About AI

Atahualpa · August 23, 2000, 17:23

Isnt it funny that we dont get much info about an enhanced AI? All we read is how the Diplomacy-AI is better and on and on. But I havent yet seen someone from Activision saying that the Strategical Thinking of the AI has improved greatly.
I mean, CtP's AI is under every civers level. When you choose 32 Civs there may be 2 Civs that are not so far behind, but most are just ways behind you and if it wasnt for their underdeveloped cities you would take them all in one turn.

Now whats with the AI in CtP? Will it be able to actually plan invasions? So that his goal is to take a certain city and that he tries to minimize his losses to achieve this goal by making one huge surprise attack? Is the AI able to make small attacks one one side of the border and after some turns. When it sees you have increased your troop strength there and decreased your strength on the other side of the border attack with a huge force? Can the AI differentiate between a RAID and an ATTACK.
RAIDS for tricking enemies or for pillaging and weaking them or just for beeing a pain in the ass.
ATTACKS for conquering cities.

Will the AI be able to stack ships? Will it be able to do a "Blitzkrieg" style of war. Meaning: First weakening the cities with Planes/Ships then landing in Armor or Fast units that take the cities and then pulling in slower infantry with high defense to defend the cities just taken?

Will there be a pool of different strategies available to the AI, or is it up to the AI to think of a descent strategy?

Can one answer these questions?

ATa

P.S.: I hope the tactical descisions and citybuilding descisions are improved. Not that it builds unproductive cities in large mountainoeus regions or in swampy regions.

St Swithin · August 23, 2000, 21:31

just about all of the AI code was rewritten, so it's difficult to answer lots of specific questions. Let's suffice it to say that it's much smarter and much more conniving, and you can expect much better attack strategies from AI players.

Big Dave · August 24, 2000, 00:29

And better diplomacy, too? Please?

------------------
Big Dave

If pro is the opposite of con,
is progress the opposite of Congress?

Atahualpa · August 24, 2000, 04:08

Big Dave: I am absolutely a 100% sure that they make Diplomacy right!!! I fear more that other AI parts will suffer since so much work is put in Diplomacy. This way we would have a great Diplomatic AI but one that can be beaten easily on the main map and well that isnt much of a challenge.

St. Within: Someone said you used code from ST Armada. Now I've heard that the Armade AI wasnt all that good. It would be good if I could see some strategy behind the computers actions (of course I should not see them when the AI attacks me). Cause in CtP1 the AI stacked his forces always near my territory. So it was easy for me to take out his troops one by one as they marched to the stacking point. Besides I could easily see where he would attack me. I hope he does more troop-movement-hiding in CtP2.

ATa

Nordicus · August 25, 2000, 09:14

Yah, I'm pretty sure diplomacy will be vastly improved. Dunno about Armada and it's code, but I've been playing that for a while and have gotten into some modding with it...and, yeah, that AI isn't that great (CtP's was better, IMHO).

hexagonian · August 25, 2000, 11:19

ATa,
I agree with you 100% - this is really the heart and soul of any game and its replayability factor.

All the bells and whistles aren't going to mean much if the AI cannot use them effectively...

But this seems to be a problem with a lot of games. Once a player is able to figure out the tendencies of the AI, he can exploit them, and the game becomes too easy. I'm finding that out with SMAC now. I still have a ways to go to fine-tune my game in SMAC (and I wouldn't even think of challenging another human player yet), but the AI isn't too bright there either - and based on the most of the gaming publications, SMAC is highly respected.

'Is it possible to program the AI to be flexible and adaptable?' is what it all boils down to. And I wonder if that is truly possible.
[This message has been edited by hexagonian (edited August 25, 2000).]

Atahualpa · August 25, 2000, 13:19

I really would advise making a public or closed beta that lasts about one or two months to playtest the AI and balancing stuff. I dont think Activisions Testers are TBS fans (if you look at the spectrum of rescent games (and some still under development): Vampire - 3D Real Time Action, Battlezone II - 3D Real Time Action/Strategy, Dark Reign II -3D Real Time Strategy, Armada - 2D Real Time Strategy...).

Also IIRC what was said about CtP's AI the time it was released - it was praised like hell. Well, it turned out to be right the opposite. Sure on some of the first games the AI did beat you on Prince or Warlord but after some games the AI doesnt stand a chance, even on deity.

