Rushing AIs... Are you turning on Aggressive AI?

[Andvare]

The problem is that the AI isn't flexible.
Any good player gather some info on his target before attacking, and that includes such things as aggressive AI, AI type (huge difference in rushing/attacking Monty-Gandhi-SittingBull), era, tech level, size and so on. This provides it with enough info on how to take down an opponent, and as the AI is rather stuck in it's way, it's rare that one gets surprised (except perhaps by it's stupidity, like when Monty attacked me when I had almost thrice the power rating).

Aggressive AI only makes the early rush a bit harder, and any human can adapt to that. The rest of the game is easier, because the AI cannot adapt.
Like Belisar said, if you know you want to rush early, you aren't playing optimal, quite the opposite. Optimal play is to play the map. If an AI is too close, yes you take it out, too far away, chances are that it's not cost effective and ,depending on the AI, it might be a future trading partner.

 

[PieceOfMind]

I can tell that I like an early rush but cause I have done it so often I also know very well that it's a situational tactic.
Starting a random game with the intention to use a specific tactic is either variant play or weak play, not optimal play.
And every smart play is exploiting the dumb AI, if you play like an AI, you are supposed to lose on higher levels.
The question is, where to draw the line, but handicaping yourself from turn 0 before you see the map would be something like a poker player that moves all-in before he gets his cards.

But you are not playing human players - that's the point. If you were playing AIs in poker, going all in before you have your cards might be an interesting way to make the game more challenging, but against human players they'd probably wonder what you're smoking.

So is your argument that the AI has improved from Warlords to the point that a fairly early rush is frequently not the best strategy, or optimal play?

Agg AI is just like OCC or any of the other game options - a variant play. It's just with Agg AI there is no mouseover telling you what the variant is about, because when Blake changed it I assume he couldn't change what it's called or what its mouseover was.

"And every smart play is exploiting the dumb AI, if you play like an AI, you are supposed to lose on higher levels."

When did I say play like an AI? Suppose our most advanced chess computers were so primitive they only respond to two human starting moves, and only play well in the game in those situations. If the human chose a starting move not in those two, then the AI would have trouble and the human easily win. The human took full advantage of the AI, but the game was very dull because it was an easy win. If the human instead were prepared to sacrifics his pride for one second, and take a move the AI understood, he might find the rest of the game challenging, interesting and may even get beaten. Does this mean the AI was better than him? No of course not, but because he was prepared to make a play that the AI could start off better from, he found the game more rewarding than simply taking the precise known easy way to take the AI down. I know the example I presented is not realisitic, but surely it gets the point across. And I think you know the point I'm making full well.

This is a pointless argument if the reason we're disagreeing is because we disagree on what the Agg AI setting is meant to be there for. I accept that most people just consider it as a game option. I am trying to put the message out there that one of the people who had a big influence on the setting was of the opinion that the setting should be for those who did early rush. I suppose it is not optimal play necessarily but that's not the point. If it's for rushers, then not rushing is exploiting it.

If your argument is that early rushing is not always optimal play, then make that clear.

Obviously, in an ideal world we'd have some happy medium, where there was perhaps a mix of Agg AIs and regular AIs. But the reality is, for some reason we don't have that. All we have is two different settings that have the AIs optimised for a particular human strategy. If the AI could observe the human in the game and see what he was likely to try, then perhaps it would be more powerful if Agg AI or reg AI enabled from that point.

 

[FlyinJohnnyL]

Bel-this is my problem. How am I supposed to know before the game if I'm going to rush or not? What if I choose Agg AI, and then end up isolated? Or what if I decide I'm NOT going to rush before the game, so I don't choose Agg AI. But then, I end up boxed in with just a couple of cities, and have no choice but to rush. Perhaps if you could flip it off after the first era or something, I don't know.

 

[PieceOfMind]

Like Belisar said, if you know you want to rush early, you aren't playing optimal, quite the opposite. Optimal play is to play the map. If an AI is too close, yes you take it out, too far away, chances are that it's not cost effective and ,depending on the AI, it might be a future trading partner.

