On harder levels (like challenging) I may do this kind of thing.  But 20 years of AI writing has taught me that fun > cutthroat.  

My first AI for OS/2 had the AI doing all kinds of nasty stuff to players that the humans could do too but they were always too lazy to do.  People hated it.

At the end of the day, we're making a commercial game. My job is to make the AI beat you in a fair fight.  So presently, I have a bEarlyGame variable where the AI cuts the player slack. At higher levels, I can take this away but it's low on the priority list.

I think that's a fair middle ground in the scheme of things.
For example, if the A.I. took advantage of my lack of early game defences when I was out exploring for resources and such, as a human player would do, I'd learn to defend first, acquire later - no exceptions.  That would slow down the start of every game quite a bit, and result in a 4X title where if you didn't basically camp for the first two dozen turns and stock defence, you'd lose.  That's really not very fun, or in keeping with the whole 4X mentality.

Once the early game stuff is over, however, the gloves should come off.  If we're 350 turns in, and I didn't defend my little outpost, too bad; I've have 350 turns to get it done.

In keeping with achieving a fair middle ground, perhaps that bEarlyGame bool could be set false at earlier times for higher difficulty levels, and simply never set to true at the highest?  So, at the easiest level, the A.I. gives you - say - 100 turns of slack, where as at the next difficulty level they only cut you 75 turns of slack.  That way, part of the difficulty increase is that the A.I. is more of a bastard earlier on, and not so much just them being granted additional bonuses?
Just a thought. 

I will be looking forward to seeing the XML code for this. When I played WOM I would always go for a co-ordinated production/research strategy to secure victory - be it archer focused, melee focused, champions with magic weapons & armor, few cities or many cities strategy. A fast expansion or a slow expansion strategy. A burn cities to the ground strategy (i.e. Sherman's March) or a traditional city conquest strategy or a demon/dragon recruitment strategy. 

My fear is that without some scripting all the AI players will end up following the same strategy.  I want to see the AIs challenge me with a wide variety of strategies.  

you know Brad its stuff like this that really bugs me.. I play games to challenge myself, to role-play and witness a story unfold. If there is no challenge for me, there is no will for me to play a game. I always start games on normal difficulty because I am new to the game, but I expect the game to challenge me in my inexperience until a point that I am experienced enough to defeat normal difficulty.

FE is meant to be strategic wargame no? I want the AI to 'try' and defeat me.. If I wanted to play a city building game, I'd play that instead. I like the AI to mimic humans to the point that realistic and historical human decisions and strategies are employed such as early rushing, even if the chance(s) is semi-random.

One of the first things I tried to do with Civ 4/5 was to mod the AI and its aggression. There should always be some AI personalities that are totally warmongering and aggressive and some that are not. There should be times that starting a game that monsters take you out if thats symptomatic of the world etc.. Thats the fun for me to recreate semi-realistic outcomes that can work well as story based AAR. I for one don't care as much for achieving a win by so many points or in so many turns as I have fun along the journey, whether I win or lose.

I would rather the AI be very good (and capitalise on everything possible) and then crippled in other areas (less hp/gold etc.) to bring it back to normal difficulty than vice versa.

oh and specifically as far as early game strategy.. bring it on!! If as a player I cannot adapt (through multiple games) to the AI taking early advantage and rushing me, then I don't deserve to be playing the game imo or at least on normal difficulty. Once again, this shouldn't be a factor for every AI and then when it is... DIPLOMACY! Two things.

1. Necessity is the mother of invention.

2. Diplomacy is the continuation of war by other means.

You have it backwards. Scripting results in the same strategy.  The AI will play each game differently because the data of each map is different.

Not to be obnoxious but Brad is talking about real feedback from real users in a real game and you're talking about your response to a hypothetical situation in an unreleased game.  I think Brad's real-world experience kinda trumps your particular understanding of the issue.

And I do think you're misunderstanding what Brad is talking about.  He's talking about making his game enjoyable.  As he notes, the AI is much better at calculating odds than humans.  An AI that plays ruthlessly is not going to be fun to play against, in part, because it wont take chances or make mistakes.  Everything it does will be to its advantage.  It will just do what works to win.

