Solver's Unofficial BtS Patch

[Kalimakhus]

Well from what I read in the Apolyton thread Agressive AI is more challenging in a militaristic sense. While normal AI can still be quite challenging especially on higher difficulties. I guess a balanced builder/warrior strategy can beat the Aggressive AI as it loses the tech lead with concentrating on building more units and getting involved in more wars. sveint post proves this to be right.

What I actually hope for is an AI that is challenging on Noble difficulty. i.e. without any unfair bonuses (which by the way are easy to beat with several techniques). It should need quite a lot of work for even to get somewhat near it.

 

[Solver]

You could say that. On the other hand, Aggressive AI really doesn't allow you to neglect a military. With regular AI, you can get away with a small army if pursuing a builder strategy and you have no psycho neighbours. With Aggressive AI, a capable army is a must.

In the end, how challenging and fun each setting is comes down to your playstyle.

 

[Kalimakhus]

Anyone here thought of trying to implement a learning AI? An AI that accumulates experience and use it later games. I think civ4 is a good environment to experiment with this idea. AFAIK it is the most moddable game so far. Allowing access to the code controlling the AI decision making.

One may think of an IA that adapt to counter the player style.

How much do you think it is possible or how far such approach can go?

 

[DaviddesJ]

What I actually hope for is an AI that is challenging on Noble difficulty.

Prepare yourself to continue hoping for the next 20 or 30 years.

 

[Solver]

That's extremely complex. Learning algorithms are difficult anyway. For Civ4, this would not only entail a significant move away from how the AI is coded, there's also the problem of experiences not being exactly translatable.

How do you get experience that can be carried over to the next game? The next game will have different circumstances, so maybe the previous experience won't be applicable. Etc, etc. - there are loads of problems.

 

[TheLastOne36]

interesting i like this idea, i have a question though, how will the multiplayer effect this?

for example will it work in multiplayer as you have mentioned, it can be worked with the 3.02 and 3.03 patches.

And will the Ai in a multiplayer game use it? even if not all members of the game have this patch?

Also if the above does not work, will it work if your the host?

 

[Solver]

For multiplayer, everyone must have the same files. That means everyone must have the patch - or not have it.

 

[IronCrown]

All players must use this patch, otherwise you won't even be able to connect.

 

[Kalimakhus]

That's extremely complex. Learning algorithms are difficult anyway. For Civ4, this would not only entail a significant move away from how the AI is coded, there's also the problem of experiences not being exactly translatable.

How do you get experience that can be carried over to the next game? The next game will have different circumstances, so maybe the previous experience won't be applicable. Etc, etc. - there are loads of problems.

Well, I know it would be so complicated. However I think of some generalizations. Something like measuring thresholds for the player's behavior like he will go to war when he gets this or that tech. Or he lawys beeline to this or that tech to found this religion. Something like the personality profiles we have for AI leaders. Actually most players have fixed tendencies that they won't change.

Anyway, I would just satisfy myself with something simple like that an AI can be able to notice that I am building-up right on his border where he can see me stationing unit after unit. I actually do this all the time and still I notice no reaction.

 

[Solver]

And the problem with that would be that the AI would fail spectacularly if you change your strategy. You attack after Iron Working 6 games in a row. 7th game, the AI prepares for it, while you build every Wonder in the era instead. If the AI tried to profile your personality, players would exploit that, and the results would be far worse. Even exploits aside, it would fail miserably against players that learn and change as they do.

 

[CivAgamemnon]

No computer algorithm can out think a human in a game as complex as cIV. With Chess and Checkers, there are known parameters, and an finite number of strategies.

With cIV, it would require something orders beyond what any computer can do right now, so I'd say asking for something that can give you a run for your money is rather out of the question at the time.

 

[Innawerkz]

That might be effective in a one city challenge, but is very inefficient to increase the culture rate in your entire empire to counter the loss of happiness in one single city. If you do that for a few turns, you're effectively losing an entire turn of research of your entire empire.

