Tweet
Well, I’ve been noticing that because the AI tends to differentiate between its marked OFFENSIVE and unmarked (defensive) units, it doesn’t “gamble” or take risks properly.
1)Now, I understand that in the human context, AI has no concept of risk and infact, may appear to take very risky actions, such as attacking with 2 hp units when a normal human player would send them back to heal.
What I am saying by risk is more closely linked to its larger scale leader AI strategizing. From watching my current debug game on PTW as well as my experience playing against it, the AI is far too often very conservative when it comes to defense of its own territory.
It usually has X offensive units spread around its territory and Y defensive units, spread roughly evenly in each city. 3 to 4 for most cities in a mid sized empire. 6 to 8 for their capitals.
When a rival Civ attacks, what usually happens is, the offensive units, escorted by 1 or 2 defensive units move out. When this force is exhausted, and a good human player can pretty much do it very quickly if they have force and technological superiority, they become sitting ducks.The AI just hunkers down. Each city for themselves.
There is a good reason for this. I think Soren did this to ensure the AI isn't caught unprepared from a sneak attack in the back. The AI is essentially hedging its bets other attacks are coming. But this is not the most efficient use of a defensive force. Human players routinely move units from its "safe" cities to deal with problems in the front. The AI is even programmed to attack by sea, since I'm sure Soren has predicted humans would do exactly this.
In the PTW debug game I am currently watching where Japan was destroyed by 3 Civs and I got to watch a lot of city attacks, I see a lot of sustained cavalry charges (usually 8 attacks per turn) against a city of say 3 infantry and maybe 2 rifleman conscripts. While these cities survive the initial assault, after 2 or 3 turns of sustained attacks, they fall pretty easily.
Japan could have defended several of its cities, (and it had all its cities hooked up by RR) effectively had it shifted some of its surplus defenders from further in their empire. Instead, it just depended on the unit production from each city and conscription to bolster the defenders of the front line cities.
What does this have to do with risk taking?
The AI is unable to quickly shift its forces, like a humans would, and take a gamble that if the enemy forces show show up form behind, it can quickly shift back.
The key issue here is AI passivity to defending against offensive stacks it can easily overcome, and this is especially true with 3 movement units. The AI seems to not notice if you place units 3 tiles away from its cities, giving 3 movement attackers a great advantage. 16 cavalryman vs. 3 infantry is scary. But move a few infantryman from the inner core cities and suddenly you can have 16 vs 6, or 16 v 8 etc. The odds shift in the Defender’s favour.
2)Another problem is that the AI sticks too closely to its unit tags. Offensive units attack almost exclusively (they defend in captured cities) and untagged units defend almost exclusively (I’ve seen the occasional attack to finish off a straggler, but it is rare).
I understand the need for this system for the sake of making it easy on the AI and our computers when it comes to unit management, but on some level, the AI has to recognize that a non offensive unit can be used to do an offensive unit’s job, and they should use it more frequently. This is especially important for Civs like Japan, where their Samurai with A=4 D=4 confuses the AI. In the dubug game I was watching, the AI built a good 10 or so Samurai in its capital. All of the 10 were untagged so they sat in the city, while their army was fighting a war.
3) A third comment I have is the inefficient use of clearly offensive units. Again, referring to the debug games I've watched and from my experiences, it is not uncommon to whittle down a city’s defenders only to find it had a lone cavalry sitting inside…clearly an offensive unit. So why didn't it attack earlier. why was it sitting inside doing nothing?
Against a stack, the AI should recognize and be able to do the “bounce” trick human players regularly use. That is, they can send a unit with superior attack points against a stack of units with weaker Defense points and not worry too much about losing the unit. After attacking, the unit will bounce back to the city because the stack is there and it can’t occupy the square. I’ve used this strategy to weaken a great many AI stacks, and it’s too bad the AI doesn’t take advantage of this.
