Soren Johnson: The Chick Parabola

There are definitely some very basic problems with the AI, and it reeks of a rush job.
Unescorted ranged units. Vulnerable ranged units, which place themselves next to melee units. Lack of mounted units. Units not taking advantage of rough terrain. AI being too aggressive.
All of those basic problems should be solvable.

And then there are the balance problems. An AI wont be able to misuse a balance issue, because the issue shouldn't be there in the first place, and is usually a sign that the developers haven't seen it. For example, spear vs. horseman. Horseman have the initiative (much higher move), so can decide where to attack (against the AI at least), and any promotion will cancel the spear's ingrown advantage because of the horseman's higher initial strength. This translate into spears not being a viable counter against horseman.
The more of these, the worse the AI seems. Any exploit equals worse AI performance. Just like in colonization, which was ridiculously broken.

Of course, that doesn't mean that the demands of the AI hasn't increased with 1UPT, which makes it IMHO a bad design decision. You could have most of the gameplay change with fewer AI consequences with a more flexible system.
 
Yes, definitely, but I think it would take a lot of work, and I'm afraid it would have to be done here by a bunch of civFanatics who know the game well. And unless the current code is seriously inefficient, a better AI would also require waiting for more powerful computers to arrive, since a more sophisticated combat AI would need a lot more computations between turns.

I would like to remind you of Panzer General II (1997). 1upt + competent AI. Certainly didn't need a powerful computer back then. :)
 
1UPT shouldnt be too complicated for the AI. We modern multi core CPU's, moving units effectively that doesnt take 1 minute to process should easily be possible. It really wouldnt have been hard to consult a couple of mathematicians to create an efficient algorithm to handle AI movements. Movements could have been totally predictable for the AI, hell, make it so his units always have +1 sight on higher difficulties if thats what it takes.

It doesnt take a veteran gamer to see that there is something seriously wrong with an AI that moves artillery units all by themselves. Not to mention the fact that when they have a gun perfectly placed inside of a city to potentially deal a lot of damage, they move it out of the city where one of my infantry units can easily take it out. Now I can shell the living hell out of the city w\o retribution, thanks!

For a game thats primarily single player, the AI seems to have been thrown together at the last minute. Ive been playing Civ V for like a month now (and I have never played a previous Civ) and I can consistently beat the AI on Immortal. I have taken half a dozen cities whilst killing dozens of units while losing one or two units. The AI being able to produce units on a 2:1 scale is no problem when you can kill them on a 6:1 scale on the battlefield.

The lack of stacking creates extremely intricate, and stupid, pathing problems. The lack of stacking also creates vulnerability to getting bushwhacked, compounding the issues. Add in too few movement points, too few hexes, and it is just not a well posed problem. I'm afraid that I doubt it's soluble - and if they did get something working it would probably be super-annoying rather than challenging.
 
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.

How often have there been intercontinental invasions in history? I'm relieved that I don't see the AI sending boats on multi-decade trans-oceanic invasions that are doomed because by the time the slow moving boats arrived their technology was two generations behind.

I've seen the AI launch shorter marine invasions in Civ 5. Napoleon once $%#^& slapped a little Coal-grab I made on an island near his territory. That's a much more realistic scenario than the time in Civ 4 that Napoleon sent a boat loaded with knights and catapults across half the globe to smash their heads against the treads of my tanks because I wouldn't adopt Hinduism.
 
The lack of stacking creates extremely intricate, and stupid, pathing problems. The lack of stacking also creates vulnerability to getting bushwhacked, compounding the issues. Add in too few movement points, too few hexes, and it is just not a well posed problem. I'm afraid that I doubt it's soluble - and if they did get something working it would probably be super-annoying rather than challenging.

Hell, forget the AI's problem with it. I have that problem. My units are always getting logjammed.

I think it's safe to say that 1upt was a mistake. This isn't Panzer General. We need units to explore, build improvements, settle, etc. It's not just a war game. The distinction between civilian and military units helps a little, but since that distinction doesn't carry over to the AI's units, not enough. It's also the reason we have to have archers that shoot over a range that, when scaled to the map, would be like firing a bow in New York and hitting somebody in Ohio.
 
I do enjoy the embarked unit method (instead of building transports), but in BtS the AI had no problems building transports and moving huge armies across water and protecting them in transit.

The big AI issue here is 1upt. Its created a big problem for the AI, not for the players. I wish they would take a step back and reconsider some of the 1upt restrictions. Being able to stack civilian units and have them move through tiles that contain peaceful units from another civ are key, and it might clear up a lot of the pathing issues.

The other issue that causes pathing problems is the automated-movement AI. When you ask a unit to move to a tile that is more than one turn away, the AI handling the path has one rule it always follows: Always use up all the MPs allocated for that turn. Thus it will never keep an arty 1 step behind its melee units if it has to sacrifice an MP to do so. Similarly, units that guard against flanking (I use Cavs and Tanks for that) will never give up some MPs to maintain a tactically sound formation.

