Will Civ ever conquer its late-game malaise?

[Trav'ling Canuck]

All players in a game should be trying to win the game, whether they are AI or human. If 1 of those players is about to win the game, the other players should be able to recognize this and make every attempt at not losing the game. Those are basic tenants that make games competitively enjoyable. My opponents not trying to win is not enjoyable.

You can find this experience in multiplayer.

Right or wrong, it's not how single player is designed in Civ 6. See leader agendas as proof that the developers take a different view.

Personally, I don't view the AI avoiding certain actions (like spontaneously nuking a fiend friend LOL) as the equivalent of the AI not trying to win. Ultimately, the AI exists exclusively as a challenge to the player, who is the only active actor in single player and therefore the only one really trying to win. Many players, myself included, consider the diplomatic aspects of the game as an enjoyable aspect of trying to win. The more AI civs you can convince to be friends, the fewer you need to worry about defending yourself against. That's a more interesting (and varied) form of challenge than knowing that all AI civs will automatically declare war every game once a known and predictable threshold towards victory is tripped.

 

[Abaxial]

Perhaps we should stop calling Civ 6 a game, and think of it as a pastime. I find tinkering around with my empire is just a relaxing way of filling time. As to matching victory conditions to history - well, history doesn't have a game end or victory conditions. Various nations have thought of themselves as the "winners" throughout the ages, but play continues ...

 

[Gedemon]

??? I'm arguing that the AI should try to win based on the defined victory conditions. If you literally remove all of the victory conditions, only a bugged slop of spaghetti code AI would estimate that you're a threat to win and go for a backstab. Considering that you can't possibly win with those settings, why would we be discussing the AI backstabbing a "winner" with them active? That's is a red herring at best.

Granted, I wouldn't put that past Firaxis to make an AI think the player is near an impossible win condition, but it SHOULDN'T happen.

I just wanted to make crystal clear that I wasn't playing Civilization like I would play "Mario Kart by bumping into walls", so I can present other arguments after proving that I'm not a fool.

Now that's done, so let's go over those for your chess and race game.

The AI goal in those game is not to win, especially in race game.

For chess, now that it can beat any human player, it's goal is to teach the human how to be better at the game, which means it has to be tuned down depending on the human's level.

For any race game, the AI is specifically coded to not be perfect, so that taking over another car could still be possible (and because there are already Train Simulators out there if you want to play with fixed positions on rails), and in some arcade race game, the leading cars are slowed down so that the human can catch them.

Granted they are not coded to bump into walls every time, but still, in fact they are, just a bit, so that one or two AI car may lose control during a race, while, I agree, Civ6 AI give that impression in almost all it's diplomatic decision...

IMO a Civilization AI should know how to win, of course, but, again, winning should not be it's main goal, it's main goal is to entertain the human player(s).

Providing a challenge is part of that, so mechanisms allowing smaller civilizations to create alliances between each other in late game so that they can still weight against leading civilizations are needed and more than welcome, but breaking the fourth wall in such a dramatic way as mass-backstabbing a Civilization because it's "winning" is the worst possible way of making the late game more interesting.

Using the race game example, for a simulation you don't want a "slow down the leader" mechanism or a "I'm second, he's going to win, let crash into him to prevent that" mechanism coded in the AI, but in a real race (most obvious example being non-mechanized races) you may see opposite teams working together in a pack until they catch up on the leader.

And finally, let's suppose for the sake of argument that the AI should be coded to prevent the leader to win at all cost, if the leader is human and with the state of the military AI, declaring war is the worst option, it would be much easier from a coding perspective to implement other mechanisms allowing the AI to catch up by working together than coding a tactical AI able to win a war against the human.

Excuse me to use your example again, but an AI civilization declaring war to a leading human player is very similar to an AI car deciding that to drive into a wall in a race game is a good decision.

That line of logic doesn't translate to Civ. The US isn't about to "win the game" in any fashion IRL. There is no game to be won. The world isn't ending in 2050 with a prize going out based on what nation has the highest score (my money is on China if there was).