But it also wont be fun because the AI is the game.  If there is an exploit or an unbalanced element and you don't know about it, you lose.  If there is a strategy that you don't know about, you lose.  If there's a quirk in the map that you don't know about, you lose

That's pretty much what you get from playing against an anonymous human over the internet.  And there are many people for whom that is the pinnacle of gaming.  But there's a significantly large number of gamers who don't particularly like to play against anonymous humans over the internet for this very reason.  Because they are interested in exploring the game.

In which case the role of the AI is to add challenge to the game in a way that's forgiving of human inefficiencies.

Another point is that Brad's letting us watch the sausage as its made.  It's unclear how this particularly element of the AI's personality will be implemented in the final product.  As humans, lacking other information, we tend to project personalities onto behavior.  If Brad never exposed you to the underlying mechanics of its decision making, you might never have noticed.  And even if you did, you would chalk it up to a more peaceful or cautious AI personality this time through.

Yuck. That's completely the opposite of what I would like. I don't play games like this to have them over as quickly as possible, I play them to enjoy them. I enjoy exploring the map and expanding my empire. If I go to war with an AI player, I do it for reasons that make sense immersion wise; because they are controlling something I need, because they're doing something I don't like, because they attacked my friends, or because they are close to reaching a victory condition. I hardly ever go for a conquest victory, but even if I do I don't go rushing for them from turn 1. I want to be immersed in the world, and reaching a victory condition should be something that flows out of that eventually. The AI should play the game the same way. Or perhaps I should say they should roleplay the game..

I'm not really asking for a cutthroat AI.  It's just that if the AI is forced to build early defenses (to fend off the human player) while the human player isn't, then the human player has a huge advantage right out of the gate.

This is why I liked Civ 5's cities that defended themselves.  The early defense came for free to both the AI and the human.  The AI can't spoil the human's fun by nailing undefended cities, and it's fair to both AI and human.

Perhaps monsters will solve the problem?  If they attack undefended cities, then that's enough incentive to defend them.

I think that cut throat AI should be implemented for anything past normal difficulty.

I am primarily a builder in these types of games and if I had to deal with Zerg rushes game after game I'd quit playing it.

That way, I can enjoy my style on normal and people that like cut throat AI can enjoy it on the more challenging levels.

Sounds fair to me.

I thought that cities will have standard guards that prevent rushing scenarios? 

My suggestion in that regard was to give the AIs "luck" against monsters.  As in, AI/monster battles are fought twice in the background, and the AI gets the best result.  (As a toggle of course)

My justification is that if humans are unlucky vs monsters, they tend to save/reload.  This would be a compensation.

If monsters just ignore an AI and attack a human, it's an obvious cheat.  If an AI wins a 40% battle vs monsters 64% of the time, that's not obvious.

The question that I'd be asking is why is the AI being squashed by monsters?

I'd also like to see AI's store player tendencies into account, and use that to a degree against humans. 

For some sorts of folks, fun= cutthroat AI.  I think a cut-throat AI option is a necessity in any game that lacks a solid MP component.  It's not for everyone, but the lack of such an option can be felt in some games.  The downside to a cut-throat AI is that it can sometimes make a game play only one way if the game has a design flaw.

Also, an AI does have to get more cutthroat as a game evolves, as players will learn to exploit the AI.

Ah, that makes a lot more sense. I would imagine there's a lot of "move crunching" going on in tac combat.

Generally speaking, AI decision making for a given scenario relies on two principles, the raw number crunching and a thing called heuristics. Heuristics means limiting the amount of decision trees you have to go through by eliminating obviously false ones. This is done based on experience, "common sense" or other forms of analysis, e.g. fuzzy logic.

Chess computers do this too. In order to reduce the number crunching necessary, they eliminate obviously poor moves by identifying patterns on the board. That cuts down on the decision trees and allows the computer to calculate further - more moves - due to the freed capacity.