Ata

Daniel Frappier · August 26, 2000, 00:53

An adaptable AI would be nice but in my opinion all of the game that I saw have what I would call an unfinished AI. For example in Civ II the AI would not rush buy wonders or use caravans to that effect, it very rarely buy cities or steal technologies. Those are BASIC strategies. CTP's AI is unfinished in the same way, it doesn't stack slavers, it doesn't stack ships, etc.

Before tackling out advanced nuclear physics, please lets do some simple calculus first.

Azmel2 · August 26, 2000, 16:25

There are three major parts to the AI in CTP2. They are the strategy engine, city management and diplomacy.

The algorithm used for the strategy engine is the same one used for CTP1, but it has been rewritten (for Armada). The purpose of rewriting it is to make it easier to add the refinements that distinguish a so-so AI from a truely clever one.

In the case of Armada, the strategic engine really only came into play during instant action games; the rest was scripted. Unfortunately there was no time to tune the strategic engine like I would have wished since most time was spent on the scenarios. The buildlist system in Armada also was not tightly integrated with the strategic engine and that is responsible for most of the stupid behavior you see out of the AI.

The advantage of using the same algorith as CTP1, and starting with existing code from Armada is that we have had a year to tune and tweak the engine to play CTP. This has resulted in an AI that plays the game more efficiently. It's important to be efficient, so that you have free cycles for detecting and reacting to special cases. An AI appears clever if it can identify and exploit special situations.

The city management system has also been greatly improved. The player can utilize the same AI used by the game (via Mayors) to control their nation. Exposing the AI for player use has made it easier for us to tune and optimize. The ability of an AI to grow its civilization has a profound effect on how good an opponent it is. The files used to tune city management (for the AI or humans) are also completely customizable by anyone who cares to fire up a text editor.

Last but not least, our diplomacy improvements come in two parts. The first is just having a more comprehensive system; mainly new types of proposals, and response options. The second part is a flexible way to write the rules used by the AI for choosing new proposals and responding to your proposals. We're currently in the process of expanding, tuning and tweaking those rules to be rational and to express the desires of our different AI leader personalities.

I'd love to go into more detail, but I've got a game to work on. Believe me when I say we're doing everything we can on the AI front to make your game experience enjoyable.

Best wishes,

-- Richard Myers
AI Programmer, CTP2

August 26, 2000, 17:16

i can hear a voice yelling "what are you doing again on that site Richard! get back to work!"

welcome to our forums Azmel...

TheLimey · August 26, 2000, 20:01

Sounds good.

If more of it is exposed, are you also still using script based hints for it to follow ala CtP1?

There were some limitations in the CtP1 hinting that still could not address some limitations of the AI's general playing style; power projection and regrouping. Have these concepts been considered? They really are at the crux of how a player utilizes miltary power best.

Will suggested stack definitions still be hardcoded? I hope not... One game feature that human players regularly exploited, but the AI did not was stacking a slaver. This gave the human a big advantage. I'd personally like to see some kind of script definition for stacking, so that the inadequacies of the default behavior can be altered later to improve it. Also, the feature would be nice for scenarios and mods later on.

I wrote a fairly in depth post regarding my thoughts on AI implementation in CtP2, that I'll repost here, since its buried, and is maybe relevant to the discussion.

MidKnight Lament · August 27, 2000, 08:11

Thanks heaps for letting us know what's going on. We really appreciate it.

- MKL

Atahualpa · August 27, 2000, 08:35

Thanks for the clarification Azmel!

I hope for a challenging AI.

One in thing in CtP that I didnt like was how the AI handled wars. IIRC there were 4 or 5 stages and they were the same for every war.
Well, but a landwar is different from a sea war and a modern war differs from a rennaissance one, so that you have to use much more combined attacks in the modern times.
Also sometimes I thought: "What a stupid AI, I took their city and only one Machinegunner survived the battle, but very damaged. Just why doesnt he take his stack of troops outside the city and do a counter-attack?"
I am sure he would have beaten me and the city would be his again.

Ata

Peter Triggs · August 27, 2000, 14:15

Azmel2,

I second everyone's thanks. All the info coming out about CTP2 sounds really good, so I was planning on buying it anyway it , but it sure is nice to get some info on the AI.

Being able to access the FLI and AIP files has convinced me that this is the most fascinating - and probably the most difficult to program - part of CTP. I'm dying to find out about what the new code is like but I guess that will have to wait till I get the game out of the box. In the meantime, I would really appreciate it - and I'm sure others would too- if you could recommend a good basic source on how the sort of AI you're using ,works.

Thanks again,

Peter