So...the customers wont be cheated this time?

[RibbonBlue]

The AI isnt challegening
I feel so pitiful.... I have yet to be 16 impossible AIs at once.

[Saarud]

lol I must be kinda pathetic I guess. While I have won against AIs on small maps with few opponents I have not yet won a game against many (14+) opponents on a big map. The sad thing is that I find this AI to be much harder to beat than the GalCiv2 AI.

[Gandalf Parker]

When you do get there, let me know and I will show you how to ally the AI's with each other so they will gang up on you.

Check out the WE vs THEM scenario.
http://www.dom2minions.com/WEvsTHEM.shtml

Or the Random Opponents program which also allies the AIs
http://www.dom2minions.com/Opponents.shtml

The "Poke in the Eye" scenario is another way to boost the difficulty of solo playing
http://www.dom2minions.com/PokeEye.shtml

[ioticus]

I didn't think there was much wrong with Dominions 2. If Illwinter only fixed the remaining bugs and didn't release a Dom3, I'd be happy. As far as the AI is concerned, Dom was designed as a MP game first and foremost, so it's no wonder the AI did not really evolve. I'm mainly a SP gamer, so any improvement in the AI is appreciated, and I'm not going to look a gift horse in the mouth. It's unfair to compare Dom AI with GalCiv AI, since GalCiv was designed around the SP AI in the first place. The 2 games have different design goals. Also, Brad is one of the best AI programmers in the business. If I were rich, I'd personally pay Brad to develop an AI for Dom, but I'm only dreaming now.

[Saber Cherry]

ioticus said:
If I were rich, I'd personally pay Brad to develop an AI for Dom, but I'm only dreaming now.

The combat model in Doms is in a different league than GalCiv.

Brad has done a very good job with strategic AI in GC2 - last night, for example, all the AIs realized I had the smallest military and started taking me out. My tiny navy clustered around a fully-upgraded Military Starbase, and were mostly unassailable. So, eventually, the AI started ignoring my planets and fleets to go straight for my military starbase, and they almost killed it... and whenever my attention slipped for a second - like, when I intercepted their fleets so they couldn't attack my starbase - they'd sneak a fast troop transport in and steal a planet. All in all, I was very impressed.

However, you can't even evaluate strength and weakness in Doms2. An Ice Devil with a Hell Sword can kill 2000 militia, and die to 500 longdead. The militia can kill the longdead, but die to a mage casting Foul Vapors. The mage would die to the longdead, who would be killed by 20 priests, who would be killed by the militia, who would be killed by 500 Salamanders, who would be killed by 25 Devils, who would (maybe?) be killed by 25 Angels, and etc.

Writing the AI for Dominions is not easy, or even difficult; it's mind-blowing. Rule-based AI just can't work perfectly because there are too many rules.

I would compare Dominions to English (or some other highly hybridized human language) and Chess / Go / GalCiv to computer languages. Sure, GC2 is nondeterministic and more complex than the other two, but they all have fairly simple rulesets without many exceptions. As long as something is defined by rules, it can be taught to a computer - that's why compilers can be auto-generated. But anything that is largely defined usage, exceptions, or context becomes exponentially (with size) more difficult to teach to a computer. If Illwinter had stopped with just Pythium and Arco (without any mages), then it would be possible to write a very strong AI. But with spells, damage types, gems, items, immunities, unit attributes, and so forth, a complete and accurate ruleset probably would be larger than the universe.


I think that Illwinter has done a very good job on an AI that copes with this complexity, and can still beat intermediate players... and advanced players too, unless they follow very specific sets of steps, that would also allow them to beat virtually any human player who doesn't know the recipe for countering them. Perhaps the AI should be given some "recipes" that it seeks - like massing Druids with ivy crowns and summoning vine-creatures, taking a 9/9 Earth/Water pretender and rushing to Clay Men, starting a Tartarian Factory + Gift of Health, or rushing Staff of Storms + Wrathful Skies. It might be interesting. But no matter what recipes are given to an AI, nothing you see in this decade will ever detect the use of an advanced recipe, much less correctly choose and employ a counter-recipe. Consider how much work Brad put into making his AI detect that an opponent is primarily using Beam Weapons, and eventually compensate by developing Shields? That's in a game with 3 different damage types (and no immunities), or something like 9 basic units... and the AI is still slow to react, when it does at all, because it has to also consider every other player in the game, and other possible uses of research points. Now consider the scenario with 1100 units and 500 spells, mutated by various combinations of Dominions and blessings and items. Still seem easy to write an AI that counters some unseen strategy that a player cooks up?


In summary, I think Dominions II had a decent AI with a few important flaws, like castle-building, TAC-AI spell selection, ignorance of supply constraints, poor site-searching, sending pretenders to the deathmatch, choosing random scales and paths on pretenders, and the inability to use any powerful recipes (like putting good items on an Ice Devil, rather than random items on a scout). If those sorts of things are addressed by Dominions III, it will be much stronger in singleplayer... and it won't be obvious how to make it better. But even with flaws, it still beats most people over and over and over for months, or until they come to the forum for advice... or even then.

