How Should Beyond Earth Handle AI Difficulty Scaling?

^ Re the above few posts, putting aside the issue of feasibility for Civ, Game AI whose strength is based on depth of search/number crunching would arguably not be a fun AI to play against anyway. It might be academically interesting but I doubt fun, when the only way to get better at the game is to spend longer and longer thinking to deeper depths each turn yourself. Civ already has that intrinsically built in with economic micro - we don't need any more!

IMO. Of course some find trying to beat e.g. chess AIs fun but it's not "fun" in the same sense as most people use the word :lol:
 
I only have 2 major complaints with how AI bonuses have been handled in the past. I can agree with all of the OPs points though.

1. Unlimited happiness, which allows the AI to expand throughout the game indefinitely, which the player cannot do.

2. Huge production bonuses, allowing the AI to spew out units to such an extent making warring more of a chore than fun. It doesn't increase difficulty beyond the early eras and becomes just an annoyance.
 
Having CPU power for high end AI is already here... take a look at Deep Blue in the game vs. Kasparov, a match and computer design back in 1997. We're in 2014... do you want me to believe that the hardware of '97 > '14? Or more recently, the game of Vladimir Kramnik vs. Deep Fritz. Guess what was that AI running on? 2 Intel Core 2 Duo CPUs... So this whole argument about not having the hardware is bogus.

Smarter AIs can definitely be possible, it just comes down to how committed they are to the project. If they're so stubborn as to persist on the 1UPT system, treat the AI that runs combat with a Chess mentality and treat empire and territory management with Go mentality. There are very high end AI programs for both games, so it is very well within the realm of reality to create a system that can manage both. And to throttle the difficulty of AI, limit the number of moves/options it calculates ahead equal to the difficulty of the game the player wants to play in. If a masochist enjoys the game so much that they prefer the AI have insane bonuses and play a completely different game, give them that option. In the same way in Civ4, raging barbs, no barbs, always war/peace can be turned on or off.

As of right now, playing higher difficulty levels with the AI is like playing chess with a player with them obeying completely different rules. Instead of starting out with only 1 queen, the other player starts with all of his back row pieces with queens and of course 1 king. Oh and he gets to move 2 pieces per turn. Oh and your queen piece is replaced with another pawn. Oh and even on the attack when your piece can legally attack and remove one of his, you've got to roll a loaded die to determine whether your attack with your Rook against his Pawn succeeds.
 
I certainly won't argue that the tactical AI can't be made better (it surely can) but Deep Blue and chess is not a good example or a good model to follow. Chess has a limited number of tiles and pieces and no random elements, and only one piece can move per turn; there are only a limited number of possible moves for each piece on a given turn, and only a finite number of positions of pieces on the board. Deep Blue operated by analyzing every possible move from each turn to 10-20 turns into the future, and comparing the outcomes with thousands of recorded grandmaster games. This kind of brute-force evaluation is not viable for a system as complex or dynamic as that in Civilization. And Deep Blue was a custom-built supercomputer that could evaluate 200 million positions per second, compared to the 8 million positions per second of the PC version that faced Kramnik in 2006 (though the latter had more efficient algorithms, to be sure).

It is interesting to note, though, from the Deep Blue vs. Kasparov matches, that imperfections in its algorithms caused Deep Blue to occasionally make erratic moves, which Kasparov misinterpreted as "deep intelligence and creativity." This reportedly demoralized him to the point that it threw him off his game in later matches, and afterward he became convinced that human operators had intervened, and he accused IBM of cheating. So, if your AI is competent, minor imperfections can actually make it seem more human.
 
TheGhostAgent said:
Smarter AIs can definitely be possible, it just comes down to how committed they are to the project. If they're so stubborn as to persist on the 1UPT system, treat the AI that runs combat with a Chess mentality and treat empire and territory management with Go mentality. There are very high end AI programs for both games, so it is very well within the realm of reality to create a system that can manage both.
Click to expand...

Do you realize how much research went into creating capable AIs for those games? Do you think it is realistic that a game company expends that kind of effort.
 
