Let's discuss AI

So, 78.6% of players have the Treasure Hunter achievement, which means the number of people who own the game and have played it can be no lower than that. 5.6% of players have the achievement for winning on King, which means if we assume only that 78.6% of people who own the game have played it, that still means that the number that have beaten it on King is only 7.1%.

Granted, beating the game on difficulties higher than King doesn't give you the lower-level achievement, but: 1. while there are likely a lot of players who skipped Settler-Chieftain, it seems unlikely that as many players skip King to go straight to higher levels. 2. Assuming the best-case scenario for this argument and saying that literally no one who has the achievement for higher levels has played the lower levels, that cumulatively still makes a total of 10.94% of players who've beaten the game on King or higher.

Dux1 said:

No, the AI doesn't care if you're winning. It will try to win itself, but if you are 10 turns away from winning a landslide World Leader vote, for example, the AI doesn't make any attempt to prevent this from happening. It just goes about trying to pursue victory on its own terms, ignoring the fact that it's own victory plans couldn't possibly come together until after the player has already won the game.

I would much rather the AI try to prevent the player from winning. In the above example, the AI should be trying to buy off or destroy City-State allies of the player. If the player were on the verge of cultural victory, the AI should attack the player's cultural centers. Etc., etc.

Click to expand...

It does exactly that if it thinks that those tactics have any greater likelihood of success than just trying to win itself. Fact is, in the majority of cases, by the time you get that close to a victory, there may not be any tactic that will win it, and the AI pursuing the most likely one may still result in it losing.

The AI is still incredibly weak once you take away its gigantic advantages in creating things. It does many illogical things just to get in the humans' way. For instance, a distant civ sent a settler across the ocean to try to settle a wasteland between 2 of my cities when there were plenty of prime spots much closer to home. So I had a unit get in its way and for the next 50+ turns, the 2 units went back and forth over the same same tiles. The AI had no memory of my past moves and thus never changed its movement.

Second example is the Mayan attempt to cross the channel and attack my lightly held city. Instead of putting all their units into the channel at once and overwhelming my 2 ranged units/city, they never put more than 4 in at a time. With my double shots and city, that gave me 5 shots per turn, easily sinking most every turn.

Or when units just sit there and get bombarded into the ground. When I am getting slowly killed, I run away. And yet they also will continue to bombard a city long after it is out of hit points.

And they are too set on their ways with their build queue. How many times do we have to watch them build the Colossus or GL with minimum access to water tiles? Or they build settlers while at war that they are losing, and then promptly walk it right into the enemy. I like to up the population to 4 before anyone can build a settler, and it really helps.
And finally, they have no concept of the human reactions. I frequently get one civ to declare war on another that has a defensive pact with a 3rd. The 1st seems to have no idea why I am offering it 3 luxuries, and just goes for the greed. There is no real feedback from us to them for things like denouncing, settling too close to me, spying, etc. My only 2 choices are to put up with it, or wipe them out, as anything short of total destruction and they will go back to doing the thing that annoyed me to begin with. They hve no idea how annoying their constant begging is. Enough already, I am NOT giving you 75% of my gold for nothing. When I catch them spying and they agree to stop, they dont and I have no options but war. I want the option that tells them that the next time, I will consider it an act of war, so tread lightly.

Polisurgist said:

It does exactly that if it thinks that those tactics have any greater likelihood of success than just trying to win itself. Fact is, in the majority of cases, by the time you get that close to a victory, there may not be any tactic that will win it, and the AI pursuing the most likely one may still result in it losing.

Click to expand...

Nah, the AI doesn't really do that. One of my warm-up games for BNW, on King, had Assyria just sit idly and denounce me a few times while I raked allies for the diplo victory, despite him having a huge and advanced army and easy access to several of my CS allies. Ashurbanipal practically handed me the victory by not reacting.

Take a look at the AI in games like Total War and then tell me the technology isn't really there yet... It's there alright, its just that Firaxis didn't want turn-times to go south to much and choose to make the Civ AI pretty light.
Would you trade 50% longer turns for a much better AI? I would. 100%? Yepp. 200%? Hmm... 400%? Eh, no... How much patience do You have? Then realise that we live in the "instant gratification"-age... Most young people today cant wait. For anything.

