I watched an interesting video recently called AI and Difficulty in 4X Games.  Some key points from the video:

*The bonuses that the AI receives to make up for their suboptimal play can be so large on the highest difficulty setting that it forces players to play a certain way to combat these bonuses.  This reduces the number of viable strategies, which makes the play experience less flexible/fun.

*It’s difficult to combat snowballing in 4X games and the endgame can become boring after you have snowballed, especially if you are going for a peaceful victory condition.

*How do you combat the player snowballing against the AI without frustrating the player?

*A couple interesting ways to combat the player snowballing against the AI and creating a boring endgame:

1. There are mods for Civ VI and Endless Legend that increase the bonuses the AI gets as the game progresses rather than give a constant flat rate from the beginning.  This gives the player more flexibility in how they play the game as the bonuses the AI receives aren’t so big from the start.

2. Adaptive Bonuses: Once the player starts to snowball, the bonuses the different AI receive depend on the type of game they’re playing (a military focused AI might receive a reduction in cost to making units, etc.) 

*Should the point of the AI be to beat the game themselves, or to provide the player with a good experience?

I assume the Fame mechanic was designed in part to help combat some of the issues described above, as it helps to simplify the problems.  Having a single victory condition  (apart from wiping everyone out I assume) should make it more clear for both the AI and the player on the kinds of things they should be focusing on to win the game.  

It seems like adaptive bonuses would be easier to program for a 4X game with one victory condition.  For instance, if the player starts to snowball against the AI in Humankind, perhaps the threshold for getting Era Stars could be lowered for the AI, allowing them to possibly catch up.  

As far as the question of: Should the point of the AI be to win the game or to provide a good experience for the player?  I feel like on Normal it should probably be to provide a good experience, making the game more accessible.  Whereas on higher difficulty settings, the AI should be able to crush players who play suboptimally.

We already had this kind of discussion in the internal forums. My take on it:

• High difficulties are for the most experienced players. The space of viable strategies always gets smaller, the more competitive you play, its the same when playing against human players. Not everyone needs to play on the highest level.
• I think adaptive bonuses are not a good solution to the "human player snowballing problem". This holds especially true, if its implemented as a rubber banding mechanism like you proposed. It feels bad if you get "punished" for doing well. I also think its a bit lazy, because:
• The reason why many 4x AIs struggle in the lategame, is because they dont recognize a snowballing player correctly and use diplomacy in a sensible manner to combat that. In any FFA-Game - doesnt matter if its 4X, RTS, FPS or 2d-Shooters like Invisigun - the meta typically revolves around teaming up to stop snowballing players. Even if the game doesnt allow for formal alliances, this kind of thing happens - often without players being able to communicate. If you teach the AI to do this in a somewhat sensible manner, lategame suddenly becomes way more interesting, even if the player is in a good spot to win. Its one of the main reasons, why the ELCP AI is better in the late game.
• AIs should ideally do both, provide flavor and a "good experience", but also have at least have some drive towards winning the game. How strong that drive is can ofcourse be dependant on the difficulty setting

LeaderEnemyBoss wrote:

• The reason why many 4x AIs struggle in the lategame, is because they dont recognize a snowballing player correctly and use diplomacy in a sensible manner to combat that. In any FFA-Game - doesnt matter if its 4X, RTS, FPS or 2d-Shooters like Invisigun - the meta typically revolves around teaming up to stop snowballing players. Even if the game doesnt allow for formal alliances, this kind of thing happens - often without players being able to communicate. If you teach the AI to do this in a somewhat sensible manner, lategame suddenly becomes way more interesting, even if the player is in a good spot to win. Its one of the main reasons, why the ELCP AI is better in the late game.

That definitley does sound like a better solution.  It woud just appear that type of solution is hard to pull off historically, at least in the 4X genre.  Here's hoping they can make it happen in Humankind :)

Hello everyone,

That's a very interesting subject you taking on here.
As LeaderEnemyBoss pointed out, relying on bonuses can feel really lazy and unrewarding for the player. As an AI dev I am also concerned about the fact that it can also "hide" some of the AI's bad behaviours behind its profusion of resources, and that's something we would like to avoid as much as possible.

Snowballing in a 4x is quite common, and experienced players will often reach that critical point before any AI can get "big enough" to oppose a decent challenge through the late game. Although we are doing our best to make each AI empire a worthy challenge for the player, we are also preparing for "that" moment where a player is snowballing, and the main tool at our disposal is, as LeaderEnemyBoss said, diplomacy.

We still haven't exactly figured out what we want to do on this subject, and are discussing the impact of such AI behaviours:

- If the behaviour gets too mechanical, you would see two sworn enemies allying each other to fight you because you are getting really strong.

- If the behaviour gets locked by "diplomacy state", you would see empires you're at peace with do absolutely nothing as you steamroll to victory

- Should AIs focus the player with most fame ? Even if he doesn't present any threats currently ?

- Should AIs be biased towards the player ?

- Should the difficulty setting only affect the AI's diplomacy with the player ? Or with every other empire ?

