Yeah it's funny how people can report wildly different behaviors between games, and assume their's is the one that holds true across all games. Like, I haven't personally seen a faction wiped out, much less attacked by a monster, since like 1.03 or something. I see the AI attacking monsters to get experience....but I don't see the 80+ monster armies chasing the AI around like they do me.

A lot of it comes down to how players play too. I aggressively scout for all the available food resources early in game and capture them, much like the AI used to do. So in almost every game of any size, I hold at least 60% of the food spawns in the world. That drastically slows down the AI. In almost every game, by the time I find the other factions, my power rating (on a normal game) is 2x the lowest faction and about even with the highest. Then I make some attacks against the strongest AI, knock them down a few pegs, negotiate the start of a few wars between factions, and it's game over for the AI.

Oh you've got me all wrong. My goal isn't to "beat" the AI. My goal is to lock it down and see how it behaves. I never kill AIs, that would ruin the point of what I'm doing. (Plus I like to play Diplomatic empires that just negotiates wars between other nations, backed by a superior military that could crush anything. [insert country here] basically )

So I like to try and find out how the AI behaves when it's only really got proactive options left. It seems now it understands the difference between a resource it can capture with it's ZoC, and a resource it should build for. At least late game. Because I've watched AIs not build new cities to capture resources adjacent to their ZoCs anymore. That's nice. At release the AI would build a new city for every resource that wasn't in its ZoC, every single one. Now it seems to only do that early game, and changes its approach late game.

At first it looks like they're just not paying attention, but I think that's what's going on now. Unfortunately, sometimes they NEED to rebuild immediately, and they don't. Late game an open resource should be a red flag to all AIs who know about it that they should go drop a city on it ASAP. Running ahead of the AI and seeing how long it takes them to expand is part of why I do the scouting too. 

***

You know, I think there needs to be more of those kinds of flags. The pacing in Elemental changes drastically as the game goes on. You want it to be fast and punchy and give quick rewards early on. But you want it to slow down and become more tactical mid to late game. And of course, you want DRAMA in the late game so you know the AI is still thinking, and so you can have the big finish that you've played 2,000 turns to be impressed by. 

But rather than time, maybe what the AI needs to account for when they "shift gears" is faction power rating. Time...that's all well and good, but it's not dynamic. A dynamic AI set up would say "Do this until you've found x new empires. Then switch to this [production mode.] When total faction power = [some number/number of factions], switch to [mid game military production mode]. When a single faction power = really high number, switch to [aggressive attack/defense mode]."

Isn't that how it would sort of work IRL? The smaller "the world" gets, the more hostile everyone becomes in general. AIs should be just as willing to attack you out of fear as they are willing to settle a peace treaty.

Rather than planning one gianormous AI that can handle every permutation of game play, write bundles of AI behaviors that shift based off of easily readable global cues like faction power. Time is a good start, but frequently it won't account for the actual situation on the ground. (I.e. players gets too strong before the allotted 100 turns, or the AI does.)

I think one of the main problems with the AI is the city development phase.

 In both 1.1 games I played I found capital and lvl3 AI cities that were full of research buildings and materials buildings. To the point that nothing else could be built since all 50 city plots were taken. You can understand how crippling that is, since until you research technologies to build command posts, archery ranges, schools, etc, you can fill up your first cities with useless buildings. And when you get command posts and archery ranges and barracks and war colleges, there's no room to build them - unless of course you demolish plots but that is plain stupid and would be almost impossible to code.

They did produce military units - and mind you, the AI did put me in tough spot in the beginning since they invaded me on contact - but they never produced any groups or archers.

This would be very easy to fix with a hard coded safety for the AI, never build more than 2x of a building in a city. I never built more than 2 of the same building in the city and never needed any more materials/research, so i guess it should be enough for the AI as well.

Another problem I encountered was undefended cities. I can understand why that happens since the AI is trying to hunt down things and get XP. It is a pain in the butt keeping my cities defended as well, but I do make a point of keeping at least two parties (that's the four man group) of soldiers in each city. I suggest you hard code the AI to do the same - keep one party of troops in it for every city level (a level 3 city, 3 parties). It won't stop a determined attack but it will be a good starting point. Having the AI marshal it's forces back to town when invaded won't be too effective due to the fog of war and low movement rate of units.

I also noticed the pioneer problem. Encountered more pioneers that would be needed to settle all of the land mass just standing around.

Don't listen to this guy...

Because everything is written in XML, balance is pretty easy to tweak.

The main problem is that the AI behavior is weird and unpredictable. It does not like to commit, and it does not calculate things properly.

-If the AI properly attacked when it has a strong combat rating, we can make it build things that have a strong combat rating, or make the things it builds stronger in combat.

-If the AI built cities with somekind of direction, we could make that direction stronger.

