martin

here's my first attempt, just some colour rebalancing of one of them and moving stuff around etc. I'm going to do a real one later:

But then you need to script an AI every time there is a new unit, and every time someone makes a mod for the game. My way I make a mod, with some new men, then I need to script the entire AI or just run the game a bit with learning and saving or experience active

well mine is a learning AI by definition - because it uses a reinforcement lerning system What you said above about planes being good against men being known by the person who scripts the AI is not true, planes arent always good against men - what if their men are always armed with an AA gun? The entire system would also hae emergent properties to define the entire AI strategy from the simple reinforcement system...

Q1) OK, so say you are invading the AI island with transport ships. Image that the island has "strategic locations" (either: randomly distributed, near collections of buildings or near battle hotspots [probably2]) - so you take a strategic location using your invasion that originated from ships at the time the AI was responding with planes. So the AI now knows that planes aren't awfully affective and will use some kind of reinforcement learning system to learn this. Q2) Well if you atack with ships and the AI responds with planes that's a pretty linear battle (implies only one attack unit and one response unit). So if there is a reinforcemnt system the battles will become a bit more complex, the computer wil learn in this case that responding to approaching transports with plane will not work, so maybe it'll use boats or artillery this time, however when you land say you unload some men and some artillery. Now when the computer attacked your island earlier in the game you attacked its army with artillery and it killed the artillery with archers (say), so now its reinforcement learning will come into play and direct it to use archers. It will also have learned earlier that planes are good against men, so it'll send in some planes, which will probably succeed - if this response succeeds then the learning for attacking artillery with archers and men with planes will be reinforced, however if the artillery gets to a strategic location the the archers will be proved ineffective and a new node will be created for armies of men and artillery (linked to archers and planes with a negative connection and a neutral connection to everything else). If of course the combined approach works then no new nodes are needed. Q3) pretty much answered this one, the AI creates a load of strategic points (around clumps of buildings, near resources whatever) - these points cannot be seen by people but they are areas that are likely to be hotspots. If an army captures a strategic location (destroys the buildings that made this a strategic location, killed the villagers mining the resources that made this a strategic location etc) then the response the AI used is ineffective. Q4) umm... that's answered above at great length and I'm not going to repeat it. Q5) + Q6) eh? rephrase the questions :?

I think we're talking abiout different types of AI - you're talking about unit AI I'm talking about enemy comander AI

well what i meant about learning was not totally concocting new strategies but having lots of different strategies and learning which to apply in certain situations - land invasion from ships, I should respond with planes because that worked and ships didn't; that sorta thing.

I like either random small island maps, mediterranean or structured maps - ones which have a small chokepoint with lots of resources in -forcing constant battle for basic resources.

just a quick suggestion - AIs that negotiate -> I hate games where you send the AI 1000000000wood and it says "hah, your pettty offerings will not wave my determination to crush you empire" despite the fact he only has two villagers and a house left.

I'm afarid I must admit to not reading all the suggestions - so I apologise if this is an old idea. I really hate it when, in games, someone gets shot, takes 50 damage and carries on as usual. I would prefer it if you take an arrow and, say, there is a 20% chance of hitting the body - which kills you, or a 10% chance of hitting either arm - which disables any weapons or shields on that arm etc. SO rather than units being AOK or dead you'd have men limping around with one leg - which means that they cannot run so fast, and one arm - which means that theyhave no shield and take damage from swords much more quickly etc.

Well everyone here seems to be talking bout coding AIs, how about some sort of flowchart for non programmers? I mean stuff like dragging out if food lower than X then do X etc, it would make for simpler programming of the AI. And what about a learning AI? This would mean that you never have to program a specific AI with tactics etc, just play against it. However the real benefiot of a learning AI would be tactics adaption. Whenever I play an AI I can use the same tactic time and again - when I play my brother (a human ) I have to keep coming up with new tactics to evade the defences he has built against my old tactics.