Restating: I agree to an extent.. If it was a piddly ice city, let them 'die' and watch the population slip a few points. However - if it was my capital running bureaucracy - those multiple unworked mature cottages can become quite a hindrance to your overall game-plan.

Depending on my 'excess happiness' available through luxuries (or available luxuries for trade) I may only have to move my slider up one notch for 4 turns. Recovering from 2 or 3 population points on a high-value city on Epic or Marathon speed is where I base my logic. They can hit my cities ad infinitum with the and I will maintain. Can't say the same for multiple .

How much of a EP cost would you suggest they differ?

 

[Backpack]

I already have a Learning AI in CIV.

His Name is Blake.

No need to wait another 20-30 years.... just wait till the next patch.

 

[Roland Johansen]

Restating: I agree to an extent.. If it was a piddly ice city, let them 'die' and watch the population slip a few points. However - if it was my capital running bureaucracy - those multiple unworked mature cottages can become quite a hindrance to your overall game-plan.

Depending on my 'excess happiness' available through luxuries (or available luxuries for trade) I may only have to move my slider up one notch for 4 turns. Recovering from 2 or 3 population points on a high-value city on Epic or Marathon speed is where I base my logic. They can hit my cities ad infinitum with the and I will maintain. Can't say the same for multiple .

Maybe it has to do with the number of cities we have in our empire. I usually play on a huge map and one city just is inconsequential even if it is the capital which was developed as a great science city and I'm using bureaucracy. If I have 15-30 cities, then I can't just use 10-20% of the commerce of all of these cities to save a population point in one of them.

However, if you play on smaller maps and you have 7 cities and the capital is hit and will lose population and it is a hugely important city in your empire, producing say 40% of the science, then I can imagine that you take this action.

Still, I would even let those citizens die in that situation. You'll lose one pop in your capital and that will be say 1 out of 12 and would maybe contribute 2% of your science output (that could be true in a small empire). But you need to set the culture slider at 10-20% for several turns to save the guy. I just do the cold calculations and I let him die. I'm a calculating ruler.

There might be some situations where I would choose to use the culture slider. For instance if one or two turns at 10% would save a citizen in a hugely important city which (re)grows slowly, then I would consider it.

How much of a EP cost would you suggest they differ?

30-50% higher cost for the foment unhappiness than the poison water supply. So when the poison water supply costs 200 EP, then the foment unhappiness would cost between 260 and 300 EP. It's hard to balance it exactly right. The difference in effect between these 2 missions depends a lot on the city that is being hit. But still in general, I would guess a 30-50% increased cost for the foment unhappiness is about right.

 

[kniteowl]

I'm just wondering if this is an exploit, If I'm pursuing a Religious Diplomatic Victory with the AP wonder and I need to Spread the AP religion to at least 1 city for all the Civs in the game but one Civ is running Theocracy so I Can't spread the religion to that Civ.

Now here's the Exploit, instead of Using a Spy and forcing it to Change Civics I just gift them my missionary and they spread the religion for me and I Win the Diplomatic Victory without wasting 1 EP point.

An Easy fix would be to make it that Missionaries can only be gifted to Permanent Allies.

A more complex Fix would be to Make the AI be more Aware of the AP Diplomatic Victory and force them to Spam Missionaries to their highly populated Core cities to deny you the Diplomatic win if they can determine you're nearing a AP diplomatic win. Forcing you to play Diplomacy better.

I have never seen an AI spy force Civic changes on me or other AIs. I can see it being a very useful tool in harming the Economy.

For example IF I'm running a Cottage Economy and my Opponent is running a Specialist Economy, I Force him to change out of Caste system using a Spy, therefore slowing his tech rate for 5 turns 6 turns including anarchy and the same can be done the other way round, If I'm running a Specialist Economy and My opponent is running a Cottage economy, I can force him out of Free Speech with a Spy and he loses 2 commerce to all his town for 5-6 turns.

