Honestly, I don't think that EL AI could win a supremacy victory against a reasonably playing human on any difficulty.



Even discounting the poor state of its military, the AI is mostly unaware of victory conditions. That keeps it from targeting capitals the way a human player would. And the AI just doesn't see elimination as a worthwhile pursuit. It just doesn't attack to destroy.



This might be by design, and some players might appreciate the impoverished AI, but I find it suspicious that this design also means less development time/effort, and I think it's probably the latter that's more important. It's far easier to dumb down good AI than to spice up poor AI.



Now, if anybody wants to play a strategy game against good AI, I'd recommend Go. Free clients, free AI that is relatively good (but still far inferior to the best humans), deep and elegant game.

Darvon wrote:
"what skill was I playing? It was normal."



/thread

This thread was started BEFORE the last update which made significant improvements to the AI. So for that alone, I'd say it is no longer valid.

"what skill was I playing? It was normal."



/thread

Maverik wrote:
I do not think the solution is multiplayer for all, play multiplayer requires rules that not everyone is willing to take (fast speed, aggressive games, many time limitations, auto resolve battles, personally I do not enjoy the game with these rules).

Agreed; AI is very important in any kind of strategy/tactics game and telling people to go to play MP is not the answer.



Anyways, this thread worries me about future re-playability once I master the basics.

I think we sometimes tend to get bogged down with philosophical terminology and high end concepts when it comes to game AI and it doesn't really help.



No reasonable person is seeking actual artificial intelligence from game AI. No reasonable person should truly expect a single AI player to beat them on even terms if they are sufficiently familiar with the game.



The reasonable expectation is to have AI that doesn't continuously sabotage itself or miss very obvious opportunities (such that the player is glaringly reminded on a constant basis that it's playing against a dumb computer). One that is sufficiently able to navigate the game mechanics so that the player feels like they are playing against several opponents with varying personalities and/or goals. Collectively, your AI opponents should provide a reasonable challenge. And more importantly, an interesting and fun game experience.

Darkscis wrote:
Did you have a peace treaty with the Mezari for the bulk of the game? If so, the AI probably considered you blood brothers due to long lasting peace and trade and wouldn't declare on you anyway. I'm with EzekielMoerdyk on this one, I don't particularly think EL's AI is any worse than the bulk of other 4X games out there. Quite frankly, until mankind develops a true AI, the only way that gaming AI will compete with a competent player will be to give it significant advantages ("cheats" if you will).

True AI is conscious, but this here is just a compute and knowing the rules and finding mathematical optimal solution issue. 4x games can have AI that is better than humans. But right now compute limitations would limit the fun factor quite heavily as it might take 'some' time per turn. ;-)



https://xkcd.com/1002/

Darkscis wrote:
Did you have a peace treaty with the Mezari for the bulk of the game? If so, the AI probably considered you blood brothers due to long lasting peace and trade and wouldn't declare on you anyway. I'm with EzekielMoerdyk on this one, I don't particularly think EL's AI is any worse than the bulk of other 4X games out there. Quite frankly, until mankind develops a true AI, the only way that gaming AI will compete with a competent player will be to give it significant advantages ("cheats" if you will). Can it be optimized? Can it be made better? sure. Will it ever compete with a competent human, who has the ability to backstab, manipulate, exploit? No.



Take a look at one simple strategy, used in any 4X game, by pretty much every human player.

- Human player makes peace with his direct neighbours to give himself some breathing room while he prepares, in advance, an army of conquest.

- AI players starts accruing long lasting peace weight with human player.

- AI gets various diplomatic "points" that govern it's response. Lets say +4 long lasting peace, +3 trade, -2 "our military is stronger". Overall, AI is +5, likes the human and won't declare war.

- Human bribes its "friend" into war with someone else.

- Human then declares on its neighbour, supposedly a friend, and snag's a bunch of cities while the AI's armies are on the other side of the map.



