ai engine research project HELP!!

[aiResearch]

hi all!!!

I've been posting around like crazy trying to find out info and have been told by a few that this is probably the best place to go.

Basically, I'm doing a computer science research project officially titled: "The AI engine of Sid Meier's Civilization III".

I realise that the actual soure for the AI is closed and intellectual property of Firaxis, but I'm trying to find out all else that I can regarding how it actually works.

I've piecing things together and am coming up with some results, however any help and or guidance would be much appreciated from those like myself with a keen interest in this area of the game.

many thanks
p

[notyoueither]

How much do you know already? Maybe if you put forth an out-line some of us could help you fill in the blanks.

Have you emailed Soren for any info he would be at liberty to give you? You can't email him directly from Poly, but you could send a note to Dan Magaha and ask him to relay it to Soren. Contact Dan using webmaster@firaxis.com. You might butter him up by addressing him as the Almighty Sombrero.

Anyways. The AI is quite flexible. It's been designed to decentralize decision making processes (or so I've heard). It is still subject to a high degree of predictability. However instances of it surprising the player are much more frequent than with any game I've ever played.

It seems to adapt very well to a wide variety of conditions (Mods). Sounds like it is doing what it was designed to do very well. For instance. My current game I have edited the tech research values to prolong the various ages and to allow me more time with any given unit as being dominant. The AI civs have almost no problem adjusting to these conditions. I haven't seen any obsolete units being used against me yet (aside from some Ancient UUs that have no upgrade path).

Soren seems to take great pride in making his AI as capable as possible of providing a challenge. I just wonder if he is holding anything back. Maybe the final version could have a final difficulty level. The *Soren's Promised Death* level. AI gets no cheats, but gets all of his most wicked strategies. Heh. Love to see that. Kind of like the director's cut.

[notyoueither]

Of course. Soren might have a keen interest in a *research project* to find out just how much we've found out by now. Might help him make adjustments. Right Soren?

It's still April isn't it?

[Dominae]

aiResearch, interesting (and fun!) research topic. I probably haven't thought of this much yet, but here's a few basic points you should consider (here, by "the AI", I mean the Civ3 AI):

1. The AI doesn't do any sort of learning. This cuts off of a big chunk of the AI field (if you don't like learning, you picked a good research topic!).

2. The AI lets the Governors make all the build decisions. In other words, it doesn't explicity determine "what's should I build next?", but relies on a fixed algorithm (the Governor preferences). This is why it will continue building Swordsmen when fighting Tanks; it just told its Governor that there is a state of war, so building the best military unit is probably a good idea. This point demonstrates that the AI is highly component-based.

3. The AI is probably divided into a number of agents, each responsible for a different aspect of a successful game. So, there is probably a "expand" (found cities) agent, a "research" a agent, and so forth. Which agent gets prority at various stages of the game is certainly an interesting algorithm. The AI is definitely not programmed to specialize.

4. Within each agent lies a large set of condition-action rules. This is especially obvious in combat, where you can guess where the AIs units will go (usually chase a random Worker or Scout). These rules (the heart of the entire AI engine) were hardcoded by people who know how to play the game, and play it well. All the nifty little strategies that the AI employs are represented by these rules.

5. As for unit movement, this a straightforward search problem. Based on what the unit is meant to do that turn (explore, attack, defend, settle, improve, etc.), a different value function is assigned to each square. In combat, for example, a high score is given to a severely injured unit with a low defense rating that is reachable this turn.

6. Planning. I'm not sure the AI does any planning. It certainly does no high-level planning ("I'm going to kill the English which will give me a precious source of Gunpowder allowing me to advance on the Persians, etc. etc.). Such plans are the very reason why us humans can "outsmart" the AI.

These are just some things to get you going, in case you're just starting your research. I'm making no guarantees that these points are accurate (partly because I haven't thought about it enough, partly because my opinions are solely based on observations of the AI's behaviour)!

Good luck!


Dominae