I don't disagree with that. If you think that is what I disagree with then you are missing my point.

If people are just going to keep responding telling me optimal play is about playing to the map, and adapting to the situation, the point is lost.

I admit it is a subtle and tricky to handle argument I'm juggling with here, but a few of the posters at least have made it clear they know what the point is, probably because they too were part of the debate nearly 2 years ago.

Seriously if another poster tells me what playing optimally is, I'm going to have to just delete the OP. It's too difficult to keep having to explain the argument to each person that comes along and challenges what they believe the the OP is trying to make.

 

[Andvare]

I don't disagree with that. If you think that is what I disagree with then you are missing my point.

If people are just going to keep responding telling me optimal play is about playing to the map, and adapting to the situation, the point is lost.

I admit it is a subtle and tricky to handle argument I'm juggling with here, but a few of the posters at least have made it clear they know what the point is, probably because they too were part of the debate nearly 2 years ago.

Seriously if another poster tells me what playing optimally is, I'm going to have to just delete the OP. It's too difficult to keep having to explain the argument to each person that comes along and challenges what they believe the the OP is trying to make.

I do get your point, but my point was that I disagree. However, if you don't want to get "optimal play" responses, you really shouldn't say something as generally untrue as this:

Generally it means rushing your neighbours near the start of the game for a land grab, giving you that early edge.

Even if that isn't your own opinion of what optimal play is, it isn't what the mantra is, which is "play the map".
But enough of that, as you said, it isn't what the question was.

I am a warmonger, and as such play very aggressively. But I don't find aggressive AI to do anything but make the AI less flexible, which is bad.
Yes, it does hamper the early rush, which I do enjoy, but at too high a cost.
In the end, all it does is change the map. It makes being boxed in harder, but everything else easier.

Of course, difficulty level might affect this. I play on emperor (or monarch if I just want to relax), but I suppose that an aggressive AI could change the playing field on higher levels quite a bit. Dunno. (I did a deity once, just to see how broken an Inca rush against Gandhi would be, now that was silly).

 

[lilnev]

I play Agg AI on Immortal with large maps (usually fractal, random settings). I like these settings because they generally provide a game that is very challenging right to the end.

Early rushing is seldom a good idea -- unless I'm badly boxed in or the setup is just perfect. My first war is, on average, with trebs, though it varies a lot (sometimes axes, sometimes not 'til rifles). So, I claim land as best I can, which is not nearly as fast as the AI. I try like hell not to get DOWed, or to survive if I do. When I get a military opportunity, I go for it, but the resistance is stiff and the costs are high.

Yes, the tech rate is somewhat slower -- Liberalism might still be available in 1200 AD. But this is where the large map comes in. Someone somewhere will get nice land and be teching along at a proper rate. And it's the rate of the tech leader that matters more than the average rate.

And then there's the snowball effect to keep the late game interesting. More AI-AI wars means more lopsided outcomes -- someone gets bigger, and someone gets vassalized. Which makes the next AI-AI war more likely to be lopsided too. Have you ever faced Hannibal, with far more than his original land plus three vassals, in a modern war?

So, just my experience, but it's a setting that provides me with challenging play right through to the modern era. And I disagree that, "If it's for rushers, then not rushing is exploiting it." It's a setting that makes war more difficult in all eras, makes AI DOWs a lot more dangerous, and encourages AIs to vassalize each other, but it also slows the tech race. If that makes for a more enjoyable game, go for it.

 

[bestbrian]

...under revision...

I wish that you'd left the OP and then revised it in a separate post; I'd have liked to have had the opportunity to read it in light of the responses that followed.

 

[Airefuego]

At this stage, measures like putting Agg AI on may make the game 'fairer' or 'more realistic' in some people's eyes, but they don't address the real problem - ie that AIs are fairly inflexible and appear unable to make the sort of complex long-term strategic decisions that humans can.