[WraithLord]

This is a good analysis, and from what I've learnt from my *humble* experience of programming AIs I completely agree.

I hope that at least some the obvious AI flaws that you mentioned will be fixed in DOM-III.

[DominionsFan]

I must agree with my brother, however I wasnt really playing singleplayer after a period, because of the unintelligent AI, I used to play multi most of the time with friends 2-3 weeks after Doms2 release. If I remember correctly, the game was sold as a sp/mp game with a nice sp ai. Even if the AI failed as 'challenging', its ok for a release, just consider that AI bugged or whatever. There kicks in the developer's support. If I understand correctly, the devs werent supporting the singleplayer part of the game (I stopped following the singleplayer - AI discussion after a period, since the first patches contained no AI fixes), and I think that it is sad, since it was (the singleplayer) a part of the game. This is like a half support and its never good.

Now as Saber Cherry said:

Saber Cherry said:

In summary, I think Dominions II had a decent AI with a few important flaws, like castle-building, TAC-AI spell selection, ignorance of supply constraints, poor site-searching, sending pretenders to the deathmatch, choosing random scales and paths on pretenders, and the inability to use any powerful recipes (like putting good items on an Ice Devil, rather than random items on a scout). If those sorts of things are addressed by Dominions III, it will be much stronger in singleplayer... and it won't be obvious how to make it better.

-> if these AI errors will be fixed for Dominions 3, the AI will be really challenging.

As for the Galciv2 AI, I dont agree with Marcus. The Galciv 2. AI is weak also, but just because there are some ways to beat the AI easily, and the AI cant do anything about it. Those are major AI flaws, like the Doms 2. major AI flaws.
Well if I am not using any "cheap" tactics in Galciv2, the AI is much harder to beat of course. The "general" AI is better in Galciv2 for sure, since it doesnt have major flaws like the Doms 2 AI. In Doms 2 the AI rarely built castles/recruited/summoned strong troops for example. In Galciv2, the AI builds kickass ships/starbases,colonizing hella fast etc. The only problem with the Galciv2 AI is that its "allowing" players to use tactics what the AI cant beat. So overall yes, the Galciv2 AI is surely better, but the Doms 2 AI problems can be fixed for Doms 3, and then, the Doms 3 AI will be >> Galciv2 AI imo.

[Gandalf Parker]

Saber, You are quite right about the headaches of programming AI in Dominions. That was shown in the forum when we asked for flowcharts and ai scripts which would fix things. Some tried and quickly shifted from whiners about the ai to just low grumbles.

Eventually (maybe Dom4) Id like to see dominions broken up. The maintaining would be a headache so it ight have to wait for multiple programmers. But Id like to see the host seperated from the tools and the client. DomHost.exe, DomTools.exe, and DomClient.exe, then probably a DomMain.exe to tie them together so that its transparent to the average user.

Then Id like to see changes made which would heavily support 3rd party scripting. Im not much interested in Dom3 for multiplayer but Id be VERY interested if it got to the point that really involved scripts that made the host take hours to do a turn could be generated. That would turn the AI into something exciting. Someone who plays Arcos really well could script an Arcos-bot, and someone who plays Ctis well could do the same for Ctis. Eventually we would even end up with variants such as an Ulm which plays as a massive army and one which plays guerrila warfare. Setting up grames which randomly selected various styles and the server spent most of the time between turns actually thinking.. *sigh*.

Anyway, If I won the lottery that would be on my list.

[Nerfix]

Saber Cherry said:Perhaps the AI should be given some "recipes" that it seeks - like massing Druids with ivy crowns and summoning vine-creatures, taking a 9/9 Earth/Water pretender and rushing to Clay Men, starting a Tartarian Factory + Gift of Health, or rushing Staff of Storms + Wrathful Skies. It might be interesting. But no matter what recipes are given to an AI, nothing you see in this decade will ever detect the use of an advanced recipe, much less correctly choose and employ a counter-recipe.

I think it could help the AI a bit, but I have no idea how it would be in practice, and how hard/unappealing it would be to make. Perhaps there could be a set of "guideline" pretender setups for AI so it won't pick completely strange scales or magic.

[Kristoffer O]

MStavros said:

!! You cant be serious Kristofer. I hope that you are just kidding, I mean what is the point in NOT fixing an AI?

Can't I
There are exactly seven points in NOT fixing the AI.
Three of them are PR related.
Three are secret.
Last tone has to do with strange cults and the orbital mind control lasers of the swedish 'Svärmöga', who sees it all.

Seriously though: there is no *point* in not fixing stuff. Fixing is good. That is why people do it from time to time.