If you can't quite follow what I am talking about, there is a simpler example. Think of a bunch of units taking up position on a road or railroad (straight line, like a queue), and you are trying to move another unit beyond them on the same road (lets say its a worker). If the worker comes up to the point where the unit line/queue is formed and still has some MPs left, but not enough to go through all the units and come out the other side, the AI will make him leave the road/railroad so that all the MPs for that turn are used, instead of clicking "do nothing" and waiting until the next turn. In the next turn, the AI will move back to the railroad tile and continue moving. Where I doing it, i would have just stopped on the road and continued moving in the next turn.

TL;DR The AI needs to be taught that giving up MPs in a turn can be a good thing.
 
That's a much more realistic scenario than the time in Civ 4 that Napoleon sent a boat loaded with knights and catapults across half the globe to smash their heads against the treads of my tanks because I wouldn't adopt Hinduism.
Isn't that exactly what happened with the crusades? Well, except that the middle east didn't have any tanks, and Hinduism was replaced with Christianity.

The main problem with the naval AI is just that they don't focus fire. If they sent their navy first and actually tried to kill your navy an invasion would be much more successful, but at the current stage they just move their navy around randomly and sometimes shoot a shot here and there. They will almost never actually sink one of your ships.

I haven't played Panzer General, but since it's a tactical war game I can't imagine they would send their artillery as cannon fodder. The limitations are definitely not that "it's so complex that's it's almost impossible to make a better AI" as some seem to think.
 
Isn't that exactly what happened with the crusades? Well, except that the middle east didn't have any tanks, and Hinduism was replaced with Christianity.

The main problem with the naval AI is just that they don't focus fire. If they sent their navy first and actually tried to kill your navy an invasion would be much more successful, but at the current stage they just move their navy around randomly and sometimes shoot a shot here and there. They will almost never actually sink one of your ships.

I haven't played Panzer General, but since it's a tactical war game I can't imagine they would send their artillery as cannon fodder. The limitations are definitely not that "it's so complex that's it's almost impossible to make a better AI" as some seem to think.

Yeah, no, I agree about the AI. I was commenting on the trans-oceanic invasions that would happen on every single map where I had an island to myself (my preferred playing conditions--which I seldom get in Civ 5, by the way).
 
The thing I find disturbing is that Soren's blog had far more insight into Strategy game design than anything that come from Firaxis in the last 2 years....

Hence, I guess: CivRev, Colonisation 2 and Civ 5
 
Insightful remarks. I don't remeber reading that before. I like the last sentence and I think it explains why I don't look for a challenge in most computer games.

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.
I've seen the AI invade via sea, it does happen and this was pre 1.0.0.62. I was doing pretty much as Soren described and deliberately ignored military units on an archipelago map as I was under the impression the AI wasn’t fully capable of a sea invasion. I was only saved by the AIs woeful tactical usage of its units once they landed and the lack of bombard support from the naval vessels.
 
The quoted article reminds me why I have never been much of a fan of Mr. Johnson. He undoubtedly can express his opinion very well, but this doesn't change the fact that his opinion often is not right.

One game mechanic that pushed Chick over the edge with Empire: Total War was amphibious invasion. The AI was simply incapable of coordinating its land and naval forces together to launch a coherent and effective invasion of an overseas target. Smart players would quickly learn that if the AI could not attack amphibiously, then the strategic balance can be gamed easily. Maybe England’s troops are not such a threat after all?

This problem is not unusual; strategy games with transportation units almost always suffer from ineffective artificial intelligence. Coordinating land and naval units to be ready in the same place and at the same time – along with the necessary escort ships – is a non-trivial task.

Rise of Nations, Big Huge Games’s historical RTS, presented a blunt but effective solution to this problem; land forces which approach the shore simply turn into boats to carry themselves across the water. Once they reach their destination, the boats transform back into the original land units. No transportation ships ever needed to be built or managed at all.

With one simple stroke, Brian Reynolds, the game’s designer, removed a classic AI problem from the game, enabling water maps to remain interesting for veteran players. The design may have sacrificed the “realism” of requiring the player to build transport ships along with other naval units, but the upside was extending the game’s longevity significantly.

I think the conclusion (bolded) is just plain wrong.

I will go a bit into detail since this is one of the problems of Civ5,too.

By enabling this "embarking", water (lakes, oceans, even coastal hexes) loose their function: to be an unsurmountable obstacle. Literally, water becomes just another kind of terrain, similar to grassland, plains, hills and so forth.
Yet, the *units* change.

And that is a core problem in Civ5. The AI is just not aware of the fact that this special terrain "water" transfers its units into something completely different. The multi-promoted swordsman out of a sudden just becomes a boat. Slow, moving on a terrain without any shelter at all and completely weak to ranged attacks.