So, I'm sorry for the lecture, I don't know the answer to your question after all I would like to hear an answer from Brad too about how the AI desides what a given unit will do in tactical combat, what kinds of heuristics does it go through and does it build a decision tree based on them, calculating future moves or at least the immediate results of possible moves (how much damage would each do, what kind of positional end result would happen) and picking the best option based on that. Or does it go for heuristics only and doesn't bother with calculating the effects of the move.

Dunno how far Brad is going to answer that, but thanks for a good question, TheProgress 

For those thinking of the challenging at all costs AI vs a fun AI, here are links to a very good presentation on the topic by Soren Johnson. In case you haven't seen it yet. The presentation is an hour long, but the second link has just the slides, if you want to browse through it faster:

http://www.youtube.com/watch?v=IJcuQQ1eWWI

http://www.designer-notes.com/PlayingToLose.zip

So it looks like Brad is trying to create a combination of a good and fun AI (in terms of that talk).  However, he wants it to be good enough so that it only needs minor tweaks, and only on higher difficulties, such as compensating for the human reloading at start (better starting positions for AI), or reloading during the game (if he implements the "play the battle twice and choose the best result" suggestion).  Ways that are more "provide compensation or a small to medium advantage that is not really noticeable, such as compensating for the AI not being able to reload, but the game is fun AND the AI is good enough that it can challenge the player, but just a little bit better at higher difficulties".

What do others think?

I really have no idea what you are trying to say to say in that first sentence. I've been gaming since the 80's and have played WOM since the beta... My comments are about general AI in reference to FE. Theres nothing to really argue or comment on other than fun vs good AI.

In response to your further comments.. If you had of read my comments properly I argued for 'some' AI to be warlike (a random variable) so as to more closely mimic human play style. Most people i believe would state that playing strategy games are more fun when playing against human players rather than AI because they are more ruthless and efficient. That is 'their' fun. A realistic AI will also mimic diplomacy which mitigates the need for war or early rushes and something which also needs to be factored into the equation. The AI 'is' the game because the object of the game is to win by eliminating your opponents etc. If you want to explore a game, play a game where there is no war or there is a very good storyline and or rpg element OR if you are challenged early game against a warlike AI.. do something not too historically dissimilar.. use diplomacy and pay tribute!

And please dont tell me about human inefficiencies in a game like FE. Brad said it himself above when referencing how complex the decision tree is in FE.. If you were talking about a game like chess fair enough, but FE? A computer AI is always going to be a crippled against an experienced human opponent. Knowing your opponent is half the fight. Something the AI will never know unless algorithms are created to analyse human playing style for patterns or there is a preconfigured optimal build style written in for the AI to counter.

I dont play games to be over as quickly as possible either.. I play for immersion also. I never play a game to rush the enemy and go for a conquest victory either..

However unless a 'real' strategy game challenges you, or role-plays, what is the real point (not talking about casual gaming)? To me every new game is a new story with a different outcome which is crafted by the AI and the decisions you make. Sure if the game is rigged so you die an early death every time, that is no fun. But i play and when every now something happens and I lose because of occurrences out of my control, and or decisions I make or not react appropriately to, then I role play it as such. 

It's safe to assume that if Brad is successful, and we're all hoping he will be, the AI will kick your butt.  It just will try to do it in a way that makes the game more interesting instead of devolving it into a spreadsheet game.  You're acting like the AI is nerfed, but these posts by Brad are about making the AI more competitive.

Let me clarify my point that you didn't understand from before.  If Brad, the experienced game AI developer says "making the AI act this way is frustrating for players and makes the game less fun, I know because I made an AI like that in an earlier game and the response was incredibly negative" then I believe him over your "I've been playing video games since the 80s and the idea of this annoys me" rant.  Does that make sense to you?

Ironically, your reply was preceded by several posts disagreeing with this premise.  Your gut-feeling about people preferring to play against other humans rather than the AI is the opposite of mine.  But whether most people playing strategy games think like you or like me is irrelevant.  There are plenty of people on either side.  But there are a crapload of games with sub-par AI that are built around multi-player.  And there are a number of games with ruthless AI.  FE is not intended to be like Starcraft or LoL and it's not anything like AI Wars. 

