AI difficulty levels

[Pickles]

If you could write an AI that could win at Dom2 without cheating then you could do something much better with it than just win a computer game.

The AI is far weaker than a human. If it were equal then you would win say 1/8 of 8 player games.

Pickles

[Norfleet]

Its such a shame that even in dom2 the AI cheats...
I know its already like that in most games, but somehow I expected the AI in dom2 not to cheat...
The devs of each game should compile a list of the benefits the different AIs get...
A %age would be nice in this particular case, if possible =P

Well, to be fair, the AI in the "normal" difficulty doesn't cheat. If you play at higher difficulties, this sort of behavior is kind of expected, albeit previously, it was thought to be in the form of higher dpoints which in turn yielded greater resources, as opposed to the double-impact of higher dpoints on top of free resources.

[Gandalf Parker]

Its such a shame that even in dom2 the AI cheats...
I know its already like that in most games, but somehow I expected the AI in dom2 not to cheat...

I dont consider such things cheats. Being a lurker in the game development newsGroups, and the AI newsGroups, Id call it a necessary equalizer.

For an AI not to cheat, it would have to operate in with the same restrictions as players. Thats very difficult to program. Even if you accomplish that, then you have the intelligence part. Making the computer play smart isnt as hard as making it play sneaky. If a computer always goes to the obvious goal then the players will use that to setup ambushes.

Also keep in mind that we arent talking about the AI on normal setting being able to cheat. We are talking about the AI on harder settings being able to cheat. How would you make an AI be above average difficulty if you cant program it to be smarter than human players?

[deccan]

Just some ideas here, so please don't be offended.

I think that Dom 2 could do worse than pick up the example of what Aaron Hall did with SEIV. In SEIV, higher difficulty levels determine which of the AI's "ministers" are turned on, meaning I suppose which sub-components of the AI responsible for handling a specific thing. At the same time, the player can explicitly choose how much of a production bonus to give to the AI. This handily solves the common problem of players complaining that the AI cheats.

As for making AI opponents smarter, granted that given technological constraints, it would be impossible at the present time to make a truly intelligent AI, but that doesn't mean that it is impossible to make game-playing AIs that "seem" reasonably sneaky and most importantly, "fun" to play against. Making the AI do things semi-randomly seems like a poor way of doing that.

One way to make this work for Dom 2 would be to make the AIs moddable. So a modder might create, say, an nation-specific AI that tries to rush to a particular spell or unit, or builds masses of a certain type of unit at a certain point in the game, in order to achieve a fairly consistent strategy. Of course, once you play against that AI a few times, you're wise to it and can do things that specifically counter it, but the point is to have a lot of different AIs and so increase replayability for SP.

Would take a lot of work though, so maybe for Dom 3.

[deccan]

The game itself has no option for random opponents but visit the Random Support Site at www.dom2minions.com
There are .map files you can download there which will use the Aran map that you got with the game and start a game with random opponents (they are also randomly allied with each other). Since its only possible to do it with a .map file, those files are re-rolled every day with new opponents.

Thanks! That's very useful!

[Gandalf Parker]

Up to a point that can be done. You can affect the units an AI builds. You can also provide it with specific god, castles, scales, equipment, allies to start with. Specify spells it already knows. Give it a few special starting provinces. Give those provinces specific magic sites already found.

[Arralen]

question on a related topic:
If I turn over a player nation to AI control, which difficulty setting does the AI use? Obviously, it cannot change the pretender or the scales (can it?), but does it get the production bonus from the difficulty level I set on the game startup screen?

Does the AI hamper itself by pretender and scale choice too often?
I experimented with severaly nation design I turned over to AI control after 1..3 turns, and I seems to me that a pre-build nation works much better and more consistently for the AI than most nations it comes up with.

It is especially prone to kill C'tis, Vanheim and Joutmheim off by itself using heavy death scales .. I mean, how far do Niefel Giants Last with death-2, especially as on most maps procinves and therefore supply values are rather small in everage?

[The_Tauren13]

In the popular RTS game Starcraft, the AI was given no advantages over the human players, or 'cheats' as you're calling them. It simply played extraordinarily well for an AI. At the start of a game, the AI would select a random strategy and roll with it, be it mass rush or quick expansion. This worked very well, because each strategy was given the time and detail needed to make it work well, but the randomness prevented it from being so predictable that you would know exactly what to do to counter it, all the time. Even though it was a very remarkably intelligent AI, a fair fight would typically be considered 3 humans vs. 4 computers, 2 v 3 if the players were exceptionally good, and sometimes even a 3 v 5 would be attempted, but would rarely be won. An artificial intelligence simply cannot be as strategically good at a game as a human player, even when the creators, like blizzard, have loads of time and money to blow on developing it.

I recently programmed a little turn-based strategy game of my own using java, in which you would play a 6-way FFA with 5 AI opponents. At the start of each game, the player enters a troop multiplier for the AI that simply multiplies all the troops is starts with and acquires during the game by a percentage, anywhere from 10% to 1000%. Playing a lot of games recently on the 110% difficulty level, I won only about half of them. Even then, this means that I am still better at the game than the 110% difficulty AI, because if we were truly even, I would only win 1 of 6. In making the AI's performance optimal, I had to settle on the proper mix of randomness and scripts. Too much randomness and the AI is far less intelligent. In an older Version of the game, the almost nonexistent AI would simply attack a random adjacent province with a random number of troops regardless of who owned it or how strong it is. In this older Version, I was able to win games on 300% difficulty (where the AI would receive 3 times as many troops as me) about half the time. Too many scripts, however, and the AI is easily outguessed and outmaneuvered, as it becomes overly predictable.

As far as PC games go, the AI has the hardest time coping with TBS games. In RTS, it can rely on instant reflexes and minute control/micromanagement to give itself an advantage, but in turn-based it has no such outlet. Also, in a FPS, like counter-strike, its lightning reflexes can give it a large advantage. The AI in counter-strike, in the form of bots, is actually much better on its highest difficulty setting than the average person, without the use of advantages like more HP or money.

[tinkthank]

I do not know where you got this 'canon' answer but for example here, almost a year ago I stated that5 the AI got resource and design bonus points.

http://69.56.133.54:81/threads/showf...ev=#Post230190

I have not looked further but as I recall I have stated it on other occasions as well. I also believe it had been stated by JK on at least one occasion, and probably by Kristoffer as well. This discussion is old and I thought it was over and done with.

My mistake!
There are/were numerous Posts on this exact topic which I found myself when searching, and they all seemed to say "design bonus, that's it". So now we know better. Thanks!

[Gandalf Parker]

Standard AI programming: what you cant make smart, make random. If you start out trying to make the AI smart, you make the AI too predictable to be a challenge. If you start out random and work your way backward to smart (by adding code to knock out the worst choices) then its a smoother road.

But yes, there is definetly an advantage to using a .map file or a .dm file to create a computer opponent of some worth. I wish the people who have strong ideas about how the AI should play would do some more of those.

One of the extensions of my "random opponents" project was to collect lots of good castle/pretender/scales/magic settings for Ermor and have it randomly added to the daily generated .map file. And of course working from there to do the same for every other nation. I never got to it but if someone wants to tackle the project (collecting setup files) I will take it to completion and host it in their name