Because of this unawareness, the AI tends to often embark units to surpass some kind of obstacle (may it be a mountain, just the 'normal' traffic jam on its approach towards the target, whatever), thus making them prey for the human's ranged units.
The result being that the AI loses multiple hitpoints if not the complete unit, that way giving the human a big advantage.

The only way to solve this problem *and* keeping the game feature of "embarkment" would be to enable the AI to have some kind of understanding of "(battle) areas" and "escorting". The latter, as we all know, is completely unknown to the AI.
Therefore, the AI doesn't screen the area in which the embarked units move.

Now, under the assumption that these principles of "changed unit type", "(battle) area" and "escorting" could be solved, in other words that the AI could handle them, the problem of naval operations (especially invasions and landing operations) would be solved, too.

The point is, "embarking" doesn't solve the problem Mr. Johnson pointed out, but leads to the fact that the same problems as with naval operations are now observed in what otherwise would have remained to be just the normal fighting on land.

It was not a solution, it has increased the problem.
 
I would like to remind you of Panzer General II (1997). 1upt + competent AI. Certainly didn't need a powerful computer back then. :)
The AI in that game was mostly just defending fixed positions, while making small counterattacks and putting in replacements wherever it could. If you force it to switch roles with you, and play as the attacker, you can see just how bad the AI is.
 
Yes, definitely, but I think it would take a lot of work, and I'm afraid it would have to be done here by a bunch of civFanatics who know the game well. And unless the current code is seriously inefficient, a better AI would also require waiting for more powerful computers to arrive, since a more sophisticated combat AI would need a lot more computations between turns.
Based on observing the computer power available and the resulting game speed I assume that the current code *IS* seriously inefficient.

Yes, it doesn't seem like they realized how much work it would be to implement this properly.
And that is something which completely puzzles me.
Mr. Shafer has been reported to have played Panzer General, leading to the design decisions for Civ5.

Well, in Panzer General there are quite some chokepoints in the scenarios, and even an experienced human player has to heavily think about how to move his units through these chokepoints.

It speaks for quite some overestimation of one's own capabilities to think being able to transfer such principles into a civilization world.

1UPT shouldnt be too complicated for the AI.
It is too complicated.

With each movement point spend, the situation completely changes. It is not only about whether my swordsman is situated in front or behind my trebuchet, but also about whether it may be benefial to put the sword or the treb onto a hill, thus having to leave the road.
Does leaving a unit on the road block my reinforcement-horse from approaching the frontline?
And so on, and so on.

As said in my previous posting, the AI completely lacks the understanding of "battle theatres" and "escorting".
It has to be taught these principles first, and then the code even has to be optimized.
I don't expect this to happen within the next 10 years, actually.

I would like to remind you of Panzer General II (1997). 1upt + competent AI. Certainly didn't need a powerful computer back then. :)

Panzer General II is a completely different game.
You have minor and major victory fields (the cities, giving you prestige points when conquered), but apart from that you just have some terrain types with different specifics.
In Civ5, not only you have the different terrain, but it may have a different meaning. A horse pasture on a plains after all is still a plains.
Yet, if you don't block access to it, the opponent may still be able to build his mounted units.

Futhermore, in PG you don't have production at all. You are just living off of your initial prestige points (needed for reinforcements and "healing", which may be stocked up when conquering a victory field or a minor city.
In Civ5 you have to take the whole map into consideration.

It might be a bit different if the design decisions for Civ5 wouldn't have scrapped the experiences from PG:
the unit to tile ratio
retaliation fire
higher unit movement
less cluttered maps
 
MoM provides challenges with the various monster lairs and node guarding creatures -- essentially barbarians which are not part of the AI empires -- that often are more difficult, and thus more interesting/fun, than anything the AI will generally throw at you.
Agreed. Most of the battles in MoM are actually "player vs. environment", not "player vs. AI", and this contributes to making the game interesting and enjoyable despite its rather clueless AI. The "Heroes of Might and Magic" series chose that approach too.

By enabling this "embarking", water (lakes, oceans, even coastal hexes) loose their function: to be an unsurmountable obstacle. Literally, water becomes just another kind of terrain, similar to grassland, plains, hills and so forth.
Yet, the *units* change.

And that is a core problem in Civ5. The AI is just not aware of the fact that this special terrain "water" transfers its units into something completely different.
I agree. It actually surprised me that Soren chose embarkation as a positive example for simplifying a mechanic - after all, he was the one who introduced working naval invasions into the Civ franchise. It seemed as if he was dismissing his own work.

And that is something which completely puzzles me.
Mr. Shafer has been reported to have played Panzer General, leading to the design decisions for Civ5.

Well, in Panzer General there are quite some chokepoints in the scenarios, and even an experienced human player has to heavily think about how to move his units through these chokepoints.

