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

[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.