So many points that we are still discussing and testing internally, and we would love to have your thoughts and ideas on these :)

What I can say right now is that our goal is to give an experience as authentic as possible, and having AIs with "understandable" trails of thoughts.

We've already built the foundation of our diplomatic AI with such behaviours in mind, so we hope we can achieve something really sweet in terms of lategame diplomacy.

Cheers

Chocossimo wrote:

Hello everyone,

<snip>

So many points that we are still discussing and testing internally, and we would love to have your thoughts and ideas on these :)

What I can say right now is that our goal is to give an experience as authentic as possible, and having AIs with "understandable" trails of thoughts.

We've already built the foundation of our diplomatic AI with such behaviours in mind, so we hope we can achieve something really sweet in terms of lategame diplomacy.

Cheers

One thought is that maybe the shift in AI diplomatic behaviour should take place over a couple of steps, not just one.  The flip from "normal AI" to "AI is trying to stop me from winning" can be jarring and immersion breaking if it happens all at once. 

I could see four possible stages, though there may be room for improvement on this:

1. Normal AI behaviour
2. We are concerned: reduced willingness to co-operate/trade with the leader
3. You are a threat: increased willingness to take aggressive action against the leader
4. You are THE threat: the AI co-operate together against the leader

Ideally, individual AI civs will move into these stages at different times. Neighbouring civs may be more sensitive, and move up the scale faster.  Civs that are closer to you in Fame/military strength may be slower to react than civs who have fallen further behind.

As to whether this should be based on Fame or threat level, it's very difficult to opine without understanding better how the Fame system will work.  However, if the idea of Fame is such that it, in itself, doesn't snowball, then in theory a civ that generated a lot of Fame in past eras isn't a threat to do so again this era.  If that's true, then the AI might be better served focussing on earning it's own Fame, and the above diplomatic system could be focussed exclusively on traditional snowballing metrics of empire size and military strength.

Chocossimo wrote:

We still haven't exactly figured out what we want to do on this subject, and are discussing the impact of such AI behaviours:

- If the behaviour gets too mechanical, you would see two sworn enemies allying each other to fight you because you are getting really strong.

- If the behaviour gets locked by "diplomacy state", you would see empires you're at peace with do absolutely nothing as you steamroll to victory

- Should AIs focus the player with most fame ? Even if he doesn't present any threats currently ?

- Should AIs be biased towards the player ?

- Should the difficulty setting only affect the AI's diplomacy with the player ? Or with every other empire ?

In terms of the behaviour getting too mechanical, one way that might make for sense for the AI to suddenly declare war or leave an alliance could be espionage.  For instance if you have a high enough level spy within an empire maybe you could do a False Flag operation. 

In terms of AI focusing on players with most fame, I think one thing you'd want to avoid is if it ever made sense for the player to purposfully stay slightly behind in the Fame race to avoid heat from the AI.  I assume that wouldn't be a fun way to play Humankind anyway.

I definitely agree with LeaderEnemyBoss that the AIs shouldn't be biased towards the player, as to me it makes the game world feel more alive.  I always like when an empire I haven't even met yet suddenly gets taken out of the game as it makes me think woah, what happened there? I hope whoever took them out isn't next to me.

If you do want to have this kind of AI, please make this either an opion, or make it conditional on victory conditions being turned on, and allow us to turn victory conditions off. I want to feel immersed in a world, not in a game that I'm trying to win (because I'm not; I care about enjoyment, not about winning).

nlspeed wrote:

If you do want to have this kind of AI, please make this either an opion, or make it conditional on victory conditions being turned on, and allow us to turn victory conditions off. I want to feel immersed in a world, not in a game that I'm trying to win (because I'm not; I care about enjoyment, not about winning).

Even without victory conditions you need some motivations for the AIs to band together. For example: If you are a war monger and begin to conquer one AI after another, you would expect that the others are starting to band together. Them just watching passively as you do your thing would be the opposite of immersion in my book.

I agree with TravlingCanuck that you want an AI roleplaying as its current civ initially. But they should gradually warm up to the fact that there is a runaway that needs to be dealt with by cooperative action. 

TravlingCanuck wrote:

As to whether this should be based on Fame or threat level, it's very difficult to opine without understanding better how the Fame system will work.  However, if the idea of Fame is such that it, in itself, doesn't snowball, then in theory a civ that generated a lot of Fame in past eras isn't a threat to do so again this era.  If that's true, then the AI might be better served focussing on earning it's own Fame, and the above diplomatic system could be focussed exclusively on traditional snowballing metrics of empire size and military strength.

I second this. It depends on how many of the fame gaining strategies are based on aggressive behavior, whether that is military, ganging up on someone diplomatically, wonder races, and how many are based on just maximizing your own culture. Which of them are engine builders, and which are engine stealers. I think the exact strategy in each era should be based on what their current strength is, as well as what they were in past eras. An aggressive civ, whether militarily in other areas, will of course want to team up to attack the strongest player if they already have a fame lead, but even more passive ones may be able to enter alliances with the aggressive ones, basically sacrificing things like money or tech in exchange for an attack dog. They could chose to point their ally at either a snowballer or someone who is beating them at their own goals for the era.

