Soren Johnson: The Chick Parabola

I think the comments regarding the lead design not being the ai coder are good. Reading the comments in Sirian's code for the map scripts also shows that civ V game designer didn't understand what the number of resources he wanted ment in terms of random maps.
I don't think 1upt is the worst mistake or cause of bad ai, though. It makes the tactical ai bad, but the strategic ai is probably worse, with diplomacy, alliances and backstabbing badly implemented.

Zhahz said:

And yes, at the very least, the AI shouldn't be making utterly moronic mistakes like moving ranged units into melee or out of positions of superiority. Even if the overall AI movement and coordination is lacking for whatever reason, it cannot be too hard to create a rule in code that says "don't freaking kill yourself" or "don't move out of a good position where you can inflict dmg and into a position where you're gonna get 1-shot"

Click to expand...

Another issue is that the AI simply moves too much. The AI needs to find a position for each unit and have them stay there unless there is a reason to move (moving to an offensive position, moving to a defensive position etc.). There's too much moving for the sake of moving (just look at your nearest CS and how they shuffle units around within their borders for absolutely no reason). If units weren't constantly moving around, there would be fewer units moving and in turn fewer pathing/blocking issues.

Ktulu said:

There's too much moving for the sake of moving (just look at your nearest CS and how they shuffle units around within their borders for absolutely no reason). If units weren't constantly moving around, there would be fewer units moving and in turn fewer pathing/blocking issues.

Click to expand...

I agree 100% I am sick of seeing this AI unit shuffle polka. This is what i was getting at when I wrote above about the AI just calculating based on the "now" rather than any long term strategic plan. I am yet to see an AI hold a fortfied line, or build forts. It either send its units on a death ride against you or it runs away like a cat from a live wire.

Rat

I have never viewed Civ as a game in the symmetrical paradigm of Starcraft. Never. By that measure, Civ 4's AI was a total and complete failure and it was obvious from the first day and has only marginally increased. Civ 4, Civ 5, indeed, none of the Civs were designed to be played symmetrically on a competitive level. One needs only to observe what happens in MP games to be certain of this.

Perhaps the problem is that most players of Civ 4 were somehow fooling themselves into thinking that they were playing symmetrically, and that beating Deity was somehow something of an achievement? I confess that I have never felt this way. Civ was always a builder's game for me. Still is.


If this is what Tom Chick is referring to in his Civ 5 review, then I have to give him even less credit for understanding the game that I already do, and I gave him less credit that IGN's reviewer, which says a lot.

There's another piece to the parabola, which is the power of the internet. With all the information readily available, players can quickly climb the learning curve and master the game in very little time. This means that without a sufficiently robust AI to challenge the players, the downhill phase can start very quickly.

As an example, I was struggling playing civ 5. Played 4 or 5 games, but hit a wall at Emperor level. Went on the forums here, picked up some pointers and moved my play level up significantly.

Without the internet, I would have spent much more time learning the game. Would that have been more frustrating or more enjoyable? I think the answer is both, as people thrive on challenge.

In any case, interesting article.

wolfigor said:

As far as I know he is responsible for Frontierville and not Farmville.
He is also responsible for CivII and SMAC, so he already reached the pinnacle of glory in gamedesign.

After so much good (with Civ and SMAC) he had balance with some evil (Frontierville) to keep yin and yang into balance

Click to expand...

Not to mention Homm2, I can't remember if it was the Price Of Loyalty addon, but one of Homm2 versions.

wurstburst said:

There's another piece to the parabola, which is the power of the internet. With all the information readily available, players can quickly climb the learning curve and master the game in very little time. This means that without a sufficiently robust AI to challenge the players, the downhill phase can start very quickly.

As an example, I was struggling playing civ 5. Played 4 or 5 games, but hit a wall at Emperor level. Went on the forums here, picked up some pointers and moved my play level up significantly.

Without the internet, I would have spent much more time learning the game. Would that have been more frustrating or more enjoyable? I think the answer is both, as people thrive on challenge.

In any case, interesting article.

Click to expand...

Really good point. In Civ2 I probably never got to the top of the parabola because I never went online to talk about it (well, and I was also a lot younger, but still).

aimlessgun said:

Really good point. In Civ2 I probably never got to the top of the parabola because I never went online to talk about it (well, and I was also a lot younger, but still).

Click to expand...

Reading strategy and discussion on the internet is so fun, yet it also puts you over the parabola since you'll learn how simple and stupid the AI actually is. There is a real trade off here. Whenever a new game comes out I think for a moment that I should play the game for a month without reading anything on the internet. I'll be in a discovery phase for so much longer and the game will last longer. I can never wait long enough though.