But the reason Brad is committing so much time and energy to AI development (other than for his own enjoyment) is that he believes that his customers will value the game based on the experience of playing against the AI.  If it's scripted, then it will quickly become predictable.  If it's nerfed (as you fear), then the game will be considered boring.  His judgment is that if it's dynamic and seemingly intelligent and doesn't play like an anonymous jerk on the internet, then that will magnify the perceived quality of the game.

Interesting take. I look forward to seeing how the AI works in the beta.

My fear was that without some scripting that you would not see; for example, one of the AI's use the the Sherman's March to the Sea Strategy - where the AI builds a large army, captures a city (or outpost), razes it to the ground and moves on to the next city. This streamroller approach was very effective in WOM and the AI never used it (seems that all the AI's were too nice, never wanting to raze a captured city).

How about (I keep suggesting this, but AI's should be as customized as possible, as every player has different optimum preferences) a ruthless AI that roleplays.  A truly Ruthless AI would backstab everyone like a human, but I generally don't do that as a player.

Ultimately, what I want from an AI is for it to play by the same rules they do.  Not everybody plays by my rules.  In GC2, there were a few things I didn't do, because I felt it exploited the AI, or that the AI would not do those things, so it was unfair for me to do them.

This is an excellent video. I've watched it a few times before and I couldn't agree more. Civ IV stuck the perfect balance IMO. I much prefer a fun AI that helps and facilitates you telling a story.

However, it would be cool if a few of the AIs were naturally cut throat or plain bat sh*t crazy (even on normal) like Monty was in Civ IV. It would help differentiate the factions, give them personality  and really help with the immersion IMO. Also having an option to make the AI's personalities random would be good as well too.

Really I don't understand where your getting off.. Once again if you had read my comments you would understand I took exception to the comment whereby Brad wanted to nerf some of the cut throat and rush styled AI elements, not Brad's attempts at a good overall AI. I believe there is a place for 'some' of this style of AI play. 

And I still dont understand why your making it a me vs Brad experience in AI argument when I've already explained my rationale. Brad's comments are more than likely correct. Many people do not like an overtly aggressive AI and losing which equals no fun. But good AI should be optimised so that it is the priority and balance becomes a secondary priority. If your designing a crippled AI for normal balance, then your effectively working on more than one AI at more difficult levels.

ahh... where are the 'several' posts disagreeing on my premise??!.. There are millions of people playing multiplayer games because of this very fact. People love multiplayer or hotseat capabilities. Look at Civ 5. A pretty big single player game that had many people hammering for better multiplayer and hotseat capabilities because they believed the AI wasn't sufficient enough in single player. I would say that a fair chunk of the criticisms of Civ 5 have been around the AI and a crippled AI will always disappoint in the long run. If you want to stop the AI from crushing you.. either use force or diplomacy in return.. Diplomacy shouldn't be simply to relay that your being crushed and now at war, it should be to engage with the opposition to create a state where there is no war. This means a route out of rushing etc.. ( although no AI should know where you are located and purposely make a beeline towards you so it can knock you out without factoring in other factors, distance and weights etc.)

yes thanks to Vallu751 for the links also. 

I must disagree with Civ 4 though. I was really disappointed with the AI and the immersion in vanilla Civ 4..  i had minimal fun and started modding it straight away, although it got better with BtS etc..

I see that.  I would try to clarify, but I have no confidence that anything I say will be intelligible to you.  I'm really not trying to be combative or argumentative.  It just seemed to me like you were squabbling over something of minor consequence that'll probably be easy to mod out and is from my perspective most likely a correct design decision.

Again, I just don't see any evidence that Brad is designing a crippled AI for normal balance.  What I do see is a commitment to a robust AI that's capable of playing competitively without cheesing/spamming/rushing or any of the other effective but lame strategies so commonly employed in multi-player.

But the AI is supposed to cheese spam rush. I just want it to be unpredictable. I want Bastard AI AND angelic AI. Friendly and Hostile, Stupid and Smart, Might and Magic. What I don't want is an AI that never rushes me, I just don't want it to ALWAYS rush me. I think I'm with Vaul on this one.