I guess the real answer is (eventually) to develop a Near-Human AI - ie one that will play the map, rush aggressively sometimes - if the situation warrants it - be flexible, plan long-term, etc etc.


I also realise that is a very very difficult ask and not likely to be achieved for Civ IV. Fingers crossed for Civ V or VI Also, while pipe-dreaming of a near-human AI, I want to break free of the usual endless bagging of the AI - after all the Civ IV AI (esp BTS) is undeniably awesome when compared to AIs in previous Civs. So I'm grateful for that, and for the continuous improvement that is going on

 

[TheMeInTeam]

At this stage, measures like putting Agg AI on may make the game 'fairer' or 'more realistic' in some people's eyes, but they don't address the real problem - ie that AIs are fairly inflexible and appear unable to make the sort of complex long-term strategic decisions that humans can.


I guess the real answer is (eventually) to develop a Near-Human AI - ie one that will play the map, rush aggressively sometimes - if the situation warrants it - be flexible, plan long-term, etc etc.


I also realise that is a very very difficult ask and not likely to be achieved for Civ IV. Fingers crossed for Civ V or VI Also, while pipe-dreaming of a near-human AI, I want to break free of the usual endless bagging of the AI - after all the Civ IV AI (esp BTS) is undeniably awesome when compared to AIs in previous Civs. So I'm grateful for that, and for the continuous improvement that is going on

IMO, the settling algorithm needs a look, and it needs to be set (permanently) to emphasize hammers, commerce, or great people (based on the tiles the AI finds). If someone were able to just program that aspect the AI would get a lot better.

From there, it could be made further dangerous by having it behave erratically. Sometimes it will whip axes like crazy and hit someone (not necessarily the human) when it has copper, sometimes it just uses them for garrison like normal, etc.

The above would make maps more challenging in two ways:

1. The AIs would be less predictable even while they handled their land better.

2. Such erratic behavior would have a higher chance of seeing one AI get in front of the others, and large empires are much more challenging to deal with in many cases.

That might be as good as you can get with simple algorithms, but I bet it would play considerably better than it currently does. If only I knew how to program .

 

[lilnev]

I hope we can agree that the AI for BtS is:
1) Much better than past versions of civ.
2) Still not very good.
3) A Hard Problem^TM. Actually harder than that. Chess used to be the Hard Problem^TM. Now it's Go. And Civ is far less amenable to the brute force decision-tree approaches that work for those games.

Actually, if I were independently rich and thus unemployed, I'd like to see what I can do to improve it. I'm not, and won't, but if I were to start it would be with worker control. How many times have I captured a size 10 city with 4 improved tiles? A competent human playing with automated workers is at, at least, a one level handicap. That tells me there's at least one level of AI to be gained by properly controlling workers.

 

[Airefuego]

IMO, the settling algorithm needs a look, and it needs to be set (permanently) to emphasize hammers, commerce, or great people (based on the tiles the AI finds). If someone were able to just program that aspect the AI would get a lot better.

Yes. I agree that the AI needs to improve dramatically its city planning and specialization. It constantly makes questionable decisions about settling, land use, wonder and Nat Wonder selection, etc.

Plus its planning should include cities that it doesn't have - or doesn't have yet. For example - human decisions might sometimes include:
- Recognising that it has no hammer-heavy cities, and settling accordingly even if there are 'better' spots available.
- Not building NEpic yet, if an enemy AI capital has uber-seafood and could be captured soon.
- Long-term planning of forest use (eg leaving one city forested for NPark).
etc.

From there, it could be made further dangerous by having it behave erratically. Sometimes it will whip axes like crazy and hit someone (not necessarily the human) when it has copper, sometimes it just uses them for garrison like normal, etc.

The above would make maps more challenging in two ways:

1. The AIs would be less predictable even while they handled their land better.

2. Such erratic behavior would have a higher chance of seeing one AI get in front of the others, and large empires are much more challenging to deal with in many cases.