A good story can cover up the AI's shortcomings I think. I think this is why SMAC is so beloved. The AI's have tons of personality in the tech quotes and wonder/secret project videos and just enough programming backing it up. Now that I've played SMAC so much I can run circles around the AI and I can see how little they play differently. They're all the same overaggressive AI that never terraforms enough!

pat4 said:

I have never played panzer general but from what I heard doesn't the AI just defend positions on pre-set maps?
Also did the AI have full visibility of where there opponents units were?

That wouldn't be near as hard to program as a randomly generated map where the AI is expected to go on the offence.

Click to expand...

I can't speak for Panzer General, but the game "People's General" had a Combat AI that *really* pushed me to the limit. Although they did leave units defending key positions, the AI wasn't afraid to send out appropriate units to really make your life seriously difficult!
Also, I recall a number of scenarios where I was the defender, having to fight off a pretty determined AI attack on my holdings-using some fantastic tactics. Just FYI.

Aussie.

wurstburst said:

There's another piece to the parabola, which is the power of the internet. With all the information readily available, players can quickly climb the learning curve and master the game in very little time. This means that without a sufficiently robust AI to challenge the players, the downhill phase can start very quickly.

Click to expand...

That's exactly the reason why usually stay out of strategy forums, don't read strategy guides, etc. It's a lot more fun for me to discover these things myself, and the fun lasts longer this way. Interestingly, I still haven't exhausted Civ4 - even after 5 years of playing, I'm still honing my strategy, changing things here and there, trying a different type of economy, etc.

Great article.

Aussie_Lurker said:

I can't speak for Panzer General, but the game "People's General" had a Combat AI that *really* pushed me to the limit. Although they did leave units defending key positions, the AI wasn't afraid to send out appropriate units to really make your life seriously difficult!
Also, I recall a number of scenarios where I was the defender, having to fight off a pretty determined AI attack on my holdings-using some fantastic tactics. Just FYI.

Aussie.

Click to expand...

Interesting I might try check it out. But in relation to the article, did the AI push you if you were playing the same or similar rules (same units/reasonable objectives)?

Also was the map anywhere near as complex as a civ map can be. Like we all know how a computer can beat any human at chess because there is no variation.

Like we all know how a computer can beat any human at chess because there is no variation.

Click to expand...

They still can't do it with 100% reliability. Don't underestimate our good ol' brain.

Lone Wolf said:

They still can't do it with 100% reliability. Don't underestimate our good ol' brain.

Click to expand...

They cant intuitively on algorithm only, but they can beat humans - even GMs if they have good a games/openings library.

The point made is a fair one. If the AI is limited to a set of maps/scenarios where there is a limited or no variation the AI can be programmed for that scenario and be decently competent. But if you try to code an AI for a pretty much random scenarios with only a set of general guidelines its limitations will be apparent.

I would guess an AI made for fighting a specific battle or war on a specific map could be coded to be quite challenging. To be fair, I have seen some real stinkers like Napoleon 1813 which was just a single war.

I think Sid needs to put his hands more into the development again or get someone of Reynolds' caliber again. Games like Sid Meier's Gettysburg, Crusade in Europe and Civ II - all Reynolds' work were pretty solid.

Rat

There's another issue to be taken with the parabola curve. I'll take Napoleon Total War, as I'm a fan and modder of that, as my example. It's also the direct follow up to Empire TW, so apt for the original example.

NTW is much more preficient at amphibious invasions, than its predecessor ETW. It isn't particularly good at them, but they exist, and have to be acknowledged as a possible threat. Also, the battle ai has been improved (and then improved further by modders), but an experienced player will basically never lose a battle unless outnumbered/ outclassed in troop type.

So you play on the hardest levels, reach the top of the parabola curve, realise the ai isn't really any good, and then that's it. You lose interest right?

Well no, not in my and many other's case. You see, as Psyringe said, if there are enough interesting game mechanics you will keep playing (basically as long as you are still having fun). My point, is to add that for any game I really enjoyed on a long term basis, I make up my own house rules. In NTW, I always insist on having as large a Navy as I can afford, basically to honor a mostly non existent threat. I also try and play as historically as possible.

Europa Barbarorum (something I know very well too)is one of the most popular mods for Rome Total War. Even with all the bonuses the ai is given in its script, the ai is still pretty hapless in battle, so virtually everyone uses their own house rules. Even in Civ IV (after learning and using/abusing every mechanic/slingshot), I played with house rules even at the harder levels. For instance, no troops ever allowed to be built without a barracks (where do they train otherwise?). In Civ IV, a huge amount of the replayability for me was down to the maps, and using random personalities (I could never understand why folks played against the same leaders, time after time...Oh its Gandhi, I don't need many troops....Oh its Mansa, I'll befriend him and us him as a "friendly" research center.) The maps, to me, are always what made Civ what it is/was (delete as appllicable). Having terrain really matter (and I played Civ IV mods which accentuated this), made the game for me, planning out where cities should go was a massive proportion of the fun.