One of the reasons I'm so excited about this game is the notion that the journey matters more than the destination. And in reliving human history, I'd like it to be as realistic as possible.

The fact that the AI changes its behavior depending on if you're winning is, to me, a potential threat on the immersion.

The way I would go about difficulty (making a lot of assumptions about other game design elements) is by introducing two dimensions to it and then allowing the player to choose from any combination of those two for further customization of gameplay experience.

The first dimension would be the "goalpost", which increases the total fame the AI has at the end of the game by a multiplier. That way the player has to up their game in higher difficulties as they have to generate more fame to beat the AI, but they would not be "at a disadvantage" regarding their ability to generate fame throughout the game.

The second dimension would be "hardships", which would be the traditional AI has bonuses / is more aggressive approach. It is this dimension that would determine if the AI reacts or not to your performance.

I think this way you would be able to satisfy players like me who believe the AI shouldn't break the fourth wall and other types of players who believe the AI shouldn't let the player win as easily.

afarteta93 wrote:

<snip>

The way I would go about difficulty (making a lot of assumptions about other game design elements) is by introducing two dimensions to it and then allowing the player to choose from any combination of those two for further customization of gameplay experience.

The first dimension would be the "goalpost", which increases the total fame the AI has at the end of the game by a multiplier. That way the player has to up their game in higher difficulties as they have to generate more fame to beat the AI, but they would not be "at a disadvantage" regarding their ability to generate fame throughout the game.

The second dimension would be "hardships", which would be the traditional AI has bonuses / is more aggressive approach. It is this dimension that would determine if the AI reacts or not to your performance.

I think this way you would be able to satisfy players like me who believe the AI shouldn't break the fourth wall and other types of players who believe the AI shouldn't let the player win as easily.

I view the purpose of the AI similarly.  re your first point, I call that "pace car": how fast you have to go in order to win.  Your second I call "speed bumps": how frequently you need to react to the game/AI, rather than just going on auto-pilot with your pre-determined strategy.  Both should increase (fairly steeply) as difficulty level is increased, so that players on low difficulty levels get a very different game experience from those who choose the higher settings.  I'd consider our discussion here to be part of the "speed bump" (or "hardships"), and presumably it could be limited to higher difficulty levels.

The other aspects of the AI (creating interesting opponents to play against and differing the game experience based on who you start beside) can be fairly consistent across all difficulty settings.

Chocossimo wrote:

Hello everyone,

That's a very interesting subject

we are also preparing for "that" moment where a player is snowballing, and the main tool at our disposal is, as LeaderEnemyBoss said, diplomacy.

- If the behaviour gets too mechanical, you would see two sworn enemies allying each other to fight you because you are getting really strong.

- If the behaviour gets locked by "diplomacy state", you would see empires you're at peace with do absolutely nothing as you steamroll to victory

- Should AIs focus the player with most fame ? Even if he doesn't present any threats currently ?

- Should AIs be biased towards the player ?

- Should the difficulty setting only affect the AI's diplomacy with the player ? Or with every other empire ?

So many points that we are still discussing and testing internally, and we would love to have your thoughts and ideas on these :)

- If the behaviour gets too mechanical, you would see two sworn enemies allying each other to fight you because you are getting really strong.

-> In a way and in some situations this would actually make sense. Banding together to fight a greater evil is something humans tend to do. So I wouldn't mind these types of events, but ofc there should be a decent weighting (really big threat) for this to happen.

- If the behaviour gets locked by "diplomacy state", you would see empires you're at peace with do absolutely nothing as you steamroll to victory

-> To keep things interesting, friendly empires wouldn't necessarily need to go all out on you. Like in real world, they would probably engage in covert operations, secret alliances and diplomacy to undermine your rule. Open warfare would be the last resort. Your "friends" would smile and compliment you while you're pulling hair with riots popping up left and right for no real reason.

- Should AIs focus the player with most fame ? Even if he doesn't present any threats currently ?

-> If Fame is the one and only key to victory, yes. But again, not necessarily all out on every front. If you're pumping out Wonders for high Fame, maybe they could start to race you on the architecture front to keep you on your toes. That, or if you're not posing any threat indeed, they could sieze the opportunity and stab you in the back. Nice lands you got there neighbour, just what I needed, you know.

- Should AIs be biased towards the player ?

-> Absolutely not. In an ideal world the AI would treat all forces equally and only react to growing threats step by step. Being biased against the player kills immersion really fast, makes for frustrating diplomacy, illogical AI moves (as in, unintuitive -> "why would he all of a sudden move all his troops to MY side in his empire even though I'm not advancing"), and makes the game feel gamey.

- Should the difficulty setting only affect the AI's diplomacy with the player ? Or with every other empire ?

-> I guess all empires to keep things balanced. Would look, again, unimmersive and a bit silly if all AI's would live mostly happily with each other while sending death threats nonstop your way.