You will never get challenging AI where it doesn't cheat unless we get truly sentient AI. Your problem is you don't understand how and why game AI works.

Great video!

I think a lot of people expect too much out of AI and don't realize how difficult it is to design an AI that is complex and knows how to respond to a variety of situations, and it can generally screw up easily. I mean, Nuclear Ghandi anyone?

the_cat_did_it wrote:
I think a lot of people expect too much out of AI and don't realize how difficult it is to design an AI that is complex and knows how to respond to a variety of situations, and it can generally screw up easily. I mean, Nuclear Ghandi anyone?

I'm pretty sure that specific example was kept as a joke rather than being an actual bug.

You can create a "Custom" Dificulty easily. Go to the Endless Legend folder on Steam, search the .xml file " SimulationDescriptors[GameDifficultyBonus] " and edit the parametrers as you like. At this moment, I am playing on a dificulty edited manually more harder than serious but easier than impossible, because the early spam of armys on impossible its hard for me.

the_cat_did_it wrote:
I think a lot of people expect too much out of AI and don't realize how difficult it is to design an AI that is complex and knows how to respond to a variety of situations, and it can generally screw up easily. I mean, Nuclear Ghandi anyone?

The question can then easily become: 'Why do developers keep designing games with mechanics so complex that they can never hope to program an AI to use them effectively?'



That's been the general trend in 4x games for a while now. You can't expect the average player to even understand what implementing an effective AI means. Most of us here, if we don't work in the programming field, only have that knowledge because we actively pursued it.



You have games with all manner of asymmetrical races and complex governance mechanics and then you add things like unit design in and say: "Here's an awesome game! Go play!"



The average guy comes along and, after a couple games, realizes that the AI is doing foolish things or being ridiculously inefficient to the point that it is easily exploitable, even with the difficulty up a couple of notches. He asks for better AI. Should he need a degree in AI programming, or slavishly read through thick books on the subject to decide that he wants a better play experience?



Maybe 4x games should all come with a 'gimped AI' disclaimer? I don't know what the perfect answer is, but I don't think it should involve starting by placing blame on the player who expects decent AI.

Slashman wrote:
The question can then easily become: 'Why do developers keep designing games with mechanics so complex that they can never hope to program an AI to use them effectively?'

I won't defend the AI and agree that EL is complex, but not every game can be Chess. I think that depth and narrative/sandbox experience is what most people want from 4Xs (i.e. "Fun", not "Good").

It's really annoying that we've gone from flat shaded polygons to beautiful 3D graphics, yet still seem to play with AI from the 80s.



Why can't anyone, not even the Civilization guys, get this right?



Do I have to wait another 20 years to play a challenging single player game where the AI doesn't cheat?

OP,



Having "hard" minor factions/barbarians increases the difficulty.

Slashman wrote:
If I'm looking for a good narrative, I can go play an RPG which will more than fulfill that need.

Yep.

Narrative = RPG

Sandbox = 4x on easy / city builders / empire builders (ie: ANNO)

Challenge = 4x?



4x to me is about the strategy. Applying my brains, tactics, skills versus the opponent (the AI). How the AI responds identifies how strong of a game it is. I kinda disagree with Slashman here though. I think the new features are what bring the players in, the developers just need to make a consistent effort to develop the AI early on instead of leaving it to be patched on at the end.

I was just trying to point out the fact that creating an AI isn't a simple process, and the Nuclear Ghandi issue (which was originally a bug that was kept on for funny factor) was a result of the fact that AI isn't actually intelligent, it's simply a program (or set of programs) made to respond in specific ways based on set logic/rules. The AI can come out broken or bugged when faced with real player, because the decisions and the way we respond as players is not constricted by the same underlying rules and can end up "confusing" the AI or eliciting inappropriate responses.



No you don't need to study programming and thick textbook to know what you want out of the AI, but it's not an easy fix either. Due to the complexity of AI, you might be able to fix one element only to find that when you start testing it's gone and completely screwed up some other part of the AI.



AI by default ends up being predictable because it is set to respond to the player and (and other AI players) in specific ways, it will never really be able to replicate the way a human would play because the system is based on logic.



Providing feedback to developers on what kind of elements you think are broken or weak is never a bad thing because it lets them know where there are problems to look at and resolve. The most important part, in my opinion, is to be sure that the interactions between all the elements are work together (instead of making Ghandi a nuclear warmonger). It can be really annoying when there are a lot of bugs and broken elements, but it's nice to know that in the case of EL the devs pay attention to the forums, so hopefully they are working on addressing some of the issues players are bringing up



PS. Please correct me if I have said something wrong, I am not a programmer or an AI specialist (obviously) so my understanding is pretty general and potentially wrong hahaha.

the_cat_did_it wrote:




PS. Please correct me if I have said something wrong, I am not a programmer or an AI specialist (obviously) so my understanding is pretty general and potentially wrong hahaha.

It's not that you said something wrong. I just took your post and used it to make a point that sometimes when we see an irritated player asking why the game AI is so dumb or something similar, we shouldn't jump on them and assume they should know the ins and outs of AI programming and its challenges...even at a basic level.