What I'm probably trying to say in this long ramble, is that the Civ series was never really about "beating" the game, and that was what gave it it's 1000's of hours of replayability. It wasn't about having some amazing ai you were constantly competing against, and that's why you never really reached the top of Soren's hypothetical curve. You just started yet another game afresh, with 100's of "what if's" running through your head. The anticipation was in fact what drew you back, time after time.

Anyways, ramble over...............

MadRat said:

They cant intuitively on algorithm only, but they can beat humans - even GMs if they have good a games/openings library.

The point made is a fair one. If the AI is limited to a set of maps/scenarios where there is a limited or no variation the AI can be programmed for that scenario and be decently competent. But if you try to code an AI for a pretty much random scenarios with only a set of general guidelines its limitations will be apparent.
Rat

Click to expand...

Case in point: One of the best "AI" opponents I've ever encountered in a computer game, was in the various (I think there are 200 of them in the stock package) scenarios in Norm Koger's The Operational Art of War III or TOAWIII for short.

This engine is a purely military engine (sadly) so it doesn't really compare with Civ. However, it was built to be a general purpose, "modern" combat engine that allows for operational (and tactical to a certain extent, depending on where you consider the boundary b/w the two) scenarios ranging from Napoleonic era to 21st century.

Firaxis would've done well to study up on this game IMO before starting to develop Civ5.

An game that combined something like TOAWIII's military engine, and even something as simple as Civ3 social/cultural/political/strategic dimensions would be an absolute stunner, with one major caveat: AI.

The "AI" (which it was not called by the TOAWIII crowd, instead that called it "PO" for "programmed opponent") in TOAWIII could be _quite_ challenging. On high-difficulty settings it could be next to impossible to achieve even as good of an outcome as the historical side you were portraying. The secret to this "great AI?" The whole game revolves around scenarios that are dependent on designers to fully encode them, and which have highly defined durations, objectives, scripted events, etc.

Aussie_Lurker said:

I can't speak for Panzer General, but the game "People's General" had a Combat AI that *really* pushed me to the limit. Although they did leave units defending key positions, the AI wasn't afraid to send out appropriate units to really make your life seriously difficult!
Also, I recall a number of scenarios where I was the defender, having to fight off a pretty determined AI attack on my holdings-using some fantastic tactics. Just FYI.

Click to expand...

As far as I'm remember ... the "defensive-mission" was against overhelming numbers ( if playing against Chinese )/quality ( if playing against UN forces ) of AI-units ...

Panzer General, People's General, Allied General were all scripted games, with fixed parameters.
That's the best way to make a "AI" work without having to do massive calculations which will take ages to complete. I know these games, just as all AOE's, TW's and sofort.

Scripting is the easiest and most efficient way to make a AI "more challenging". Problem is that once you know how a certain "script" works; there is no surprise anymore and you learn how to respond on the AI "scripted behavior".

For CIV it is extremely difficult to make a good battle-AI; the main reason for that is because of all the random parameters involded. small maps/huge maps, fast play/slow play etc.etc.etc.
Then you have numerous units and random maps. While it is a builders game "through" the ages, the AI needs to calculate what units to build when also.

There is so much involded, that i don't think that it is possible to make the AI "smarter" with the current mechanism. I don't even speak of longer thinking times to compute, don't forget the computer have to calculate more then 1 player, depending on the game you play it shall be alot more; it also have to compute one after the other, while it are all "independed" players.

Scripting also works best, when there are more fixed parameters. Again, there's little fixed in CIV's, units can build anywhere on the map, whichever map that may be. Making scripts to work for CIV 5 will be difficult if not practicaly impossible.

Jediron said:

There is so much involded, that i don't think that it is possible to make the AI "smarter" with the current mechanism. I don't even speak of longer thinking times to compute, don't forget the computer have to calculate more then 1 player, depending on the game you play it shall be alot more; it also have to compute one after the other, while it are all "independed" players.

Click to expand...

Although I agree about the difficulty to create a good combat AI for a game like Civ with its countless units, hexes/tiles and whatnotever, the calculation of the computer opponents could be done during the human's turn. After all, there is multi-core support now.

All what would have to be done is to check whether any action of the human changes the related "threat-level". As long as this does not happen (not having met the human or the human actions are not considered threatening for the respective AI opponent) the calculation can just go on.

The problem is that the Firaxis guys have never identified their core problem: they (or better, the "AI's" created by them) aren't investigating a certain problem, they are running through the units and then check what could be a problem.
See just how meaningless the AI moves units back and forth. This is so stupid that alone for this one could slap them into their faces all day long.

pi-r8 said:

So if the land units that turn into boats are good for the AI, how come the AI is unable to do any intercontinental invasions in Civ V? I've never even seen them try.

Click to expand...

Playing England on an archipelago map, I saw Japan try to launch an invasion against me. It was a disaster, since my two caravels and two ships of the line simply ate four embarked units per turn for three turns, and then there was nothing left.