Putting it all together
A related note here is that the AI shouldn’t blow its load immediately when war is declared. Offensive units, especially the mobile ones, should hang back in their frontline cities so they can do what I just described above. It would be much more challenging for a player if they march up their stack to an enemy city only to find their stacks are being mowed down by offensive defenders who popped out from a nearby city and popped back in to hide. Not to be unfair here, the AI does do the bounce trick on occasion, but it seems more like a chance occurrance than a realization of this particular strategy. This will also make it doubly hard for human players to quickly whittle down the AI forces in the open phase of a war. All to often, the AI will have all of its offensive and good units out in the open on their way into enemy territory. The human response to these is usually defensive. They hunker down and use the bounce trick. They wait until a stack is close enough to one of their cities, move in mobile attackers, and wipe out the said stack. 5 or 10 turns later, all attacking stacks are wiped out, AI is on the defensive, and the humans move in with their large stacks of offensive units.
My first comment ties in to my third. When the human players attack with their big ass stacks, the AI sit passively, barely reinforcing their frontline cities even when they can afford to and there is a good chance a vigorous defense will weaken the assault. What essentially happens is Each city becomes an isolated island that can be divided and conquered. There is no effective joint defense here.
And since human players can see and count the number of enemy units within their range of sight, the AI should know this and react to this. If extra code need to be written to knock them over the head with it, I think it is worthwhile, even if the AI has to somehow cheat a little to get to this conclusion. What I think is sorely needed is for the AI to see stacks, recognize the size and type of the force it is facing and adjust the forces in the frontlines to deal with it.
I have some more comments regarding potential human exploits if a system like this is implement and possible counters to this. I’ll write it and post this later today. It’s 2:30 am and I need to finish my homework.
Thanks for reading another long rant from me.
June 1 2003 ADDENDUM:
SOME of what Sirian has to say about loopholes after 1.16f still applies to the game post PTW 1.21f.
Comments on 1.17f still applicable today
1)Now, I understand that in the human context, AI has no concept of risk and infact, may appear to take very risky actions, such as attacking with 2 hp units when a normal human player would send them back to heal.
What I am saying by risk is more closely linked to its larger scale leader AI strategizing. From watching my current debug game on PTW as well as my experience playing against it, the AI is far too often very conservative when it comes to defense of its own territory.
It usually has X offensive units spread around its territory and Y defensive units, spread roughly evenly in each city. 3 to 4 for most cities in a mid sized empire. 6 to 8 for their capitals.
When a rival Civ attacks, what usually happens is, the offensive units, escorted by 1 or 2 defensive units move out. When this force is exhausted, and a good human player can pretty much do it very quickly if they have force and technological superiority, they become sitting ducks.The AI just hunkers down. Each city for themselves.
There is a good reason for this. I think Soren did this to ensure the AI isn't caught unprepared from a sneak attack in the back. The AI is essentially hedging its bets other attacks are coming. But this is not the most efficient use of a defensive force. Human players routinely move units from its "safe" cities to deal with problems in the front. The AI is even programmed to attack by sea, since I'm sure Soren has predicted humans would do exactly this.
In the PTW debug game I am currently watching where Japan was destroyed by 3 Civs and I got to watch a lot of city attacks, I see a lot of sustained cavalry charges (usually 8 attacks per turn) against a city of say 3 infantry and maybe 2 rifleman conscripts. While these cities survive the initial assault, after 2 or 3 turns of sustained attacks, they fall pretty easily.
Japan could have defended several of its cities, (and it had all its cities hooked up by RR) effectively had it shifted some of its surplus defenders from further in their empire. Instead, it just depended on the unit production from each city and conscription to bolster the defenders of the front line cities.
What does this have to do with risk taking?
The AI is unable to quickly shift its forces, like a humans would, and take a gamble that if the enemy forces show show up form behind, it can quickly shift back.
The key issue here is AI passivity to defending against offensive stacks it can easily overcome, and this is especially true with 3 movement units. The AI seems to not notice if you place units 3 tiles away from its cities, giving 3 movement attackers a great advantage. 16 cavalryman vs. 3 infantry is scary. But move a few infantryman from the inner core cities and suddenly you can have 16 vs 6, or 16 v 8 etc. The odds shift in the Defender’s favour.
2)Another problem is that the AI sticks too closely to its unit tags. Offensive units attack almost exclusively (they defend in captured cities) and untagged units defend almost exclusively (I’ve seen the occasional attack to finish off a straggler, but it is rare).