-Tactical AI is dumber than bricks and doesn't use magic. Not sure how tactical AI works or if it's hardcoded.

-The strategic AI doesn't really use magic either, I think.

Right now the AI has no sense of purpose or direction. It's just doing things. I'm not really sure what.

----

Frogboy, I know your AI techniques are secret, but does the AI learn things overtime? Over a period of games? 

From a programmer's point of view, here are things you can analyze, and points to record for reweighting AI priorities

-What do good players make? How many of what buildings? What techs do they take? How many cities do they have? How many champions? What stats get leveled? What factor is distance? What spells do they research? How often do they cast spells, and what do they cast it on? What do they spend resources on?

-Record these statistics in any given game. Record them for the AI and players.

-At the end of the game, arrange the factions according to power level, and who won. Generally the winner and the top two statistics are the most relevant. Enter them into your Baynesian filter.

-calculate, reweight, and try again.

-Having your 'hooks' into these build statistics are really really vital. Just building the 'right' things and 'enough' of them can really make a difference.

PS, I totally applied to work for Stardock 2 years ago.

I started handing out new EXEs to people on #elemental today. I think the first one made a pretty big difference.

However...

[1] The AI is still not getting groups of units fast enough.  Knowing what the AI should do and making it do it are two very different things.

[2] The AI is still not building up its cities in the way I'd like to see it do it.

[3] The AI is still not designing units well enough.

[4] The AI needs to be more flexible on units it can build.

[5] The diplomacy options are pathetic. The AI needs to do much better in this area.

So this is what I'll be doing during the Christmas break. Those of you who are really into this kind of thing can hang out with us on #elemental and I'll be sharing new EXEs from my Dropbox account.

Nvm, I wasn't connecting to irc.stardock.com properly.

You can also reach #elemental from Impulse on the community->Chat tab.

Glad to have you in the channel.  We're gonna have some fun!

I'm playing an island map (Islands of War) and it seems to solve most issues, sadly I still don't see AI using terraforming spells to come at me or even ships.. they sure keep their pioneers in their borders.

Altar out-teched me  in armors by several levels, still took them out with magical weapons. Capitar only ever researches warfare in my game so they may be worthy opponenets in the future . Seems factions use tech properly in my game, maybe because of slower pacing since exploring in island map is a bit harder. I think the turning point may be squads, the AI doesn't seem able to produce them.

I've yet to see AI using magic in the new version outside of familiar summons (I've to meet pariden still, I'm sure they will have more), I would suggest emphasizing the use of magical weapons by the AI (and dispelling them , when not necessary, as it seems intended). If you combine that with lvlup of cities being defaulted to random guardian units and gildar for the capital or cities with gold mines, it could work wonders. It seems the formula has been reworked and it gives more units now, 5 spiders the only time I tried. Right now, I think rushing with those units enchanted with magical weapons, would be very hard to stop early. Also AI doesn't seem to use teleport... it would change things a lot if it did.

Also it could be interesting if the AI focused on taking your capital by surprise, a possible strategy to do this is the pioneer-teleport convo (build city to have influence near them, then teleport).

Eventually the AI even seemed afraid to attack me, I didn't see enemy units trying to raze my improvements or even taking empty cities, they just moved away, which was weird (I had units nearby, that could kill them, that may be why... still razing improvements or taking the city would be worth losing a familiar )

Also most advanced spells seem to take waaay too much mana, which seems to be scarce at least in my game and I'm controlling the map an dhave more tech research then anyone. Ai can't comete in that department, simply. Maybe add some notables that add mana instead of the usual gold ones and/or lower some enchantment costs, that the AI will be likely to use. Another further idea would be to add heroes that generate mana each season.

Now code a way for the AI to effectively use boats and fight on an island map and you can sell yourself as freelance consultant to 2k and Stardock. Well, 2k cheated in latest Civ but you get the general idea

Hmm . Should i buy the game now , or wait for a better AI ?

Oh very similiar game with me, only Pariden turned out to be my big bad boss, I'm sitting at 55 and suddenly I meet 350 Pariden, so I thought epic-ness would ensure. But mostly I just sorta went through the motions, declared war on everything season 100, dealt with teeny sorties they sent after me and picked off the opposing factions one by one, some were nice enough to invest in Adventure Tech during wartime so I could quest on the way and pick up a shiny Sov shield. I had 2 champs with super high intel arcane arrowing everything to death and keeping my hero's health up, and my Sov was solo-wrecking pretty much everything else (combat focused). You would think 1 Sov vs 12-ish creature mobs with ridiculous health would be epic but more like click turn, click turn until the battle was done, rinse repeat, no variation vs AI

Though it is nice to see AI casting spells every so often, really surprises me sometimes even if it has marginal impact on battle outcome.

I *think* there is work being done on an initiative system which would help solve the issue of players having common sense and letting AI move into range and then crushing said AI with all the first strikes.