Securion said:

Take a look at the AI in games like Total War and then tell me the technology isn't really there yet... It's there alright, its just that Firaxis didn't want turn-times to go south to much and choose to make the Civ AI pretty light.
Would you trade 50% longer turns for a much better AI? I would. 100%? Yepp. 200%? Hmm... 400%? Eh, no... How much patience do You have? Then realise that we live in the "instant gratification"-age... Most young people today cant wait. For anything.

Click to expand...

Somehow I knew the problem had to be Kids These Days.

But Total War isn't a true TBS.

TBS AI has always had one disvantage. the human player has forever to play their turn. Non-TBS games can have looser AI because the AI has the computational advantage to consider more outcomes and moves than the human player over a set time scale.

Other TBS like EU have restrictive almost scenario based AI that allows the programmers to zone in on fewer variables.

Civilization's challenge is that it's giving you an open world TBS game where almost any outcome can occur and scenarios in the Civ context are glorified mod packs meant to work within the existing AI structure.

If you build a game from the ground up just geared towards a single map, the programming challenges are mitigated

I remember the original design goal with Civ5 was to make a competitive AI that realized it was playing a game and tried to win. But there was an ENORMOUS furore on here about lack of diplomacy?

Diplomacy can be a part of the game even if the AI is trying to win - Civ isn't a zero sum game and there are ample opportunities for co-operation, especially with the new trade routes.

I really only have problems with the AI in a couple areas.

First of all, while I understand that programming military inteligence ai is difficult in a game like this, is it really that hard to program city states to always keep a ranged unit garissoned in their city when they are at war? I've seen too many times where I declare war on someone, and then all my city state allies move their ranged units around, allowing them to get slaughtered by ai infantry, and then leave their city completely undefended. Even worse, sometimes they send their one military unit to attack the ai, and then have it get killed by enemy cities. The AI military could use a little work, but things like that, which really shouldn't be that difficult to program, need to be in the game.
Secondly, sometimes the ai just has piles of cash that it doesn't use. Even if they don't buy city states for victory, just amassing 10,000 gold seems like a bad strategy, and they should at least use it and generally not have more than 3k at any one time (just hard code that in to the game I suppose). It feels a little awkward when I'm about to get a diplomatic victory and a computer has 12,000 gold in the bank, even if it doesn't care about me winning, it should probably steal some city states with that money anyway.

It's not about making it difficult enough, it's about making it smarter so you don't have to give it a bunch of bonuses to make it difficult, the entire gaming experience gets more interesting when the AI actually makes interesting decisions, not when it gets 50 soldiers for free just to make it harder.

That's where I want them to focus their resources, but when they have issues such as the AI not even spending money while the player wins a diplomatic game they're far from getting there. It honestly feels like this is where gaming evolves the least, modern games haven't got much better AI than games released 10 years ago, it's an incredibly underrated feature.

pezit said:

It's not about making it difficult enough, it's about making it smarter so you don't have to give it a bunch of bonuses to make it difficult, the entire gaming experience gets more interesting when the AI actually makes interesting decisions, not when it gets 50 soldiers for free just to make it harder.

That's where I want them to focus their resources, but when they have issues such as the AI not even spending money while the player wins a diplomatic game they're far from getting there. It honestly feels like this is where gaming evolves the least, modern games haven't got much better AI than games released 10 years ago, it's an incredibly underrated feature.

Click to expand...

Maybe if there's a breakthrough in computing science that allows CPU to consider so many possible outcomes they can appear to outthink a human player. Quantum computers perhaps?

Until then, even with multicore CPUs, you're mainly dealing with linear processing with the various cores assigned to handle other subsystems and menus so the game doesn't freeze IBT and you can still sort of navigate around as it's thinking.

It's certainly conceivable that in the future, 3-4 cores might be required so that the programmers will assign more than 1 core to AI whereby 1 core is always processing in the background even as the human is playing their turn while another kicks in IBT for extra processing. That could yield some improvements. But the 'AI bonus' mechanic to scale game difficulty is likely going to stay with us for at least the forseeable future.

This thread reminds me of the 'general' forums in world of warcraft, which isn't a good thing. The top 1% of players urging the company to make the game harder?