But regarding what you wrote earlier about the AI being predictable, I think that while this is always going to be true, it doesn't necessarily mean that the AI can't provide an interesting challenge or do things that make us have to up our game. Maybe it won't win, but I shouldn't have a boring time with it either. I think that's what most people want even if they don't say it or have troubling putting it into words.



What I personally realized while playing EL is that I don't have any interesting stories of things that the AI did that were cool, or made me think more, or sweat a little. I have those stories from other games like Sword of the Stars, AoW 3, FE:LE and even Star Ruler 2. Someone earlier mentioned narrative being a draw for the game. If all that is is the story segments from doing quests and the intro and outtro, then that's pretty weak to me. I would rather have my narrative spring from the things that happened by interacting with the random elements of the world and with the AI players. If that makes any sense to anyone but me, that is.

I always thought it would be pretty if games could map out/record the way a human player is playing and then work out those patterns in order to try to simulate that play style. I have no idea how hard that would be, but I think you would get some pretty interesting results if the AI started treating situations the same way you the player does; based on probability maybe? eg. 70% of the time when *this situation* occurs the player does x and not y, so now every time this occurs to me (the AI) there'll be a 70% I do x instead of y.



EDIT. Although that might lead to just the AI mirroring the player which is probably boring.

the_cat_did_it wrote:
I always thought it would be pretty if games could map out/record the way a human player is playing and then work out those patterns in order to try to simulate that play style. I have no idea how hard that would be, but I think you would get some pretty interesting results if the AI started treating situations the same way you the player does; based on probability maybe? eg. 70% of the time when *this situation* occurs the player does x and not y, so now every time this occurs to me (the AI) there'll be a 70% I do x instead of y.



EDIT. Although that might lead to just the AI mirroring the player which is probably boring.

The hard part of that is figuring out why the player is behaving in a certain way. If a player settles a particular tile, is it because of that tile? Because of nearby tiles? Because that tile is or isn't on a coast? Because there's a threat from an enemy unit? Because it's turn 20 and that tile is within reach of its settler? 4Xs are full of situations that will probably only ever happen once in the entire life of the release, across its entire audience. There are megabytes of data available to the AI at any given moment, only a few k of which are relevant to any particular decision.

I agree that it would be hard for the AI to pattern human behaviors. But some things may be easier then others, unit design, army/unit mixtures, specific research (if the human always researches something first, maybe the AI should as well), city location by faction by resource type (which resource type is most prevalent), etc...



Edit: The reason this might work, is because each player plays different. If im the military player who is always going to go for supremacy victories, and the AI actually knows that and researches weapons early, the challenge would be there. Where as an economic player, maybe the AI doesn't immediately go for weapons techs, which would help.



It is really hard for me to describe what I would like to see in the AI... so im going to segue into an example.



As far as interesting AI behaviors I always loved SMAC (regular game not expansion, the expansion ruined the game). I know the AI there is probably 100 times simpler then games now adays, but Ive played 3 matches on the hardest difficulty in the last week and each one was drastically different (instead of the same: wait until turn 40 and see hordes of enemy units from the same direction every turn).



For example: I was playing on islands. I conquered the two other people on my island about the time the other two major empires did the same, green and white were allied and the two remaining. Green was highest ranked and kept landing troops on my shores (in quantities of 8-16 each time) and actually took a couple cities. I repelled them and went on the offensive. As soon as I passed green for first place, the third ranked player nuked a tight cluster of my coastal cities which effectively halted my offensive against green. (In SMAC nukes completely destroy cities, since I had 3 tightly spaced, I lost all 3). He didn't nuke my economic cities, or the locations where my military was the strongest. White did the one thing, against the strongest player, that could help his ally (the move was literally perfect).



I can beat the hardest AI on SMAC 9/10 times, but it is -fun- to do so, because situations like the one above make each game different and challenging. I could play another 10 games and not see another nuke launched by the AI. The AI doesn't feel extremely weak, extremely 'cheaty', or extremely scripted.

natev wrote:
The hard part of that is figuring out why the player is behaving in a certain way. If a player settles a particular tile, is it because of that tile? Because of nearby tiles? Because that tile is or isn't on a coast? Because there's a threat from an enemy unit? Because it's turn 20 and that tile is within reach of its settler? 4Xs are full of situations that will probably only ever happen once in the entire life of the release, across its entire audience. There are megabytes of data available to the AI at any given moment, only a few k of which are relevant to any particular decision.

I guess that would be an issue with the AI mapping your play style, because it can't calculate why you do something, people can be pretty illogical and careless and therefore make imperfect decisions, whereas an AI picks based on whatever formulas it is governed by - and yea, that data would be a lot to process, but it would be a cool exercise in modelling playing patterns I reckon. Lets say you did model a bunch of play styles, you could then use general patterns (although you would limit variables, because particular situations are unique) to create a role-playing style that more closely replicates what a human player would do. I mean, there are some people who have the math and the game strategy all worked out, while other players don't plan ahead much and don't necessarily weigh all of their options for an 'optimal' result.



I guess what I was getting at with the whole idea is to insert some of that randomness associated with a human player, wherein sometimes we do make imperfect decisions. As you increase game difficulty the AI would decrease in the number of imperfect decisions it makes, therefore being smarter and a more difficult opponent.



But yea, just some thoughts, probably not something that would be done or implemented though.