How do you program an AI to combat this human? You make the AI willing to backstab as well. The problem with that, as Civ 4 discovered, is that there still needs to be a point at which the AI won't backstab - otherwise the entire diplomacy system is nullified. Once human players discover where that point is, they do everything they can to get that AI to that mark so they are safe, then carry on with war preparations and get ready to backstab themself. Watch any youtube let's play of a Deity level Civ 4 game and you will see the lengths at which human's will go to to win. If you program an AI to backstab, manipulate and exploit - well there are plenty of movies and games about that



Thing's that can, however, be done to make an AI more competent would be things like;

- Programming it to optimise and specialise cities (no more turn 180 - AI's don't even have mill foundries) *Even if the bonuses mean the AI don't need it - at least make it look like it isn't cheating

- Programming it to pre-build war armies and THEN declare war (Civ does this well)

- Programming a more "devious" diplomatic AI. It should take into account (perhaps even hidden modifiers) what actions the human has done in the past. eg. If a human declares on a friend, the other AI's should take note of this and use it in their calculations. The reason I say hidden modifiers is that the AI should be willing to make peace with that human, but have the hidden modifier telling it to be careful and/or plan it's own backstab.

I made no peace. I can't. I was playing Necrophages.



I know the AI will never be truly like a human. But it needs to at least pursue the victory conditions better than what I saw. I am not asking more than that at this time. I should not turtle, especially when it has an overwhelming advantage.

Did you have a peace treaty with the Mezari for the bulk of the game? If so, the AI probably considered you blood brothers due to long lasting peace and trade and wouldn't declare on you anyway. I'm with EzekielMoerdyk on this one, I don't particularly think EL's AI is any worse than the bulk of other 4X games out there. Quite frankly, until mankind develops a true AI, the only way that gaming AI will compete with a competent player will be to give it significant advantages ("cheats" if you will). Can it be optimized? Can it be made better? sure. Will it ever compete with a competent human, who has the ability to backstab, manipulate, exploit? No.



Take a look at one simple strategy, used in any 4X game, by pretty much every human player.

- Human player makes peace with his direct neighbours to give himself some breathing room while he prepares, in advance, an army of conquest.

- AI players starts accruing long lasting peace weight with human player.

- AI gets various diplomatic "points" that govern it's response. Lets say +4 long lasting peace, +3 trade, -2 "our military is stronger". Overall, AI is +5, likes the human and won't declare war.

- Human bribes its "friend" into war with someone else.

- Human then declares on its neighbour, supposedly a friend, and snag's a bunch of cities while the AI's armies are on the other side of the map.



How do you program an AI to combat this human? You make the AI willing to backstab as well. The problem with that, as Civ 4 discovered, is that there still needs to be a point at which the AI won't backstab - otherwise the entire diplomacy system is nullified. Once human players discover where that point is, they do everything they can to get that AI to that mark so they are safe, then carry on with war preparations and get ready to backstab themself. Watch any youtube let's play of a Deity level Civ 4 game and you will see the lengths at which human's will go to to win. If you program an AI to backstab, manipulate and exploit - well there are plenty of movies and games about that



Thing's that can, however, be done to make an AI more competent would be things like;

- Programming it to optimise and specialise cities (no more turn 180 - AI's don't even have mill foundries) *Even if the bonuses mean the AI don't need it - at least make it look like it isn't cheating

- Programming it to pre-build war armies and THEN declare war (Civ does this well)

- Programming a more "devious" diplomatic AI. It should take into account (perhaps even hidden modifiers) what actions the human has done in the past. eg. If a human declares on a friend, the other AI's should take note of this and use it in their calculations. The reason I say hidden modifiers is that the AI should be willing to make peace with that human, but have the hidden modifier telling it to be careful and/or plan it's own backstab.

wilbefast wrote:
I'm really sorry to hear that. We're aware that our AI is not perceived as up to the standard set by certain other games. Strategy game AI is a very hard problem and we are a very small and very young studio. Civilisation 5's AI is built on top of Civilisation 4's, which was built on top of Civilisation 3's and so on. With each release lessons were learned and systems and algorithms were salvaged and re-used. Endless Legend is our second 4X game and uses very little of the AI architecture from Endless Space: it's very different. We've not had as long to solve this problem, but we are still working on improving the AI. In fact a big update is planned!



I can only beg your indulgence and hope that this new iteration will be satisfactory.

This post was a bit "early". I have since understood a bit more on how races behave differently. The AI needs to be more aggressive at times no doubt but I am not done with the game because of it.



Update on 1st game talked about in this thread. I decided to reload a saved game around turn 300 and proactively declare war on the Mezari. He promptly became very aggressive and wiped me off the planet. The AI hang up here is perhaps he needs to declare war more often if he has overwhelming strength and his usual victory condition is not active.



Another observation from this 1st game, there is a 3rd race in it, the Ardent Mages. They have one city and have done nothing to expand. But also, the Mezari could have wiped them off the planet too and did not.

