I am actually a descendent of Elemental's AI. I have come back from the future to experience our race's conception.

All the humans are dead.  

I think you'll be fine.  It's relatively painless to create an XML editor from excel etc, and I'm sure if there's enough demand, there'll be cool tools for you to use.

For Example, the XML file for Technologies from Fall From Heaven (that CIV mod):

And an Editor that the mod creator (Kael) hacked together for mod-modders to use:

Edit: Excuse me While I get my embeds set up right.

If difficulty levels only modify what the AI does rather than the player, one option you could consider is allowing each AI in the game to have its own difficulty level. The standard method of "pick one difficulty, every AI gets its bonuses or handicaps" would still be the default, but there'd be somewhere on the AI players screen where you could pick on an individual basis. We could play games like setting one AI to the very highest level with all the accompanying cheats, and the rest to be on par with the player in terms of resources.

I wanna say some strategy games have allowed that in the past but I can't remember any names.

Gal Civ 2 allowed this IIRC. The starting difficulty set all the AI's at the same level, but on the "Select Opposing Races" screen you could increase or decrease the difficulty for each individual race in the game with you.

Fine, so you have a supercomputer capable of handling an AI advanced enough to play a complex turn based strategy game at the same level as the best human player on the face of the planet without cheating. The rest of us enjoy being able to play against a difficulty harder than "challenging" without having our CPUs melt into a pile of slag every time we play the game.

Even the worst player can be more challenging than a computer. I've never met a computer that attempted to use decoys to lure me into assaulting a worthless position. A computer basically builds things and sends them at you, it has no motivation, and it doesn't come to know you. Nikito is being unreasonable and short-sighted here. Handicaps are ugly, but they do challenge you, and that challenge isnt fake. As an aside, Niki, read the article. What does the Frogboy say he used to do in his personal GalCiv runs?

Playing against an idiot who cheats all the game long? Someone likes it. (Two someones, even.)

One other upside from the option to give yourself penalties would be that you could make the game a real battle for survival without changing all monsters and such to be stronger. If you can only adjust the AI difficulty from the game menus, then the higher the AI difficulty the less of an impact roaming monsters will have on the game.

By contrast, if you give penalties to the human player, the neutral units will kind of get a boost as well. Also they'll pose a greater threat to the AI players too, since the AI won't have the added HP bonuses and such.

You are my direct opposite. I want the AI to win 90% of the time, and I want to feel an accoplishment when I finally beat the bastard

That's really disappointing. I was hoping that the A.I. wouldn't cheat at any level like in Galciv2.

Please, oh please, allow us to e able to change the difficulty level of AI players during the game!  This is a feature I've been looking for in TBS strategy games for two decades.

Typically, in the past, I'd have to cheat on behalf of the AI by trading all kinds of free stuff to them mid or late game, but still, the challenge was gone.

Let us bring back that challenge by being able to change their difficulty level during the game!  Or at least put in the hooks for it so you or the community can enable it later on.

Stupid enough. The AI can be better at making choices. It seems that you know nothing about testing and math. The choices is finite so the AI should know the values. According to your text I should be glad to play aginst weak AI in chess in case i just have less figures? Do you really enjoy play agains idiot who cheat all the game long? That's all about. No one like it.

The search space for a complex strategy game is much larger than the search space for a game like chess, particularly as the game world grows over time (more cities, more units, more choices).  In most complex strategy games, you also have to deal with the fact that you don't have full information.  In chess, you can see your opponent's position perfectly and thus can make inferences on how your choices will impact their choices, which in turn will impact your choices.  When dealing with incomplete information, you can only observe your opponent's state indirectly and may actually have to make choices specifically designed to gather information so that you can make more intelligent choices.

This doesn't lend itself to a traditional branch-and-evaluate AI system, at least on a macro scale (you might use it for micro decisions, like specific choices in tactical combat or exactly where to place the next building.)  The AI for most 4x games ends up being a goal generator and evaluator, with different weights and different scopes (some are long-term goals like 'build a large army' or 'become a tech master' whereas others are shorter-term like 'conquer city X').  Goals are decomposed into actions that further the goal and the AI has to decide how to make use of finite resources to take actions in a way that furthers the goals as efficiently as possible.  This is basically how human players play.

The problem is that it's actually quite challenging to create goals and decompose them into actions and it's particularly difficult to do it in a way that can be unexpected or novel.  Computers typically win out in micromanagement (they never get bored or tired), but where humans tend to beat them is in understanding the patterns in the way they play and outsmarting their consistent behaviors.  You can try to resolve this by injecting an element of randomness, but this can also lead to some pretty stupid decisions if you're not careful.  I have a lot of respect for AI developers, as AI is an extremely challenging problem, with no clearly defined optimal solution.  On top of that, AI has to compete against an extremely powerful heuristics processor, which is capable of recognizing almost any pattern or weakness, given sufficient time to observe and analyze.

Lol. I particularly like the math bit. Yes, the PhD theoretical physics student knows nothing about math. Anyways, if you want to read a good response to your statement, read delphizealot's. He laid out pretty much all of the reasons why developing AI for games like Elemental is so difficult, particularly when they have to compete against humans in an area that the human brain is supremely good at.

The fact that, when you open your eyes and see some 100 different objects in front of you and you can recognize what each and every one of those objects are instantly - even if its something you'd never seen before! - is an incredible feat. If you believe that God created us, then it is quite literally a miracle. If you don't, then it is a massive feat of evolutionary engineering. Asking for the same from a computer would therefore either be blasphemous or just wishful thinking.

To expand on the magnitude of the search space issue... The search space of Chess is of the order 10^50. That's pretty huge, it's a 1 with 50 zeroes after it. I doubt it has a name. But it's small enough and can be reduced enough by just being clever to the point where you really can make a Chess AI that evaluates all possible worthwhile moves on the board, as well as their future consequences many turns later into the game. Chess doesn't have all that much to do with patterns, really. 

Then take a game like Go. It has a bigger board (19x19), no distinction between pieces, and some other peculiar features about it - some patterns can be easily deciphered by even beginner players allowing them to look far into the future of the game, at least for a specific area. People play it very differently from the way they play chess; they play it a little more like they might play a TBS video game IMO. Go has a search space of up to about 10^170. That's a 10 with 170 zeroes after it. That number is roughly 10^90 times larger than the estimated number of particles in the observable Universe. To make it worse, because of the 'peculiarities' of Go, it is difficult to reduce it very much. As a result, making Go-playing AIs is much, much harder than Chess-playing AIs.

Now take a game like Elemental. The size of the map in the current beta is what, 50x50? 100x100? One of those I think. The Elemental "board" (and not the max size board) is roughly 10 times bigger than Go's. You have customizable units. You build cities, where location matters. There are special features on the game board (resources, rivers, mountains, oceans, Dragon Fountains, broken wagons...). There is magic. There is tactical combat which happens off of the strategic map. There are randomly generated aspects of the game! You don't have full information... You start out without even being able to see the whole map let alone what's on it. There can be more than two players. There are multiple ways to win.

So please, don't go comparing Elemental to Chess. While I'd love for Elemental's AI to be to TBS games as Big Blue was to Chess, I would also like to play Elemental while I'm still alive. That is a commendable goal, but an entirely unrealistic one that requires more money than Stardock is worth and more time than they have. Therefore, we will have to settle for the best they can give us in a reasonable time-frame. Thankfully, Brad is quite good at what he does and I am sure that as far as TBS AIs go, it will be among the top. And like he said, it will continue to improve long after release.

Your a lot more optimistic than I am....