failed, they would wage war (unless restricted by AI personality).

OTAKUjbski said:

So since the AI 'over-expanded', you MUST compensate and overcome the handicap by waging war to eventually have Z% more cities than them just to keep up. Even if you desire a peaceful UN or Cultural victory, you MUST go to war ... how effed up and backwards is that?

I think if the AI implemented better strategies, the only wars waged would be wars that needed to be waged -- wars over resources, against aggressive neighbours (AI individuality), religious crusades, etc.

Assuming equally verdant starting locations, every civilization would grow at roughly the same rate. However, this is never the case because of random starting locations (yes, luck ). This means all civs would need to play to their starting resources, leader traits, UU and UB -- not be handicapped with a mechanical arm. If all else failed, they would wage war (unless restricted by AI personality).

Click to expand...

Yeah, that's all true. I haven't played online, but don't people online who play peacefully die pretty much... 100% of the time? I haven't played online for that very reason, because I've heard all sorts of stories about, "100% of the time, expect your neighbor to go for a conquest victory. 99% of the time you'll be right." I really hate being a warmonger, so it is comments like that which I have bought into (due to their frequency), which has made me have no desire to play online. So if so many people do it, then it has to be the most effective strategy, right? Maybe not, as I haven't really experimented with it, but it seems a logical conclusion.

I think that some of the AI would ultimately be "better" than others due to their behaviors, traits, unique unit, and starting technologies - all of which the improved AI should certainly utilize to its fullest, but I think there does need to be a sort of "difficulty within a difficulty" when it comes to each of the AI (which is currently the way how it is, really). A deity difficulty Gandhi is going to be a lot smarter than a monarch difficulty Gandhi, but that still doesn't make him the ruler with a behemoth empire that your other neighbor, Alexander, might have. I've always regarded Tokugawa as the lowest difficulty AI ever (considering he's so far behind in technology), and if Firaxis wanted to keep his image of refusing to trade, then he'd still be fairly easy to beat (compared to other AI) on higher difficulties. Maybe each leaderhead should have Civlopedia information about how their AI behaves and a sort of "threat level" associated to each leaderhead. That way, a person who wants a huge challenge can go against a deity Catherine, whereas if you want a step-down from that but a step-up from immortal, you could go with another AI on deity. They'd be just as smart as one another within deity, of course, but they'd behave differently (and thus some would be "better") according to their leaderhead. Maybe Tokugawa, for example, would focus only on grabbing key military technologies and making an attempt to utilize his samurai early. As it currently is, a Tokugawa will research fishing and sailing (to get closer to Banking for Mercantilism) even if he's landlocked. Not only is Mercantilism a terrible civic in the first place (in my opinion), but it's even more terrible that someone would consider beelining to it. After all, a Tokugawa that would beeline Machinery with Civil Service (down the Feudalism path to utilize his protective archers) would be much more fearsome than a Tokugawa with Banking. The only way beelining Banking would be smart for is if they replaced the Samurai with: "Unique Unit - Japanese Banker. Special: Compared to normal bankers, Japanese bankers produce +25% income and take 5 less hammers to produce - because you don't need to make calculators for them and they're very good with numbers."

Anyway, yeah, in general I have noticed that the AI will try to beeline their Favored Civic technologies, which most of the time, is pretty stupid. I hardly even think of an AI using their favored civic as AI roleplaying anyway (which is why this argument doesn't go against supporting AI individuality), because it's not as though their civics have much of any affect on you - so why should you care what civics they're running? If they get their favored civic technology somewhere down the road and want to use that civic then fine, but they shouldn't beeline stupid technologies. They need to spend more time picking up important technologies (and technologies that give you a benefit for being the first to discover it. For example, a cultural civilization might rush to Music [for the great artist] and Theology [for religion/Sistine Chapel], whereas a civilization more technology-based would go for Physics [free scientist] and Liberalism [free technology and Free Religion]). Personally, I don't see why anybody should research Divine Right - ever. Unless you need the late religion, I think it's a really pointless technology unless maybe you want the wonders out of it (neither of which I bother building, anyway). There are so many other technologies that are a complete and utter waste of time, but the AI grabs them anyway. Just by improving this alone I think the AI would have somewhat of a nice boost - although managing tiles is the next big thing (probably the first thing, actually).

Anyway, to sum up my thoughts in a nutshell: I would love to see better artificial intelligence while maintaining a sense of AI individuality.