Just to remind everyone, including the OP:



Chess, a ridiculously simple, symmetrical, predictable and (in comparison to Go or 4X titles) relatively non-dynamic and linear game, took generations of world-wide research to get to grandmaster level. (https://chessprogramming.wikispaces.com/Artificial+Intelligence). Even then, so much of the AI for chess was brute-force methods - checking for all possibilities ahead, and then trying to weigh the best options.



Go, on the other hand, was for a long time the grail in AI research. Even simpler than chess in its rules and construction, it contains a much larger playing field, and therefore exponentially more options, that it effectively needs to be tackled with non-linear and dynamical sampling methods. Much more difficult AI to program.



Now consider EL, or any other 4X game. It is asymmetrical. It contains a huge playing field, in comparison. Each turn brings multiple choices of hundreds of different outcomes per turn, as opposed to a single choice in Chess and Go. It contains different victory conditions, different paths to each victory conditions, and, ideally, dynamic re-evaluation of grand strategy throughout the game. It contains diplomacy, as well as the will to war/peace - which is already difficult concepts to express computationally. And finally, AI for these games are developed by a smallish team at best, or a single guy (or CEO, in certain cases) at worst, and it is expected to be a) perfect, b) super fast, and c) tailored to each player's experience.



So honestly, be happy that we have a (non-scripted!) AI which can actually play the game, and with a little help, provide a challenge for our ridiculously powerful human computers. The AI in EL is definitely not good enough reason to stop playing. Hell, if we had an AI that did what the OP is expecting of it... I'd probably install an EMP panic button in every room.



And in comparison to other games - Meh. I've not really noticed that EL's AI is particularly better or worse than many other 4X games that I've played. Maybe there were some good AI's from the late 90's/early 2000's, but I might just have been dumber during those years as well.

wilbefast wrote:


I can only beg your indulgence and hope that this new iteration will be satisfactory.

Just keep doing your thing. The AI IS improving, and they will return, or not. This is more of an exercise of a concept at this point. Polishing off the AI before everyone switches over to ES2, unless you plan to have both games getting worked on after ES2 is released (this would rock by the way).

MrChoke wrote:
So I am quitting my 1st full Endless Legend game at turn 340. Just like Civilization 5 and every other turn-based strategy game I have ever played, the AI is incapable of executing a militaristic victory.

I'm really sorry to hear that. We're aware that our AI is not perceived as up to the standard set by certain other games. Strategy game AI is a very hard problem and we are a very small and very young studio. Civilisation 5's AI is built on top of Civilisation 4's, which was built on top of Civilisation 3's and so on. With each release lessons were learned and systems and algorithms were salvaged and re-used. Endless Legend is our second 4X game and uses very little of the AI architecture from Endless Space: it's very different. We've not had as long to solve this problem, but we are still working on improving the AI. In fact a big update is planned!



I can only beg your indulgence and hope that this new iteration will be satisfactory.

natev wrote:
That's an interesting blog you link, but I don't agree with that developer either. It sounds like he or she is writing, "We made an AI that was smart in one situation, but that could be drawn into a situation where it was dumb, and people called our AI dumb-- not fair!"



It's like buying a chain from the hardware store that breaks the first time you use it. When you try to return it, the store owner tells you, "Well, yeah, that one link was bad, but look at how strong this other link is!"

Not really, the main issue was that players considered the AI cheating, not the problem with retreating to the long corridors to snipe the AI from that position. The players had issues with the flanking already. And there are more examples of that kind of player behavior.



Besides that the corridor issue could be easily fixed with the AI, the AI simply could not run into those traps, but then the players get even more pissed because now the gameplay stops nearly completely in such a situation. The level designers could have removed as well such points in the map and replace them with areas with enough cover for the AI and the player and then forcing the player out of that cover with grenades, but then we are back to square one and just removed the flanking option from the AI which still would have been imo better than removing that script all together. So you have for sure a point as well, even when you ignored the main point of that developers blog.

I expect the devs are aware of the shortfall of good AI in TBS games.



They've at least allocated an entire subteam to developing the AI for ES2 so I'm very optimistic.

That's an interesting blog you link, but I don't agree with that developer either. It sounds like he or she is writing, "We made an AI that was smart in one situation, but that could be drawn into a situation where it was dumb, and people called our AI dumb-- not fair!"



It's like buying a chain from the hardware store that breaks the first time you use it. When you try to return it, the store owner tells you, "Well, yeah, that one link was bad, but look at how strong this other link is!"

BPrado wrote:
In comparison to...?

I don't know? Dogs maybe?