We are literally playing a game, and you're complaining about it being a game. This does not compute.

"Game play trumps realism" has never meant that playing a game must be a completely unrealistic experience.

I'm okay with banana launchers and "we're accelerating your car because you're not first" mechanism in games like Mario Kart, but I don't want that in Asseto Corsa.

I hate it when someone try to sneak in "We don't like you because you're winning" modifiers in a Civilization game, there are other solutions, much more elegant.

 

[TrailblazingScot]

Has any 4X game beaten the late game malaise? I think it's just a quirk of the genre. That being said, it could certainly be better than it is now.

Sid Meier's Colonization (1994) culminated in a revolutionary war against a well equipped Royalist Army. This event is dramatic and different from what has come before and makes for an excellent end game.

I'd love a re-make

 

[Chibisuke]

Yea I think the argument can't really be resolved because of the philosophical differences I mentioned before. Some people don't want the game to be designed with fun for the human as the primary focus. They want the game to be designed to try and %$&@ with the human player, because they find that to be a challenge and like that challenge. If your biggest goal is fun for the player, the AI shouldn't try to all DoW just because you are in fact winning. If your biggest goal is that you want to be %&$@'d with by the AI so the game seems as hard as possible, then you will want them to do that. I don't think that will change as long as people hold such different philosophies.

 

[shaglio]

Sid Meier's Colonization (1994) culminated in a revolutionary war against a well equipped Royalist Army. This event is dramatic and different from what has come before and makes for an excellent end game.

I'd love a re-make

There was one.

 

[TheMeInTeam]

I just wanted to make crystal clear that I wasn't playing Civilization like I would play "Mario Kart by bumping into walls", so I can present other arguments after proving that I'm not a fool.

The point of that example wasn't to judge people for doing it, but to point out that it's not a useful implementation for an AI regardless of whether people do it.

The AI goal in those game is not to win, especially in race game.

For chess, now that it can beat any human player, it's goal is to teach the human how to be better at the game, which means it has to be tuned down depending on the human's level.

For any race game, the AI is specifically coded to not be perfect, so that taking over another car could still be possible (and because there are already Train Simulators out there if you want to play with fixed positions on rails), and in some arcade race game, the leading cars are slowed down so that the human can catch them.

Granted they are not coded to bump into walls every time, but still, in fact they are, just a bit, so that one or two AI car may lose control during a race, while, I agree, Civ6 AI give that impression in almost all it's diplomatic decision...

IMO a Civilization AI should know how to win, of course, but, again, winning should not be it's main goal, it's main goal is to entertain the human player(s).

Chess AI is still allowed to dumpster human players, it has to be held back. The same is true for racing games. In both cases, AI is at a level where it can defeat elite human players without bonuses routinely. In both cases, the AI is still generally performing moves in line with the game's stated victory condition. The comp for this in Civ is not "AI shouldn't try to win". The comp for this in Civ is the difficulty setting.

And the fact that Civ AI is not in the same air as AI in the examples above and can't win against good players no matter how hard the devs make it "try" in its present state. Civ is nowhere near the point where AI has to be tuned back so it's possible to win.

Providing a challenge is part of that, so mechanisms allowing smaller civilizations to create alliances between each other in late game so that they can still weight against leading civilizations are needed and more than welcome, but breaking the fourth wall in such a dramatic way as mass-backstabbing a Civilization because it's "winning" is the worst possible way of making the late game more interesting.

Trying to win is not "breaking the 4th wall". It's playing the game. "Breaking the 4th wall" would be something like insta-targeting the player just because it's the player, without any other context about the game state.

Using the race game example, for a simulation you don't want a "slow down the leader" mechanism or a "I'm second, he's going to win, let crash into him to prevent that" mechanism coded in the AI

If you don't want these things, the proper implementation is to not include them as mechanics. Including them then having opposition not interact with them is awkward and self-inconsistent.