I read these forums and I struggle allot with the AI. I regard it as a major achievement to have moved from King to Emperor in G&K. Playing the AI on deity requires SO MUCH knowledge (who is likely to tech what, which wonders are likely to be taken, and in-depth knowledge of game mechanics) that frankly I can see why the rest of the player-base can't be bothered.

The thing is that it is not so much the AI itself that makes the game hard, but rather there is so much going on that you are likely to make so many false decisions so that the AI will be a threat. I am fully in support of people asking for better multiplayer support, but seriously the AI is very intelligent considering many other games out there. I can plan ahead (albeit only to a certain degree) and be vicious, it can backstab and pretend friendliness, it is exploitative and mean. How much more human could it get?

Of course there is room for improvement as there always it, but I don't see this as a debatable 'problem'.

Securion said:

Take a look at the AI in games like Total War and then tell me the technology isn't really there yet...

Click to expand...

Well, Total War would certainly suggest the technology isn't there yet. TW has a simplified, Civ I-IV style diplomacy system that the AI struggles to handle. Its tactics are unsophisticated and barely change with difficulty level - instead, as with Civ, it just gets more units. AI army design is rudimentary and always focuses on the same two or three unit types. It makes little use of agents for much of the game, and when it does use them it can be annoying but exhibits no selectivity in target selection.

It has the same rudimentary understanding that it should flank as Civ V AI, but little more than that - it will charge headlong, will rarely disengage to choose other targets (though on Hard and above it will sometimes do this with cavalry, but only cavalry), will either not keep units in reserve at all or will simply stand around with 'reserve' units while its fighting units are being defeated by enemies with exposed flanks (especially in naval combat, at which the AI is atrocious), it won't use its units' special abilities at every opportunity (fire arrows right at the start of a battle, but never subsequently, for example; and it never uses schiltrons/phalanxes, a particular problem since it's so incapable of responding effectively to enemy cavalry), and it has no idea at all how to execute a siege or to coordinate attacks so that defenders are engaged simultaneously (and the order in which it attacks is heavily stereotyped). AI units will invariably be set to run to the point of exhaustion, and will never withdraw to stop and recuperate (although in Shogun 2 the game pace is too fast for this to matter a lot, still the default run setting means AI units are less capable when they reach combat than they should be). Every TW release is accompanied by fans waiting on a "Better AI" mod.

Tellingly, almost none of this has shown any improvement since at least Rome (although I haven't encountered it doing anything quite as stupid as Medieval II's favoured siege tactic of sending battering rams and siege towers - one by one - to be shot by defending archers while all the infantry and cavalry stands about out of range) despite an increase in AI turn times over the same time period (and, in Shogun 2, a drastic reduction in the complexity of the strategic level). Indeed some of it seems to have got worse - Shogun 2 seems much worse for basic rushes than Rome or Medieval II, where AI units would sometimes actually withdraw if attacked by a superior force, and would actively seek cover to hide in to set ambushes. Empire's AI was so notoriously bad that it's the major reason for the game being considered the weakest in the TW series (and likely the reason Creative Assembly rushed the semi-sequel Napoleon out within a year).

AIs might well be able to do better than Civ V's on technical grounds, but comparing Civ AI with Total War games, which boast what may be the most famously bad AI in strategy gaming, is an odd choice.

dexters said:

Maybe if there's a breakthrough in computing science that allows CPU to consider so many possible outcomes they can appear to outthink a human player. Quantum computers perhaps?

Until then, even with multicore CPUs, you're mainly dealing with linear processing with the various cores assigned to handle other subsystems and menus so the game doesn't freeze IBT and you can still sort of navigate around as it's thinking.

It's certainly conceivable that in the future, 3-4 cores might be required so that the programmers will assign more than 1 core to AI whereby 1 core is always processing in the background even as the human is playing their turn while another kicks in IBT for extra processing. That could yield some improvements. But the 'AI bonus' mechanic to scale game difficulty is likely going to stay with us for at least the forseeable future.

Click to expand...

Yeah my fantasy of having actual thinking AI might be farfetched, but things like AI not firing with their archers(mainly barbs I think), not spending their money, shuffling their units around instead of just attacking etc. I feel should be possible without any leaps in technology.

pezit said:

Yeah my fantasy of having actual thinking AI might be farfetched, but things like AI not firing with their archers(mainly barbs I think), not spending their money, shuffling their units around instead of just attacking etc. I feel should be possible without any leaps in technology.

