Hello,
I am amazed how effectively the AI plays in the initial stage. But in my opinion later it makes stupid mistakes (in 2.1.8) and some of them should not be so hard to fix:
1. AI does not disband even VERY obsolete units
2. Even when intensively attacked by bombers for some time, AI often does not start to build fighters. Also, even when bothered by my ships, it usually does not build submarines although they are an excellent defender of coastal waters. The thing with submarines may perhaps be caused by the fact that they were too powerful so I decreased their parameters - but either case, it would be good to include (into the AI program) that naval units with low defense are to be used as coastal defenders, who leave cities only when they see a tempting target nearby and very seldom otherwise.
3. I have a feeling that AI does not deliberately escort full transports with warships
4. AI often keeps insufficient number of ground units in cities (0 or 1).
5. I think AI would wage war more effectively if it sometimes decided to build a big army instead of attacking with a big number of raids with very few units in each raid (this applies both to ground attack and to naval invasions to other continent). In my opinion AI makes these massive attacks far too rarely. A related topic is that AI makes invasions in my islands even when it is obvious that if AI is lucky and aptures my city or two, I will soon recapture it. Such strategy has only sense if AI wants my tech, but then at least one third of invasion forces should be diplomats. If AI were reprogrammed to send mass attacks more often, the game agains AI would be far more thrilling.
Further, the AI does not show even the smallest effort to protect unloaded Cavalry with Alpinists or to prefer a rough terrain for unloading the ships. Sometimes it keeps unloading on the same part of my island again and again, which makes the defensive measures easier.
6. I suspect that AI builds such improvements as "Port fascility" or "Airport" simply because it feels that it is time to build some improvements - while the right strategy is to build these maintenance-expensive buildings only if it intends to produce several ships/aircrafts in that city in the next turns. And such buildings should be be built only in very productive cities. More generally, I have some suspicion that AI does not weight the benefit of each building for this specific city against its build and maintenance cost.
7. (Here I am less certain if the AI does that badly but I suspect so.) AI should conquer a city with city walls in an inteligent way. That means it should attack me either if it thinks that it has enough units to conquer or after sending hordes of diplomats to destroy the Walls. The AI behaviour should reflect the habits of experienced players - for example I usually either investigate the city to see if the number of defenders is low, or sometimes I starve the city by occuping the surrounding squares. Or I attack with reasonable number of Ironclads.
8. The AI would be more dangerous if it traded knowledge more often with other AI.
9. The AI strongly underestimates the importance of terrain changes (both normal things like mining or roads and transforming mountains to hills etc.).
10. The AI seriously underestimates the importance of building Wonders and it does not realize that some wonders have much greater ration effect/cost than the others. Especially the Great Library, also Michalangelo, Adam Smith, Leonardo and Hoover. After correcting this, AI would be much more challenging.
11. The AI puts very little effort to make progress in science.
12.I believe points 10 and 11 are caused by the fact that AI builds many units even when a peace development would be a better choice. A big improvement in AI could be achieved if AI were programmed to distinguish between the following statuses:
- Total emergency - build nothing but military units
- Intensive war - build almost nothing but military units
- Peace status - build only neccessary garrisons and throw all other resources to buildings, science, terrain changes, wonders and expansion. (I wrote more about this at Forum:Improvements in AI strategy).
- Almost peace status - someone declared war to you, but you want to develop peacefully (because you know that at this stage, capturing enemy cities, if at all possible, has much worse ratio benefit/costs than improving your existing cities). Therefore, do not send your units to enemy cities. Use units only to defend your territory and put all resources that you can spare into a peace development.
By this, I do not say that there should be exactly these four statuses. But I think that AI builds many units in cases where peace development would be a far better choice.
Pavel Jelínek
--Reflame 17:59, 20 April 2009 (UTC)