The CIV AI gets often into infinite loops, seems to be not aware of any of the combat rules, which means it gets often flanked and beaten by far outmatched units, etc … and this this consistent through all difficulty levels. And as far as I know this is as well intentional. Players seem to prefer to outsmart the AI than getting outsmarted.



http://askagamedev.tumblr.com/post/76972636953/game-development-myths-players-want-smart



And personally I don´t even agree with that developer, because in all honesty, I really miss my half-life 1 AI soldiers, which actively tried to flank me, made me constantly nervous about my flanks and clearly were not just spawning behind me. I honestly miss those guys. And I wonder if I am still on topic?

So I am quitting my 1st full Endless Legend game at turn 340. Just like Civilization 5 and every other turn-based strategy game I have ever played, the AI is incapable of executing a militaristic victory. How to prove this is simple. Just set the only active victory condition to supremacy. In this game I trailed in every category especially military strength to the leading AI. It took out another AI opponent early game, as did I. Then what did it do? It turtled of course. Sure it pillaged some of my resources until we traded closed borders but that was it. I went to turn 340, waiting for this joke of an AI to stop turtling and achieve the victory condition and it did nothing. They never do.



So your 1st question may be, what skill was I playing? It was normal. I don't like cheating AIs. But to be honest I doubt with only this victory condition active that it matters. Unless someone can tell me this AI is "smarter" on higher levels. I doubt it. They always get bonuses that you don't and with it comes the difficulty because game developers can always program an efficient AI. An AI that can manage cities second to none. But this is a strategy game right? Can they ever make an AI use strategy to actually attack and defeat the human player by more than ridiculous bonuses and flawless efficiencies? I have yet to see it.



By turn 340 I was starting to see that by me just hitting end turn over and over again and doing nothing but keep my cities building something (not even military, just buildings), I was slowly gaining ground on this turtling worthless AI. I could go 300 more turns and it would still be doing this.



I quit CIV 5 for this exact same reason. But I tell you what, this was especially bad and inept. Amplitude needed to do better than this. I wish I could get e refund. What a waste of such a beautiful, innovative game, with so many neat concepts. Only to throw it all away with garbage AI. Really too bad.

Apocalypse wrote:
The Civ5 AI itself is very bad to begin with. ;-)

In comparison to...?





By the way, you cannot take any other 4x as a base to measure Endless Legend´s MP. It´s much faster even on Normal speed, and Fast speed is balanced enough that it barely makes any sense to play Normal. No one really imposes rules to new players (I will leave if an experienced one manuals everything, but hey there´s more ppl there to watch you play), and the pace of aggressiveness is just what the game makes possible - you can´t expect to able to play Bob, The Builder with your city and have a hard time.



I´m just suggesting the next step in difficulty, OP says he doesn´t like "cheating" AI but wants challenge. That´s the only way he will ever find it. Complaining about AI has never made much difference, and has a lot less merit when the person doesn´t beat the hardest difficulties.

MrChoke wrote:
I do not agree that the experience for the human player is the same if they get more resources. If you have any other victory conditions active, science victory, score, economic etc..., a higher level AI will destroy you with neck breaking efficiency and bonuses. You truly do have to be a master to win. But tell me this, what will the AI with the "supremacy-only" challenge do at higher levels? I am a newbie to this game. He was kicking my a$$ with his economy and military but did that make him smart enough to attack me? No. Will having an army size 4 times bigger than that because he is a higher (cheating) skill level, you tell me. What he still turtle?



If you say normal is semi-easy mode, I am willing to go up one level and play again. But if this level is so high that it would be impossible for me to win with any other victory condition because of his advantages, I won't play it.

They will NOT turtle depending on the faction you are against. Many factions will come at you with a large army early in the game on hard especially the cultists, necrophages, and sometimes Ardent mages. From what it seems you played against the more "turtle" based factions which seems to have skewed your point of view of the AI to begin with. But yes, on Hard mode I find the AI to be more aggressive either way, whether it be because they are more confident because they have better bonuses might be true BUT they will be more aggressive and be less likely to turtle which seems to be your goal for the AI.

Ashbery76 wrote:
When you play the game a while you see the A.I has not been taught how the play the game.Most late games the A.I does not even have trading up and running.It has no clue about diplomacy and the player is the one manipulating the political system.Amp just seem to think adding massive bonus is the answer rather than coding the A.I how to play.



Civ5 has much better A.I.

The Civ5 AI itself is very bad to begin with. ;-)