Yes, obviously finding where to put the time in to make the most useful changes is difficult.

However, if you were to focus on the most meaningful tasks first, you could probably make some significant improvements.

For example, if you were to focus on effective preparation, stacking and coordinating in groups, you would see significant improvements.  If the AI would be prepared to fight, it wouldn't have these largely undefended cities on the border while all it's forces are left behind in it's other cities.  If the AI would consolidate it's weak stacks into strong stacks, the player would have no weak stacks to pick off.  Finally, if the AI coordinated it's stacks together, either to all attack one city or to attack two separate areas at once, you'd give the AI the ability to make effective attacks.

All of these would remain no matter what other aspects are changed.  They are fundamentals of strategy and, thus, even if you changed other aspects of the game, you wouldn't lose the improvements.  Also, although my programming ability has collected much dust, I don't think these would be incredibly difficult to implement.  It just requires the AI to realize it needs to keep its units stacked together, rather than separately, along the borders, rather than in back cities, and a timing sequence so they can attack together.

Anyways, just my 2 cents.

Bump.

I'm mining this thread today for AI updating. So if there's something specific not mentioned here, go for it.

Hi Brad,

Watch Episode #1 and #4 for some of my thoughts on AI improvement  https://forums.elementalgame.com/445569/page/1/3370765.  Here are a few:

1.  I'm able to trick the AI into stop moving its pioneers simply by putting one of my units nearby.

2.  Some AI units get stuck/frozen on a single tile for many turns.

3.  Teach the AI to avoid being swarmed in a battle

4.  Do resistance checks.  Don't cast spells or abilities on units that are immune.

5.  AI spiders should cast web on melee units not ranged units because it has no real effect on range units.

6.  Have Yithril units go Berserk after they move so that they get the extra 2 movement points.

7.  Have the AI re-evaluate its location before it builds an city or an outpost just to make sure there's not a better location 1 tile away.

8.  Have the AI build the outpost or city at the end of the turn after it moves... don't wait till the next turn.

9.  Have AI sovereigns stop buying so many greaves and horses and wargs.

10.  Don't let the AI cast Wither on caravans.  By the way, the AI is doing a great job casting strategic spells now.

Want a few more? 

11.  The AI is little too passive on the higher levels about declaring war on the human player. 

12.  The AI is declaring war on too many people at once.  I know Yithril likes war but declaring war on everybody just means that he will soon be out of units and easy for the human player to walk over.  For example, have Yithril declare war early and often, but focus on nearby opponents (sometimes I see them in wars against people on the other side of the map or with everybody on the map).

13.   If the human player does not give into tribute demands, that should be very upsetting to the AI.  "I'm going to give you one more chance... give me 62 gildar or else I pry it from your cold dead hands!"

14.  When at war, production should focus on units and defensive structures.  I see the AI still building wells, studies, wonders, etc., even as I approach.

15.  Encourage AI sovereigns of the Mage class to choose Summoning traits.  I would love to see the AI using summoning spells in battle.

I'm assuming the reason you didn't write them was to save time, though ironically if you have to go back through and figure them out, it may have been better to create them in the first place. How many other staff members than you worked on the AI?

Glad you enjoy working on the AI so much, though I'd also love to see/hear from any other AI coders at Stardock, too.

I've posted before on a few different things, notably that the AI should concentrate more on defence when its on the defensive (building units, not moving units out of cities when there's a scary stack nearby). I've also had a chance to observe the offensive AI a bit now and it also seems to do some strange things. Normally I crush AI units before they get close to my cities, but I was caught slightly off-balance by the event which means everyone goes to war with everyone else (blood season?). So I had a chance to observe what it did with large stacks in my borders.

1) Strange thing number one: large AI army sat two tiles away from an undefended city and did not move. I think it was two tiles through forest, so two turns away, but even so I was surprised it did nothing. I could understand heading for my capital (which was about five tiles away and not well defended), or maybe heading back to its own territory, but not moving seemed very strange.

2) Strange thing number two: I lost a weak army to its stack when I attacked with a few garrison troops I'd managed to round up, underestimating the strength of the AI stack. Again, this left the nearest city undefended. My best stack was otherwise engaged so I was expecting to lose the city, but the AI walked right past it. At least the stack was moving this time, although I'm a little puzzled where it was going. It may possibly have been heading for my capital, but my capital was something like twenty tiles away.

Assuming it was heading for my capital, I can see this can be a good tactic if the capital is nearby, or (if you want to make it more complicated) maybe if the defending country doesn't have roads. But if the defending country has roads, it's going to be almost impossible for the AI to win a race, it's just giving the defenders time to organise. In general I think it should give more priority to thinly defended nearby targets, and it should re-evaluate this on a turn by turn basis. So in other words if the city's defenders kill themselves on the stack it should decide whether it can now take the city.

Even if it razes the city after capturing it that would help it win a war of attrition. Alternatively it could decide to garrison the city or try to take another city. But just bypassing the city seems bad strategy.

If it does some "target of opportuniy" evaluation to target things which are of benefit and not defended, that might help with its ability to target outposts and goodies as well.

Hope those thoughts are useful.

Yeah I agree, in general I don't think it prioritises production and unrest buildings enough.

Made some pretty significant progress today. I think the improve nets swill be noticeable enough as to be unmistakeable. More to come.

I'm going to check in what I've done on Sunday and hopefully it'll go into the next update.  

There's a lot more to do of course but I found some issues with pathfinding and city construction that make a pretty big difference. Also worked a bit on the tactical battles.

I'll probably need to spend a whole update cycle on tactical AI though since it's going to require a lot of YouTube watching.

I'm at the point I just wish Caravans were completely removed from the game.  Now that the AI is actually casting freeze, they're targeting caravans as they're perceived as enemy units.  Which is also irritating having to cycle through a dozen enemy caravans to actually find the real enemy units broaching my borders.

I like caravans despite annoyances. Hate to lose them altogether.

I like the diplomatic pressure having caravans create, but in practice seeing the caravans and having them work properly isn't fun/functional. They could be removed and have the bonuses still apply or continue to be improved. Right now, it's kind of in an 80% working state that causes issues with the AI/pathfinding/bugs.