And finally, let's suppose for the sake of argument that the AI should be coded to prevent the leader to win at all cost, if the leader is human and with the state of the military AI, declaring war is the worst option, it would be much easier from a coding perspective to implement other mechanisms allowing the AI to catch up by working together than coding a tactical AI able to win a war against the human.

Excuse me to use your example again, but an AI civilization declaring war to a leading human player is very similar to an AI car deciding that to drive into a wall in a race game is a good decision.

This isn't just a "finally" throw-in, it's a completely different argument. In essence, you're saying that because devs and players alike know the tactical AI is awful (nothing like the racing/chess examples cited above in AI quality), devs could in principle knowingly compensate for this by having AI collude while avoiding it's most obvious weakness: actually fighting.

It doesn't address the "should AI try to win or not" debate, rather this particular quote is arguing which approach the AI should take to try to win. You could make a legit case that the AI is so inept offensively that its win probability estimation for DoW --> offensive war should be abysmal in most cases, but that's not really what we're debating .

Yea I think the argument can't really be resolved because of the philosophical differences I mentioned before. Some people don't want the game to be designed with fun for the human as the primary focus.

This is excessively disingenuous. I can counterclaim that players who don't like the AI playing the same game don't like fun because they prefer to not play the games they play. That they are essentially insisting that the AI be like that kid who does his own thing in board games, doesn't try, and impacts the outcome of the game for everyone else arbitrarily.

"You must not like fun" is not a useful argument nor can it genuinely address any of the discussion to this point...which is why I did not make such a claim.

 

[Chibisuke]

Notice that I didn't say your philosophy was "doesn't like fun." I said your philosophy is that fun for human player isn't the primary objective of the game's design (meaning the way the AI functions). Your philosophy says that you want the AI to attempt to frustrate the player in an absolute sense. I.e., the AI should engage in last ditch desperation tactics, perhas even dirty tactics, that may not be in its interest as far as its own victory condition but instead is intended to stop the player from winning. (I think it should be evident to all of us that desperation tactics in strategy games rarely result in the victory of the person who used them and usually simply allow a third person to win.)

You believe that makes sense because the AI can't win if the player wins. But there are problems with that way of thinking.

First of all, I think it should be evident that the player winning a challenging game that requires careful planning and execution of strategy IS a win for the AI. The AI is not an independent person. It has no possibility of personal satisfaction from defeating the player. All it has is a programming function. My philosophy and that of others is that the AI's primary function is to present a game that is entertaining for the player. Whether the AI wins or loses doesn't matter to it because it is incapable of having such feelings. If the player had fun win or lose, the AI has succeeded and the player will purchase more games, thus making money for the game developer.

Another problem is that the philosophy of AI seeking to win by desparation tactics not supporting it's victory condition is that it turns a strategy game into a chaotic free for all that is probably much more dependent on luck of the draw than anything else. For example, a player in a central location simply is incapable of winning whereas a player on the coast can win. This is because any centralized player who comes close to victory will be immediately attacked on all sides whereas a player on a coast can only be attacked by adjacent enemies with limited attack lanes.

Without such a desperation mechanic, a centralized player CAN win through careful manipulation of the game's diplomatic and defensive mechanics. But if you implement a feature where all AI suddenly ignore their usual agendas and attack without care of consequence, then the players attempt to manipulate the game's diplomatic features becomes meaningless.

Fun is not a primary objective if the game is designed to make your hard work mastering it's features meaningless. Yes, that is a greater challenge, and if your primary goal is increased difficulty you will achieve that. But again, this is where the differences in philosophy come in.

 

[Gedemon]

The point of that example wasn't to judge people for doing it, but to point out that it's not a useful implementation for an AI regardless of whether people do it.

I disagree again, programming the AI to simply play the game is useful.

In a race game the AI is not programmed to win, it's programmed to do fast turns on a track while driving safely enough to avoid most crashes. This leads it to finish the race, sometime first if the human players failed to win. But it's not programmed to try to force the first player to crash to prevent him from winning.

Still, the victory condition is clear, if the player finish first, he wins, so why simrace AI aren't coded to try to make the first player crash then take its place, as the physical rules of the game would allow that with a small bump on the rear at the right time ?