Hold on there, guys :crazyeye: Yes, chess is different but I would discourage anyone from dismissing those same AI principles in a practical civ context. I hope I can get some results soon-ish so I can speak from an *actual* practical position, but so far things have held up. Gotta allocate() more time!

I mean, chess searches are characterised by sharp back and forth alpha-beta pruning between turns(amongst other things). In civ you just don't have that. It would be a mistake though to say that in civ cutting nodes is fundamentally different, therefore computationally ineffective. It has other things to take advantage of. In civ you make long series of maximising moves. The things that make sense to look for are the things that suggest a move will lead to high average value of your remaining moves. For example, a move allows and blocks certain moves from other pieces near it. You end up evaluating a lot of moves and, since moves are relatively indepedent, you have a lot of relevant data to compare things to - almost like a bunch of parallel universes where swordsman x went on to achieve different things. It's one of many heuristics outside the position evaluation itself. And there are endless things to exploit to improve your odds of never searching the unproductive or redundant stuff, but I don't think people want a random wall of text :p

Anyway, It's all about the selectivity of your search. If you can get useful info, you can prune. The question for civ is whether, in the end, the info is good enough for the amount you need to prune. Ultimately you have to get the branching factor under control, so did you manage to keep the best moves in what's left?

I guess alpha-beta can take you for a ride between turns, but civ turns aren't very tactically sharp after just a couple have gone by. I think the big thing would be making it 2 turns deep - the end of your opponent's turn. I didn't even mention pruning based on tactical blunders because computers are just so good at that already, but turn 2 sounds strong. Who lets an injured unit die after regaining control of it?
 
Lieu said:
I guess alpha-beta can take you for a ride between turns, but civ turns aren't very tactically sharp after just a couple have gone by. I think the big thing would be making it 2 turns deep - the end of your opponent's turn. I didn't even mention pruning based on tactical blunders because computers are just so good at that already, but turn 2 sounds strong. Who lets an injured unit die after regaining control of it?
Click to expand...
(sounds like you got yourself an interesting research project there :) )

You have a point here. It does however go against the notion that some people here have put forth that you could create a very scalable AI by scaling the search depth. It may very well be that increasing the search depth beyond 2 turns (if one were to achieve a scalable algorithm that could manage that) adds very little to the performance of the AI.
 
Game players want an AI which is about as smart as they are. If it's less smart they'll say it's dumb, if it's more smart they'll say it's random, go down a level till they can beat it, then say it's dumb anyway! Challenging players of a mainstream game with AI "smartness" is impossible because "smart" means different things to different players. Just make it believable, and most importantly, fun :D
 
Kid R said:
Game players want an AI which is about as smart as they are. If it's less smart they'll say it's dumb, if it's more smart they'll say it's random, go down a level till they can beat it, then say it's dumb anyway! Challenging players of a mainstream game with AI "smartness" is impossible because "smart" means different things to different players. Just make it believable, and most importantly, fun :D
Click to expand...

I think that is mostly a consideration for the diplomatic AI. When it comes to using units effectively, it becomes a lot more objective, and most players will agree on which AI is smarter. They certainly agree on things that are very unsmart, and it is probably safe to say that most players don't think it is fun when the AI makes inconceivably stupid moves.

Moreover, more efficient unit use, would mean that the AI needs less units to provide a challenge, which again most players will agree is more fun.
 
Be careful what you wish for. If the ideal is an AI that (on the highest level) plays as well as a really competent human being, what would follow?
  • If you have 8 factions you might expect to win 1/8th of all games unless you are really really good.
  • Even if you are and plough ahead, the sensible thing for the AIs would then be to gang up on you and knock you out of the competition.
  • All the backstabbing that people complain about in countless Diplomacy threads will become the dominant strategy (because it is the correct one: You don't win by holding hands with the score leader).

As long as there is Winning (and no joint victory for allies), a competent AI will lead to nothing but frustration for players.

