Your AI Opponent - by Ision

Good article! I want to add one point. The number one contributing factor to a civ doing well is starting food bonsus tiles. If the AI's first city has 2 or 3 food bonsus, that civ tends to do much better.

 

I would consider expansionist civ to lower class, since AI seems to pop goody huts very rarely. As for human expansionist can be extremely good, depending on starting location.

 

That is incorrect. Not only DOES the AI pop goody huts, but it will do so with greater regularity than the human as you move up in level.

 

Ision, I would just like to thank you for all the articles you have writen. For the longest time I could hardly be able to beat Regent. Now after reading your various articles on different civs and the AI's behaviors. I have now mastered Emperor and moving up to Demi-god. Thanks Ision.

 

Very well written Ison. Many kudos to your great research into this topic.

I want to bring up an oft observed but little discussed point however. That is, the game AI does have limited capacities to planning more than one turn ahead. We see this most often in how they place units before they go to war. Sometimes, sending them over fvst distances (several turns) before declaring war.

I'm inclined to believe there is some subsystem at work that allows for these type of planning. Granted a player can reload and save and sometimes prevent and invasion by moving things around, which again would indicate a turn by turn planning as discussed. But that would only suggest the old 'battle plan' was scuttled because of a new development. As another poster noted, the AI's strength and weakness is its ability to abandon plans as things change.

Because a lot of things going on aren't directly observable, like what technology to research next and or why the AI likes to sometimes send fishing parties into your territory (this is different from those units crossing our territory to go to war), it is hard to say definitively how deep this planning extends into the AI decision tree.

I strongly agree that the bulk of AI planning is made in one turn for the next turn with the whole cycle starting again and so on. I still believe that it has some capacity, likely scripted and very limited, in planning several turns ahead. Although the end result could be that the plan is scuttled.

 

Oh, don't give ainswood the idea for a hidden Agri/Sci/Ind Civ (or Agri/expan/ind since Sci seems to be out of favour) on the other continent so it is a real monster by the time people like me get to it!


Oops!

 

Germany has one of the best starts on higher lvls when they get bonus units though, as they get both spears and archers instead of warriors.

What you say about the AI never using the lux slider isn't 100% correct, it is used but not effectively. It makes use of it later in the game atleast, but not enough during the early game it seems, and never enough.

 

To my knowledge the AI never uses the luxury slider - at any point in the entire game. Superior late game performance in happiness is due to a superior AI infrastructure (temples/cathedrals/colosseums). The AI works on a strict and consistent program throughout the game - I cannot imagine that it has 2 separate programs where it will not use the Lux slider early, but then will use it at a later point - the game program is not that sophisticated. I have never heard of an 'event' or 'turn' trigger that would cause the AI to intiate an option it could have used earlier. Besides, why would the game designers handicap the AI in this odd fashion? If you can verify that I am incorrect in this, please let me know where I can access this information.

Ision

 

For some strange reason, my web browers on this computer refuses to access apolyton although I can access it from the compuer in the living room.

I recall Soren addressing this issue in a backdated post on Apolyton's forums. He was talking specifically about the PTW release. He noted that the luxury sliders were made very sticky to make sure that the AI doesn't go 100% lux and neglect trading for luxuries. However, I recall quite clearly that he also mentioned in the same breath that the AI can adjust their sliders.

The default I believe is 10% going up to 20% in times of war. There may be extreme circumstances where its goes up even further.

I'll confirm these numbers when I re-read the thread in my living room computer.

 

Please do so - I always like to expand my knowledge. Nevertheless, if it is as you say a move from 10% to 20% and only during war - then fundamentally for all practical purposes - my contention still holds true - for the most part. The only change would be from using the word 'never', to to the phrase 'barely ever' or the 'lux slider use is insignificant.'

Ision

 

This may sound like a dodge but it's just bad luck on my part.

The thread with Soren's comments on the subject no longer exist, but I'll continue to dig around the other threads in hopes that the reference to this pops up in those threads also.

I would also like to make a correction.

Default luxury is 0% up to 20% in times of war. You can probably test this yourself by investigating cities of AI civs that have been at war for a long time. I hope it checks out. If not, I'll run through a few debug games myself when I get some free time in the coming weeks.

 

Let me suggest one thing: the AI should be renamed to "AS", meaning "Artificial Stupidity".

 

The AI does move around the sliders, both lux and science, but will only move them when starting a new research project.

 

Now this article (finally discovered) drives me into a deep crisis

It's quite annoying, how little stress programmers had put on "artificial intelligence". Those are some quite easy rules - building a settlers & workers dependent on recent (and prospected future) food development.
This AI-failure disappoints me a lot.

That's not state of the art. I don't need cool graphics and music and... but I want skilled opponents, damn it!

I guess I should try some MP-games...

Any hope, that civ4-AI will be better?

 

AI is just fine. Ignorance, Idiocy, Inefficiency, ...

 

As a side-note to AI worker production, I find that AIs are easily 'culled' by robbing them from their worker as soon as you meet them. Often this is a blow they hardly ever recover from. Provided of course you can stave off the attack they'll subsequently launch on you. But often you can negotiate peace before they even reach you.

Of course this doesn't work on the higher levels, where the AI starts of with a considerable army to get you whe you piss them off.