PTW 1.21f AI thoughts

[dexters]

Just a few thoughts on a few issues. Feel free to pitch in.

All observations based on a special sencario I created called debugtiny.bix It is a continements map. 5 players, with the human player (observer) isloated on an island moat surrounded by mountains. The map is set out to observe AI naval combat. I have put on purpose the Chinese and the Ottomans, two of the studier AI Civs in my map. Each are paid with a random AI civ on their continent.

The only modification to the rules made is the "build often" preference for the Ottomans and the Chinese. Which I hope will create some interesting naval activity in the late game. So far, the Chinese has the largest surface fleet, but that's not saying much. 1 galley and 2 caravels. Anyways, back to some of the unrelated thoughts based on my obsercvation on this tiny map debug game.

Rush Builds:

I'm glad the AI is actually capable of doing rush builds. My interest was aroused when I noticed a Chinese coastal city finish their library in one turn, whereas one turn prior, the counter noted 18 turns left. To test out my theory, I gave myself a lot of cash and contacted the Chinese diplomatically (yes, I established diplo relations with all AI so I can follow their trades and dilplomatic actions. It's better than being totally isolated and not knowing what they are up to).

I gave the chinese 5,000 gold lump sum. Sure enough, the next turn, they spent about 1/2 of it on rush builds. Attached is a picture showing a before and after. Note the Sistine Chapter build. It's ticks down by own turn, but Anyang somehow manages to finish its University in 1 turn. Rush build at work. (screenies attached below)

+1 for the AI.


AI Worker Inefficiency

I've noted this behaviour in another thread. But basically, I'm just confirming that AI needs to be tweaked on this end. Even the powerful AIs are prone to this when they run of out gold. In this game it happened with Babylon. #2 Civ, doing well, but 0 gold in treasury. I have "show all friendly moves" turned on so I see all the moves. Then I saw Xerxes kill off one of his workers, in mid work! The worker was irrigating! Then a few turns later, I spotted the CORE city of UR building a new worker. ::shakes head ::

The phenomenon: The AI will for opt for destroying its workers rathern than sell off its improvements when it has 0 gold in the treasury have a negative income (I can't confirm this conclusively because there is no in Debug mode to access the AI treasury screen). This is not an irrational choice on its own. But on a larger context it is a problematic.

The issue here is, AI workers have a finite amount of jobs to complete and they also happen to be one of the most important units in the game. Killing one of the workers off won't change the jobs they have to do. What ends up happening is that AI kills a worker to deal with a cash crisis this turn, but goes on to build another the next turn, or in two turns because the AI still needs to mine that tile, or irrigate this tile, or road another tile etc.. The argument is therefore that the whole system of killing workers is inefficient Because A) They just lost several turns of work from a lost worker. B) they don't even bother disbanding workers in their cites. C) The workers are almost always replaced by a core city producing another worker. There is ALWAYS the Hidden cost of shield and commerce lost from the population reduction after a worker is produced..

Why is this an issue?
1) It is exploitable. Humans are better at trading than the AI. Let's get that out of the way. There is already the "advisor" trick when trying to figure out the AI's gpt income. One way to exploit the AI is to grab all its income in a gpt deal. Unseen, and most likely unkown to most human players, the AI, even the powerful ones, will inevitably experience negative income as they start building city improvements, and when their treasury hit 0, workers are whacked (mob euphemism .. heh). And I am sure they are then replaced in the not so distant future by new workers built from core cities. It's a vicious cycle.

I had previously hypothesized that this is why AI Civs who are "beat down" early in a disastrous war often have a hard time rebuilding without human intervention or luck. With their shattered economies the AI begin will invevitably begin to kill off their most important assets, their workers.!

-1 for the AI


------------
Before I move on I have some suggestions

Suggestion #1) The Trading AI - Since workers are tradable under the current system we only need a fix that would have the AI immediately shop around other civs on who would buy their worker. At the current cost of 130 gold, (less if its the AI buying given the trade bonuses) we kill two birds with one stone. AI gets some quick cash, and they aren't wasting any workers. It can be seen as the AI borrowing money to consume today and deferring the use of its workers in the future.

Modifications will have to be made if this is going to be the case. Firstly, there is the issue of how to get a worker to the capital city. This is tricky. Short of making the AI cheat and sell its units regardless of where they are, I can't think of a good solution. Perhaps someone might. A second modifiation is that the AI civs needs to be programmed to buy and value workers differently. Under the Trading AI scheme, the AI should randomly shop around all Civs, so humans get more of less an equal chance of being offered the opportunity, but not an exclusive monopoly on buying AI workers. The pro-active approach of "offering" is important here because the AI can make trades within a turn, and thus freeze out exploiters from immediately calling up a weak AI civ to see if they have workers to sell. The trades are done discerely for this purpose.

Suggestion #2) While this suggestion won't exclude the first, I propose a stronger slider flexibility on this issue. Whereby AI will almost always move around its sliders to keep its workers rather than kill them. Slider adjustment should also give them a 1 or 2 turn cash surplus and a small treasury that can allow them to ride through a cash crisis without killing too many units (workers and military alike).

to be continued...

[dexters]

Before Gifting China 5000 gold lump sum

Focus on the date, Beijing and Anyang.

[dexters]

After

[Theseus]

dexters, have you been selected as a beta tester for Conquests?

[dexters]

No. Unfortunately I didn't get my application in on time.

Are you a beta tester?

[statusperfect]

interesting

they are doing a lot of undocumented AI changes

cheers

[Kaos XIII]

Hi!

Great observations as always. I'm now thinking if we, as humans should help out the AI. For example, give them some free Workers, or maybe send some Workers to build their infrastructure.

One thing I already do, is not take all their GPT. What I mean, is that when the advisor says, that the AI can pay 60 GPT for something, I will only take 50 GPT. This saves the AI some money, that it can later use to annihilate me.

[dexters]

Thanks Kaos for your kind words.

I'm hoping someone from Firaxis or Breakway Games realize that a lot of people want the naval aspect fixed to at address at the very least, some of the more atrocious AI missteps. Such as the aforementioned AI loop when landing parties sail back and forth between targets and never are able to land or waste so many turns trying to find a landing spot that the landing doesn't achieve much. With all attacks, naval landings especially, it's all about the timing.

Those landings from behind your lines Soren programmed were neat. If the AI can put on the pressure and repeat a few times and reinforce their positions, it can achieve several things.
1) make the game more challenging
2) make the AI more competitive vis a vis humans and other AI civs
3) increase the depth of the game by rewarding a strong navy. Right now, most people get away without having a navy for most of the game because they know they can survive an AI seaborne invasion. If such invasions became much more threatening, some may think twice.