Even the combat AI can't be too good. Otherwise I would expect to lose a significant portion of my forces even in victorious campaigns. The next battle will then be against a different, non-decimated faction and they will wipe the floor with me. If you wan't to stand a chance against 7 opponents you have to be quite a lot better than they are. If you don't want to start giving bonuses to the player in the way Civ V does for AIs on higher levels, the AIs need to be pretty moronic.
 
AriochIV said:
This is what difficulty levels are for. But it would be nice if on higher difficulty levels the AI player better, instead of just cheating more.
Click to expand...

The problem is you can't just turn a dial that makes the AI smarter.

What you are asking is
1. They make a good AI (we all agree on that)
2. They then make a slightly stupider AI, balancing its stupidity just so
3. They then repeat #2 for every difficulty level below the AI that they made

You then have a problem if the AI in # 1 is not hard enough for the high difficulty levels

I would much prefer
1. They make a good AI
2. They give it easily tunable handicaps for easier difficulty levels, and easily tunable bonuses for higher ones (some bonuses/handicaps can be given to the human player as well)

This way
1. They don't waste AI development time making a deliberately stupid AI
2. Difficulty can go as high or as low as they want
 
boghog said:
Be careful what you wish for. If the ideal is an AI that (on the highest level) plays as well as a really competent human being, what would follow?
  • If you have 8 factions you might expect to win 1/8th of all games unless you are really really good.
  • Even if you are and plough ahead, the sensible thing for the AIs would then be to gang up on you and knock you out of the competition.
  • All the backstabbing that people complain about in countless Diplomacy threads will become the dominant strategy (because it is the correct one: You don't win by holding hands with the score leader).
.
Click to expand...
You're assuming that there is no competitive viability in multiplayer, that a game that tests skill in multiplayer is always impossible because of ganging up.

I have words for that, but I think I'd prefer to start this post with my intuitions along a different axis.

I can see enough to tell we want different things from the game, but we can both have what we want. I want a setting where the AI actually tries to play the game, where it gives me something like a preparation for play with Humans, who exploit the rules as they are, and the fact they can't win together keeps any two in alliance from staying too cozy. But what you want is clouded by your concept that multiplayer is inherently unfair. Can you explain that? I question why you take interest in a game built with a whole dimension of said multiplayer components, only to dislike their presence. You want a certain experience, and to me it doesn't make sense that you look to 4X games with multiplayer to fulfill this want (as it's like buying a race car so as to take it apart and construct a skydiving rig from it), but I won't say the game cannot accommodate you. You want the AIs not to do something gamey like targeting the player just because they know who the Human is. I want an AI that would never do that either, but would target me or not target me because that is what a logical analysis says is the best strategy. (I think a third group, that I'm also a member of, wants a setting where the AIs act in a roleplaying manner, but still aren't tactical morons.)


I'm telling you, multiplayer in this genre is fair. In this game it's fair. Knowing that, would you still want the AIs to be deliberately stupid? I bet you have a picture of "beating all the AIs" which looks like beating each of them, one-on-one, in a mere 8-way race. It's cool if that's what you want to do, but I am sure you are mistaken to think that a field of rational enemies makes any strategic play through them impossible.

When you wrote about war leaving you weak, I didn't see a proof that slogging through 7 opponents is impossible, I saw a proof that war is weak.

The fear of getting into a fight that doesn't end swiftly is exactly what keeps the stage balanced. This tension, the way that the politics make even a position of strength a weakness, is what this game is about and is the reason I play it. I don't want to argue you need to appreciate what I do, but I am held up by your proposition that this game's true form is uncompetitive.
 
I think players see MP differently. We accept that only one player can win and if we are evenly matched its unlikely to be me depending on the number of players.
In a solitaire game of any kind I want a decent chance of winning and my intuition says that that's true for a big part of Civ's buyers.

I'm not saying clever AIs would gang up on you because they know you're the human (although it would make some sense) but that they will gang up on you as soon as you became the frontrunner, which on your way to victory you would most probably do at one point. I've got enough experience playing MP games against ambitious players to know that that's what humans do - it's what I do when someone pulls too far ahead. Of course these alliances aren't stable but they last long enough to ruin your chances of success. As I said, from humans we tend to tolerate this sort of behaviour, from AIs we might not.