Well, because no one want to play against such a frustrating AI, but in the same time, note that some humans may use this exact winning strategy against the AI, because it's not going to protest, or rage-quit, is it ?

Same approach is valid for Civilization, the AI should know how to use the different mechanisms to grow its Empire (ie "play the game"), which in turn may allow it to reach one of the victory conditions (ie "win the game"), and it can be programmed to reach one faster (and it is, yes, should be better, but it is), but it shouldn't be programmed in a way that will make it try to prevent the human to win at the cost of itself not finishing the game (or at the cost of the human player rage-quitting after 4000 years of diplomatic actions to build alliances)

Chess AI is still allowed to dumpster human players, it has to be held back. The same is true for racing games. In both cases, AI is at a level where it can defeat elite human players without bonuses routinely. In both cases, the AI is still generally performing moves in line with the game's stated victory condition. The comp for this in Civ is not "AI shouldn't try to win". The comp for this in Civ is the difficulty setting.

And the fact that Civ AI is not in the same air as AI in the examples above and can't win against good players no matter how hard the devs make it "try" in its present state. Civ is nowhere near the point where AI has to be tuned back so it's possible to win.

I agree on that, but does it means that chess and race games are bad examples when talking of an AI for a Civilization game ? edit: Because you bring them in the discussion, not me.

Still there are some point where civ AI was tuned down on purpose to remove some frustration for the human player, like rigging election in CS for civ5 or some sabotage missions in civ4

Trying to win is not "breaking the 4th wall". It's playing the game. "Breaking the 4th wall" would be something like insta-targeting the player just because it's the player, without any other context about the game state.

Anything mentioning "you're playing a game" in a game is breaking the fourth wall. That's the definition. Some game doesn't have any fourth wall, because their core design is just to be a bare game, but Civilization is about leading an Empire to stand the test of time, victory conditions are just additions on top of that.

If you don't want these things, the proper implementation is to not include them as mechanics. Including them then having opposition not interact with them is awkward and self-inconsistent.

I'm all for reworking victory conditions to help the AI, yes.

But even if I the victory conditions stay as they are, I still disagree on using "massed dogpiles" on the leader, that solution is worst than the problem it tries to solve, I prefer the diplomatic approach (see below)

This isn't just a "finally" throw-in, it's a completely different argument. In essence, you're saying that because devs and players alike know the tactical AI is awful (nothing like the racing/chess examples cited above in AI quality), devs could in principle knowingly compensate for this by having AI collude while avoiding it's most obvious weakness: actually fighting.

It doesn't address the "should AI try to win or not" debate, rather this particular quote is arguing which approach the AI should take to try to win. You could make a legit case that the AI is so inept offensively that its win probability estimation for DoW --> offensive war should be abysmal in most cases, but that's not really what we're debating .

Maybe I've missed some posts, and I apologize if that's the case, but we're not arguing "should AI try to win or not ?".

We're arguing on "should AI only play to win or should winning be only one of its objectives when playing ?"

IIRC that point of the debate started by saying that the diplomacy should be none existent in the game, except to prevent the leader to win at all cost, because the winning conditions allows only one winner.

Maybe they should reintroduce the "always war" option, I won't be against that, but I don't think it should be the default option, and I won't use it.

But IMO they should simply reintroduce shared victory options, with the "teams" being build during the course of the game, using diplomacy, not from the start.

 

[shaglio]

I agree on that, but does it means that chess and race games are bad examples when talking of an AI for a Civilization game ? Because you bring them in the discussion, not me.

To be fair to Phil, I was the person who originally mentioned auto racing - although I wasn't referring to racing games, but real life auto racing (Nascar, Indy Car, etc.).

Also, rather than using the word "gamey," maybe we should be using "metagamey" instead.

 

[Gedemon]

Well then I do apologize for having missed some posts, sorry.

 

[shaglio]

Well then I do apologize for having missed some posts, sorry.

It wasn't the greatest post, but it did get a like.