That might be as good as you can get with simple algorithms, but I bet it would play considerably better than it currently does. If only I knew how to program .

Yes I agree (although I would not use 'erratic' - which implies errors - rather i'd say 'less predictable').

Most importantly, I would like to see the AI's strategic decisions made often based on needs, opponents and the map, rather than mostly on 'personality' which seems to be the current model.

For example:
- If an AI has a neighbour which is Mansa Musa on floodplains, it should often consider an early rush even if it is Gandhi!
- If isolated, it should know it can cut back on garrison troops until at least optics - especially if it has the GWall.
- If it has no metal or horses, it should know effective strategies to survive.
Etc etc.

This does not prevent the AI leaders also having personalities. Just as many human players prefer to war, or prefer to use a peaceful religious strat, so can the AI leaders prefer different things. But - like us - they should be capable of varying from their core personalities more often - especially if the situation begs for it.

Monty may well prefer to war, but he doesn't HAVE to ALWAYS suicidally attack his vastly-more powerful opponent. Imagine a Monty who could never be trusted but who, just sometimes, peacefully beelined Rifles and deliberately cultivated a couple of loyal allies before DOWing you. Now that would be a dangerous opponent.

 

[PieceOfMind]

I wish that you'd left the OP and then revised it in a separate post; I'd have liked to have had the opportunity to read it in light of the responses that followed.

it will be back up in a moment. I really did need to revise it, but the OP will still be contained in the post.

 

[PieceOfMind]

I do get your point, but my point was that I disagree. However, if you don't want to get "optimal play" responses, you really shouldn't say something as generally untrue as this:



Even if that isn't your own opinion of what optimal play is, it isn't what the mantra is, which is "play the map".
But enough of that, as you said, it isn't what the question was.

I am a warmonger, and as such play very aggressively. But I don't find aggressive AI to do anything but make the AI less flexible, which is bad.
Yes, it does hamper the early rush, which I do enjoy, but at too high a cost.
In the end, all it does is change the map. It makes being boxed in harder, but everything else easier.

Of course, difficulty level might affect this. I play on emperor (or monarch if I just want to relax), but I suppose that an aggressive AI could change the playing field on higher levels quite a bit. Dunno. (I did a deity once, just to see how broken an Inca rush against Gandhi would be, now that was silly).

Thanks for your reply Andvare and thanks for taking my post to you maturely and making a response that helped return the discussion to where I intended it to be.

Your opinion on how Agg AI hampers the AI further, even with the early rush, is particularly what I'm interested in hearing, just as much as I am interested in hearing those who believe it does pose a greater challenge for the early rusher.

I guess that if the mantra "rush your neighbour" is now so unwise in BtS that you only do it in some games (maybe less than 50%?), then the whole point of Agg AI is almost lost. I would tend to agree that, back when the controversy first started in 2007, the typical strategy of rushing your neighbour in (pretty much) every game was becoming far too widespread that I welcomed the intended changes for Agg AI. If the AI has advanced to the point where only in some games (maybe less than 50%) is it best to rush the AI, then the point of having a setting for rushers is lost, because there would remain far fewer rushers anyway.

Assuming we are not talking about the rare cases of isolated starts or surrounded by PRO and AGG civs, do people contend that not attacking their AI early is now often not the best strategy? Perhaps I am too naive but I still feel that rushing AIs in the maybe 90% of games that it is possible is still a very very strong strategy. After all, people in the forums (in polls) do seem to still have the obsession with the praetorian, but they probably hardly represent the high level players or at least the players who play at difficulties where AIs win as many games as they do.

 

[Airefuego]

Assuming we are not talking about the rare cases of isolated starts or surrounded by PRO and AGG civs, do people contend that not attacking their AI early is now often not the best strategy? Perhaps I am too naive but I still feel that rushing AIs in the maybe 90% of games that it is possible is still a very very strong strategy.