You frequently read complaints on these forums like "Caesar DOWed me even though I was clearly ten times stronger than him. The AI is stupid."
But it isn't. It's exactly what a rational player would have to try to stop you getting too far ahead. It's of course not what a responsible, rational ruler of a nation would do because in RL his aim would be to have a stable country full of happy citizens, in the game there is no second place so you have to be prepared to risk everything in order to win the game. That is rational - and stupid at the same time.
 
AriochIV said:
Of course you can. You handicap the AI at lower difficulty levels, instead of cheating at higher difficulty levels.
Click to expand...

That's just cheating in reverse.
If the AI has to pay 200% of what I do for maintenance, that is NOT making the AI stupider
The AI is not 'smarter' if that changes to 150% or 100% or 50% or 5%.

The point that they should make the AI smart so that it can use the same rules as you on a high difficulty is NOT an issue of how to scale the difficuties (they are still scaled with handicaps/bonuses)
 
KrikkitTwo said:
That's just cheating in reverse.
If the AI has to pay 200% of what I do for maintenance, that is NOT making the AI stupider
The AI is not 'smarter' if that changes to 150% or 100% or 50% or 5%.

The point that they should make the AI smart so that it can use the same rules as you on a high difficulty is NOT an issue of how to scale the difficuties (they are still scaled with handicaps/bonuses)
Click to expand...
You can scale the smartness of the AI; you don't have to resort to handicaps. A good example of this is in XCOM: on the lower difficulty setting, the tactical AI often makes sub-optimal moves, and sometimes doesn't take opportunities to screw you. On the harder levels, the AI is more aggressive, and will rarely pass up an opportunity to flank you when you overextend or kill a solider out of cover.
 
boghog said:
I think players see MP differently. We accept that only one player can win and if we are evenly matched its unlikely to be me depending on the number of players.
In a solitaire game of any kind I want a decent chance of winning and my intuition says that that's true for a big part of Civ's buyers.

I'm not saying clever AIs would gang up on you because they know you're the human (although it would make some sense) but that they will gang up on you as soon as you became the frontrunner, which on your way to victory you would most probably do at one point. I've got enough experience playing MP games against ambitious players to know that that's what humans do - it's what I do when someone pulls too far ahead. Of course these alliances aren't stable but they last long enough to ruin your chances of success. As I said, from humans we tend to tolerate this sort of behaviour, from AIs we might not.
Click to expand...
That is how I described I see MP, with one omission. If that is what happens, then clearly it is a mistake to become the frontrunner.

You frequently read complaints on these forums like "Caesar DOWed me even though I was clearly ten times stronger than him. The AI is stupid."
But it isn't. It's exactly what a rational player would have to try to stop you getting too far ahead. It's of course not what a responsible, rational ruler of a nation would do because in RL his aim would be to have a stable country full of happy citizens, in the game there is no second place so you have to be prepared to risk everything in order to win the game. That is rational - and stupid at the same time.
Click to expand...

That is what the "roleplaying" setting might be for. However, it's still a separate issue from tactical stupidity. The AI should manage to fight a war like a not-moron on the occasions it does go to war for whatever reason, at least. Do you know about the ranged predestination exploit? It's why the AI can't fight with ships. The AI for Civ , for any game, should be free of this; it's pathetic.
 
You must log in or register to reply here.

Similar threads

T
Tier 2 social policies: Fealty a must-have in most situations?
Replies
6
Views
1K
randomnub
R
Taefin
Taefin's More AI Cheats
Replies
2
Views
666
Taefin
Taefin
T
Better way to implement AI difficulty bonuses
Replies
1
Views
500
Krikkit1
K
L
7OTM06 - Completion of Modern Age Spoiler Thread
Replies
8
Views
623
GoblinMatron
G
notque
RHQ Artificially Intelligent AI Mod
34 35 36
Replies
713
Views
18K
ggalindo001
ggalindo001
Back
Top Bottom