I understand the need for this system for the sake of making it easy on the AI and our computers when it comes to unit management, but on some level, the AI has to recognize that a non offensive unit can be used to do an offensive unit’s job, and they should use it more frequently. This is especially important for Civs like Japan, where their Samurai with A=4 D=4 confuses the AI. In the dubug game I was watching, the AI built a good 10 or so Samurai in its capital. All of the 10 were untagged so they sat in the city, while their army was fighting a war.
3) A third comment I have is the inefficient use of clearly offensive units. Again, referring to the debug games I've watched and from my experiences, it is not uncommon to whittle down a city’s defenders only to find it had a lone cavalry sitting inside…clearly an offensive unit. So why didn't it attack earlier. why was it sitting inside doing nothing?
Against a stack, the AI should recognize and be able to do the “bounce” trick human players regularly use. That is, they can send a unit with superior attack points against a stack of units with weaker Defense points and not worry too much about losing the unit. After attacking, the unit will bounce back to the city because the stack is there and it can’t occupy the square. I’ve used this strategy to weaken a great many AI stacks, and it’s too bad the AI doesn’t take advantage of this.
Putting it all together
A related note here is that the AI shouldn’t blow its load immediately when war is declared. Offensive units, especially the mobile ones, should hang back in their frontline cities so they can do what I just described above. It would be much more challenging for a player if they march up their stack to an enemy city only to find their stacks are being mowed down by offensive defenders who popped out from a nearby city and popped back in to hide. Not to be unfair here, the AI does do the bounce trick on occasion, but it seems more like a chance occurrance than a realization of this particular strategy. This will also make it doubly hard for human players to quickly whittle down the AI forces in the open phase of a war. All to often, the AI will have all of its offensive and good units out in the open on their way into enemy territory. The human response to these is usually defensive. They hunker down and use the bounce trick. They wait until a stack is close enough to one of their cities, move in mobile attackers, and wipe out the said stack. 5 or 10 turns later, all attacking stacks are wiped out, AI is on the defensive, and the humans move in with their large stacks of offensive units.
My first comment ties in to my third. When the human players attack with their big ass stacks, the AI sit passively, barely reinforcing their frontline cities even when they can afford to and there is a good chance a vigorous defense will weaken the assault. What essentially happens is Each city becomes an isolated island that can be divided and conquered. There is no effective joint defense here.
And since human players can see and count the number of enemy units within their range of sight, the AI should know this and react to this. If extra code need to be written to knock them over the head with it, I think it is worthwhile, even if the AI has to somehow cheat a little to get to this conclusion. What I think is sorely needed is for the AI to see stacks, recognize the size and type of the force it is facing and adjust the forces in the frontlines to deal with it.
I have some more comments regarding potential human exploits if a system like this is implement and possible counters to this. I’ll write it and post this later today. It’s 2:30 am and I need to finish my homework.
Thanks for reading another long rant from me.
June 1 2003 ADDENDUM:
SOME of what Sirian has to say about loopholes after 1.16f still applies to the game post PTW 1.21f.
Shouldn't paratroops be allowed to operate (like other air units) out of any city? That would put them on the map. Shouldn't the AI be trained to at least the option of a slightly more sophisticated attack routine than just "attacker/defender pairs" sent beelining willy nilly toward the nearest targets? Shouldn't there a be SOME point at which the AI's, in their defense routines, opt to use attacker-type units on defense? Some knights defending in cities would sure slow down the horsie rush in the medieval age -- or should the whole always-retreat-safely option for fast vs slow units be re-evaluated? It is awfully strong. I don't find these last couple of issues to be loopholes, but something about them is out of whack. 
The game balance in the ancient era was just fine. Closing the poprush loophole is all that was needed to correct the ability of some players to dominate in that era. Where the game breaks down is at railroads and nationalism, where the AI's are unable to use artillery effectively, to use rails effectively, and to cope with the player's ability to use rails on both attack and defense. If railroads were ELIMINATED from the game, the AI's would perform a whole lot better. But then, they also overdraft and overwhip their people, to get more units for warfare, which are then usually just wasted on stupid offensive attacks, leaving cities virtually undefended. So games of Civ3 turn entirely at the start of the Industrial era. On lower difficulty levels, a player can have fun in the ancient and midieval times, then the game's over at industrial as the player pulls ahead. Or else on higher difficulty, the player scratches and claws his way up out of a deep hole, and if he can survive to the industrial era, can then gain the means to compete.
Comment