Yes. IF I start close to a neighbour - especially if they have better land than me - I will always look to rush. Unless I have no copper (in which case I MAY rush with horses or iron - if I have them - but it's riskier - so the need must be more pressing).

Predictably, the main considerations are: proximity to neighbour, and copper (or having an early UU that is good for rushing)

 

[lilnev]

Well, I seldom rush. It's gotten a lot less profitable since Warlords, and more so with Agg AI. I pretty much need a perfect setup (adjacent capitals plus Copper) or no other choice.

On the other had, Praets are still awesome. Macemen available with IW, and for far fewer hammers? (minus the advantage vs. melee troops, I know. I don't care). They only go obsolete with Rifling, and sometimes I wish I could still build them then....

 

[Andvare]

(Snipped for space)

Well, with Praets, it isn't really a true rush, as iron working is a bit far off into the tech tree, but they are so strong that they don't care. They are so strong that it's not until rifles that they are obsolete, which means you can build an army early, and use it for practically all of the game (it's often decided won will win before rifles).

You can also predispose the map towards early rushing, high water, many AIs, that sort of thing, and in these circumstances agg AI is probably the best choice for "level playing field", however, you could also do the opposite, low water with standard AIs (never have I used fewer than standard AIs on any map) and the like.
An early rush is only economically wise if the enemy's capital is close by (rushing is IMHO a crap for capitals), if it's too far away, you need sailing and/or a huge road network, which might not be feasible, and you will take a research hit that is too big to justify the hammers wasted on dead units.
Of course, with an annoying neighbour like Monty, you could say that taking him out early would be good as well, but the warmongers tend to build a lot of units anyway.

If I start close to an AI, I will rush him (though if it's the evil version of protective, it might not succeed). But I prefer making the map less guaranteed rush than agg AI. Agg AI can be fun, but more for those many wars.

(Boy, I am too deep into writing my project, I even organize 'net responses the same way )

 

[nullspace]

I have to admit it feels very unnatural to not play in the way that would make you most likely to win. It can feel so unnatural that a competitive player just won't bring themselves to do it.

For a good game against aggressive AI, make it your goal to win by domination or conquest as early as possible (also a good way to get a high score). The optimal path for that goal is to war early and often, but choosing aggressive AI makes it less optimal and allows other strategies to be viable.

On the other hand, if you're playing a big archipelago map where teching and trading to a space victory is a stronger strategy, then use normal AI.

By picking between the two AIs, you can force the computer players to adapt to the map or your preferred strategy. Aggressive AI isn't better, it's just different. If you can easily beat both types of AI, that just means you need to move up a difficulty level.

 

[Belisar]

If your argument is that early rushing is not always optimal play, then make that clear.

I wanted to clearify this but Andvare has already pretty nicely covered the subject. You made an assumption that is not agreeable (rush always best), so the discussion was pretty much guaranted to strive in that direction.

My view is similar to TMIT, we would need a more erratic AI that is less predictable. That would be way more enjoyable and challenging than restricting yourself to a specific strategy before the game starts.

 

[bestbrian]

I wanted to clearify this but Andvare has already pretty nicely covered the subject. You made an assumption that is not agreeable (rush always best), so the discussion was pretty much guaranted to strive in that direction.

My view is similar to TMIT, we would need a more erratic AI that is less predictable. That would be way more enjoyable and challenging than restricting yourself to a specific strategy before the game starts.

And I also agree with the idea that the AI needs both better city settling and worker usage algorithms. Then again, Firaxis has had four years (?) to play with this code - I would expect improvements in Civ V (once they work through the inevitiable first year of a buggy unplayable mess).

 

[dirtyparrot]

They are starting to make games that can pick up on your habits and adjust to them. I'm sure that it's a way bigger task programming the AI to adjust on a strategy game, but it's something worth exploring. If the AI had access to how you played over your previous games, that would be great. So, if you kept rushing the AI, eventually he would adjust and that strategy simply wouldn't work anymore. It would force the player to adjust to his situation and not try to employ the same cookie cutter strategy over and over.