The same can be done during wars, If I'm fighting defensively I can Force the AI into Pacifism if I'm running it or force the AI out of Police State if I have the Statute of Zeus therefore making war costly.

And the same can be done with The Economy Civics with Corporations, if there's a small Civ dependent on Corporations for extra food/hammers and I Force them into into State Property for 5 turns they stave and lose the extra production. If there's a large Civ running State Property I can force it into Enviromentalism causing maintenance to spike and therefore slowing down their tech rate.

Just a couple of ideas to improve AI uses on spies

 

[Roland Johansen]

Note that the change civic missions are useless on a civilisation with the Christo Redentor Wonder.

I personally dislike the 'change civic' missions. They seem far too powerful for what a spy should be able to do. The effects of a spy should be local and not effecting every citizen in every city in a whole empire. But that's just my opinion.

 

[jonpfl]

Solver, I remember reading that stealth destroyers cannot defend while in a stack. Is this something you could change in your patch?

From what I read, destroyers upgrade to stealth destroyers and then they are useless in a stack since they will not defend.

Do you know if this is intended?

Thx
jonpfl

 

[46852]

That's extremely complex. Learning algorithms are difficult anyway. For Civ4, this would not only entail a significant move away from how the AI is coded, there's also the problem of experiences not being exactly translatable.

How do you get experience that can be carried over to the next game? The next game will have different circumstances, so maybe the previous experience won't be applicable. Etc, etc. - there are loads of problems.

I recall that XCOM:Apocalypse had a "brain" file where it wrote what it learned from previous tactical fights with human player, and then used new tactics in future fights. I'm not sure if I'm imagining this but I think this was how it worked.

This kind of strategic/tactical memory mechanism that'd flawlessly work in grand scale strategy game would require huge amount of coding and endless hours of testing, but I think it might eventually work. I'm not saying it's the ultimate solution for an AI to use though

 

[Solver]

Solver, I remember reading that stealth destroyers cannot defend while in a stack. Is this something you could change in your patch?

From what I read, destroyers upgrade to stealth destroyers and then they are useless in a stack since they will not defend.

Do you know if this is intended?

Thx
jonpfl

I could change it, but I believe it was intentional, so I'll leave it alone.

I recall that XCOM:Apocalypse had a "brain" file where it wrote what it learned from previous tactical fights with human player, and then used new tactics in future fights. I'm not sure if I'm imagining this but I think this was how it worked.

This kind of strategic/tactical memory mechanism that'd flawlessly work in grand scale strategy game would require huge amount of coding and endless hours of testing, but I think it might eventually work. I'm not saying it's the ultimate solution for an AI to use though

XCOM tactical fights are simpler than Civ4 games, and thus provide better learning material for the AI. Forget a learning AI - Civ4 can't even give you an AI with a memory because of limitations of modern technology. The Civ4 AI has very little memory. Now, it'd be great to have it remember enemy troop positions and movements over the last few turns, and use these movement patterns to analyze potential threats better. That can even be coded. However, it would stretch the system requirements too far - memory usage would increase by a good amount and more importantly, turn processing times would increase a great deal through the use of such algorithms. And 99% of players aren't willing to wait 5 minutes between turns even if it gave them a much better AI.

 

[Blake]

The problem with a learning AI is it's slow.

You train it for 10 months, and it takes 10x longer to play turns, and it STILL hasn't "stumbled upon" the optimal strategies which are fairly trivial to derive mathematically.

An AI which learns personality is certainly interesting, but it's just WAY too easy to exploit and would also make game comparison more difficult, a strategy which works for one player may not work for another, because their AI's are trained differently . Given the inherit degree of "willful honesty" required, it'd be easier to just provide a set of tuning parameter sliders for the player, make em more aggressive/honorable/unit spammy etc, by era if you want. It'd work better.