Tweet
I just finished watching a good chunk of my first C3C game. The game isn’t won by the AI yet, but it’s in the very early modern era and I’ve seen enough to make a few points about observed changes.
AI worker Inefficiency.
This is based on the initial thread I started on the 1.21f patch PTW AI.
1.21f PTW AI thoughts
The problem “appear” to have been partially fixed. AI is no longer seen disbanding workers out in the open, but the AI will continue to have varying worker workforce sizes. From highs of 30 workers down to 4 workers, and back to 30 workers during the railroading frenzy and back down again.
A limitation of the study is that because it is hard to visually track workers as they move about, it is unclear what happens to the workers. It would appear that they are now JOINED into cities. Basically, the camera in-between turns no longer shows workers “dying” out in the open and worker numbers go up and down. This is the basic premise I’m working from.
Behaviour however continues to be inefficient in that large cities in the late game are often diverted to produce workers, instead of other needed improvements as demands in their services increase, for example, after pollution becomes a problem. This has three negative effects. First, the lost in population point can cause a losses in gold and research income especially in the core cities with little corruption. Secondly, the opportunity cost of lost shields in highly productive cities producing relatively cheap workers should not be discounted. Lastly, producing workers usually mean AI cities are not producing something else, like markets or other improvements that also impart benefits. The cost to the AI for playing like this is substantial.
Another issue is that the AI has been observed to disband captured workers. Captured workers are essentially free units since they have no upkeep costs and they should never have been disbanded. I have not noticed this behavior under PTW debug, so I will tentatively put it in the “new” issues category.
Although the two screenshots posted (below) do not reflect AI workers at their peak numbers, one can get an idea of the kind of see-saw variability in worker numbers. This occurs several times through the ages. 1) The initial growth stage will have the AI building many workers. 2) After this phase is over, worker numbers dwindle during the middle ages. 3) As railroads become available AI again builds a large army of workers. 4) This number then declines and increases again as pollution starts to become a problem. The Sumerian industrial era screen capture is taken as the workers are increasing to deal with increased pollution.
Note its captured worker population goes from > 3 (its actually 8 captured workers) to zero through this period. This is simply sloppy and inefficient play and must be corrected.
Solutions: From observation, there appears to be a range in which the AI will try to maintain workers beyond which additional workers are disbanded and below which new workers are produced. A solution would be to narrow this range, so that the variability observed or swinging from 30 workers down to 4, and then up to 20 can be reduced. A sustainable permanent worker force is key.
Related to this solution is to give AI and absolute rule when it comes to captured units. By simply not disbanding them, AI will tend to have a larger permanent workforce at its disposal, allowing it to build less during peak demand times. Coupled with a narrower range of desired worker variability, this should hopefully reduce practices of disbanding and rebuilding workers by the AI through the ages.
Sumeria 1320-Middle Ages
AI worker Inefficiency.
This is based on the initial thread I started on the 1.21f patch PTW AI.
1.21f PTW AI thoughts
The problem “appear” to have been partially fixed. AI is no longer seen disbanding workers out in the open, but the AI will continue to have varying worker workforce sizes. From highs of 30 workers down to 4 workers, and back to 30 workers during the railroading frenzy and back down again.
A limitation of the study is that because it is hard to visually track workers as they move about, it is unclear what happens to the workers. It would appear that they are now JOINED into cities. Basically, the camera in-between turns no longer shows workers “dying” out in the open and worker numbers go up and down. This is the basic premise I’m working from.
Behaviour however continues to be inefficient in that large cities in the late game are often diverted to produce workers, instead of other needed improvements as demands in their services increase, for example, after pollution becomes a problem. This has three negative effects. First, the lost in population point can cause a losses in gold and research income especially in the core cities with little corruption. Secondly, the opportunity cost of lost shields in highly productive cities producing relatively cheap workers should not be discounted. Lastly, producing workers usually mean AI cities are not producing something else, like markets or other improvements that also impart benefits. The cost to the AI for playing like this is substantial.
Another issue is that the AI has been observed to disband captured workers. Captured workers are essentially free units since they have no upkeep costs and they should never have been disbanded. I have not noticed this behavior under PTW debug, so I will tentatively put it in the “new” issues category.
Although the two screenshots posted (below) do not reflect AI workers at their peak numbers, one can get an idea of the kind of see-saw variability in worker numbers. This occurs several times through the ages. 1) The initial growth stage will have the AI building many workers. 2) After this phase is over, worker numbers dwindle during the middle ages. 3) As railroads become available AI again builds a large army of workers. 4) This number then declines and increases again as pollution starts to become a problem. The Sumerian industrial era screen capture is taken as the workers are increasing to deal with increased pollution.
Note its captured worker population goes from > 3 (its actually 8 captured workers) to zero through this period. This is simply sloppy and inefficient play and must be corrected.
Solutions: From observation, there appears to be a range in which the AI will try to maintain workers beyond which additional workers are disbanded and below which new workers are produced. A solution would be to narrow this range, so that the variability observed or swinging from 30 workers down to 4, and then up to 20 can be reduced. A sustainable permanent worker force is key.
Related to this solution is to give AI and absolute rule when it comes to captured units. By simply not disbanding them, AI will tend to have a larger permanent workforce at its disposal, allowing it to build less during peak demand times. Coupled with a narrower range of desired worker variability, this should hopefully reduce practices of disbanding and rebuilding workers by the AI through the ages.
Sumeria 1320-Middle Ages
Attached Files
Comment