It speaks for quite some overestimation of one's own capabilities to think being able to transfer such principles into a civilization world.
I still wonder whether the fact that (for the first time in a Civ game) the lead designer didn't also program the AI was a factor that exacerbated the problem. An experienced AI programmer should be able to see pretty quickly that writing a competent AI for a Civ game with a 1upt rule was going to be extremely tough. Although an experienced player without any programming experience at all should be able to see that as well, given how much trouble the Civ AIs traditionally had in coordinating movement.

Referring to my previous post, it almost seems that Civ5 was designed with a 90s approach: Design a game with interesting rules for the player, then cobble an AI together that understands the core rules well enough to pose a challenge (if given enough bonuses). And when they found out that the AI couldn't handle the task, it was too late. Whereas games like Civ4 and GalCiv2 where designed by their AI programmers, who took care that the game's rules wouldn't pose insurmountable problems for the AI.
 
It actually surprised me that Soren chose embarkation as a positive example for simplifying a mechanic - after all, he was the one who introduced working naval invasions into the Civ franchise. It seemed as if he was dismissing his own work.
I think it's just and example that makes clear to everybody (even the less expert) about the AI problems and some "easy" solutions to it.
I remember Soren's speech at Google about AI in games.
Even there he used this same example: the planning for oversea invasion with all the building of transports, escorts, embarking troops, scouting, finding the exact location for disembark (consider that for calculation the IA has to think not about the transport but about the mix of units actually transported).
It's all very complicated and a huge exception to the normal "rules".
The IA has to apply 2 different sets of "rules" to plan production and execute the invasion oversea.

With the "trick" of self-embarking units the AI has to deal with only one set of "rules", thus simplifying the AI work a lot.
The AI will use the same rulebook for both land and sea... much simpler for the AI to plan and execute sea invasion.
But if the AI is bad to start with (like in ciV with 1upt) then you don't get a huge improvement.

Panzer General II is a completely different game.
You have minor and major victory fields (the cities, giving you prestige points when conquered), but apart from that you just have some terrain types with different specifics.
In Civ5, not only you have the different terrain, but it may have a different meaning. A horse pasture on a plains after all is still a plains.
Yet, if you don't block access to it, the opponent may still be able to build his mounted units.

Futhermore, in PG you don't have production at all. You are just living off of your initial prestige points (needed for reinforcements and "healing", which may be stocked up when conquering a victory field or a minor city.
I completely agree with you.
The AI in PG and PG2 wasn't that good at all, but it had an easier job than its ciV counterpart.
PG is a tactical game with AI mostly placed on the defensive.
The very few scenarios where the AI was on the offensive it usually had overwhelming units superiority (and some extra bonuses in units' xp).

Movement in PG was extremely complex and the human player had to do some serious planning before moving the units.
Even a small mistake could ruin all the plan.

It's surprising that civV's designers didn't understand that taking those game concepts from PG would be hell in a more variable environment like ciV.
Or maybe they understood it, but didn't care: the target audience for the game wouldn't mind a simplification.
 
With one simple stroke, Brian Reynolds, the game’s designer, removed a classic AI problem from the game, enabling water maps to remain interesting for veteran players.

Brian Reynolds is responsible for Farmville and Frontierville.

Using him as an example of getting design right is not good for your argument.
 
"its not me, its you"

Despite computing ability massively increasing every few years, the ability to actually code a descent AI has been, I guess, approximately static.

AI coding needs to:
a) stand up for itself and become established as a respectable stand alone industry and career within Game development. (especially with dedicated core/threads now available)
b) actually visit the front running AI labs in the world a bit more often (and maybe get sponsored by them in return)
c) forerunning AI gaming developer specialists to create an identifiable brand around their own ability.

I guess I'm asking for more specialized AI specialists, and the above mentioned stuff won't happen until 2020.....
 
Brian Reynolds is responsible for Farmville and Frontierville.

Using him as an example of getting design right is not good for your argument.
And...? Don't you know that Farmville is the most popular game on Facebook? Not that I play it, but actual success trumps whatever you might think of a game.
 
Brian Reynolds is responsible for Farmville and Frontierville.

Using him as an example of getting design right is not good for your argument.

Ah well, succumbing to the dark side doesn't automatically invalidate all of one's past merits. ;)

Btw, if you're interested in Farmville's influence on future game design, I recommend to continue reading Soren's blog. He's clearly interested in the topic, and he manages to remain critical of the recent developments (even seeing them as a potential threat to game design) without condemning them outright. It's one of the reasons why I enjoy reading his blog; he tries to keep a broad perspective instead of narrowing down on ideology and information wars.
 
Brian Reynolds is responsible for Farmville and Frontierville.

Using him as an example of getting design right is not good for your argument.
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 :)
 
Top Bottom