Click to expand...

Barbs are meant to be dumb, and they've been made slightly dumber (not beeline for unprotected workers/settlers) to make early game expansion a little more forgiving. Though I suspect it benefits the AI somewhat more than human players, but I can see lots of ragequit scenarios being avoided.

The area AI needs the most work is target selection of ranged. When faced with various priority targets, AI likes to spread fire evenly. Against a human player, it gives humans a chance to retreat damaged units and move in fresh ones.

Some people have already touched on the topic: Do we want a good AI or a fun AI?

Today many people don't want a cutthroat AI that plays to win, especially in a sandbox game like civ. Actually, game design nowadays leans more towards always letting the player win, otherwise he will get bored and play something else. Gone are the days when games were made to challenge the player (think old school arcade shooters). Today, the focus is on a "fun" AI that doesn't actually play the game but is there as an entertainer for the player. This is another reason why the AI isn't improving.

Early vanilla civ had a more "play to win AI" on release, but the complaints were loud and plentiful. People found it boring and thought that it behaved irrationally because it would gang up on the leader or just hated the player for even trying to win the game.

Soren Johnson (Civ4 AI guy for the uninitiated) has a good lecture called "Playing to Lose: AI and "Civilization". It can be found here on youtube for people who are interested. It's an hour long though, but really interesting.

^ Thanks for bringing that up. Had watched it before but forgot it existed.

Ankh said:

Some people have already touched on the topic: Do we want a good AI or a fun AI?

Click to expand...

This is what I expect the different levels to be for. Instead of putting the human way behind and giving the AIs tons of bonuses, have the higher levels be better.

As I am watching a war between 2 AIs, 2 more glaring stupidities become obvious. The AI has no idea what a citadel is. Even when in it, they failed to destroy it and just took the hit on all of their units who also just sat there turn after turn. How hard is it for a unit to say "hey, I took damage, maybe I should move"? Secondly, a fleet of empty carriers is surrounding an island city. For what purpose? It is as if the AI has no idea that the carrier has no attack value. Altho it has gotten better at filling carriers, they really should come prebuilt with at least 2 fighters, or attack carriers with 2 jets.

The AI also has no concept of defense. In a war using infantry and artillery, I gave the defending AI 3 x-com. All 3 were dead in 3 turns because they attacked until they committed suicide. Also, the attackers are failing to take a city with 0 hit points left because they wasted all of their melee units attacking the city with no hope of taking it that turn, instead of allowing their ranged units time to weaken the defenses.

And has anyone ever seen a parachute drop by any AI? I have not.

I understand that the AI is never going to outmaneuver most humans, but there are so many simple concepts that would make it so much better. I would prefer a deity level that did all of these simple things than one that simply gets ridiculous bonuses.

The simple reason is money. There aren't enough players who will fork over enough money to make it worth creating with the resources available. My own view is that the same people who say "making the AI better would be simple," are the same people who wouldn't plunk down the piddlest amount of cold hard cash for developers to make it happen, so it doesn't.

I visit here and like to talk strategy. I like optimizing a bit, but I like role-playing a lot more. The AI in Civ is stupid. It was always been stupid. Civ IV AI is notorious for being so manipulable that they may as well be considered player vassals, even when they're not.

I played GalCiv2 at Hard - the highest level at which the AI doesn't get bonuses. It was fun, and it was smart. But it's not as smart as a human. It still has clear and easy to surpass limitations.

I still play Civ V at Prince because I like building cute cities, big wonders, and pretty armies. I'll intentionally let an AI build itself (and even help it along the way) if I want to stage a Medieval or Industrial war with it down the line. I like growing Civs and pitting them against one another while I watch.

At no point do I play it to win, because if I did, I would either quit at turn 20 (when my win becomes inevitable), or I would move up. I do not find it entertaining to exploit obvious AI flaws, the same way I don't poke at the science holes in Pacific Rim. They're there, and they're obvious, but in the interest of having fun, I intentionally blind myself and look the other way.

I don't capture wayward AI settlers. I don't milk the AI for gold (it's too accommodating). I don't lock it into a decision loop where most of its army just sits there to get killed (though I can, if I wanted to).