Darryl, if you get time, write me up a brief description of the major improvements to AI behavior that you have been able to achieve, and what settings you altered to achieve them.
Wes

Well, I just recently implemented them but so far I have seen the ai attack in stacks of 9 for air and naval units. To do this I set the priority for attack_troops_priority to the highest priority for all the aip files. It seems when other priority's took precedence that the ai would stack very ineffectively, for example you might see a stack of two or three for air and naval units when the explore or seige priority took precedence. Since I set the attack_troops_priority the highest in all aip files it takes cities a lot more effectively and more often.
Also, it seems that adjusting the min_attack ratio and max_attack_ratio to from 9.0 and down(i have 14 different aips for the war settings so I decrease this in each file) tends to have very good results for making the ai more inteligent about where to attack and when. Although I have tested this it is only the attack_troops_priority that seems to really get the ai to stack naval units in 9 a stack.
Also the distance_from_unit as well as the other distance priorities seems to have been interfering with making the ai a tougher opponent. Setting all these priorities to 0 has made the ai attack from farther distances and also made the ai less choosy about the cities it takes. For instance the per_capita modifier would make the ai not take easily taken cities, instead waiting for what appears to be a larger city to come into play. If that larger city doesn't come into play the ai seems pretty stupid and will just pass by or wait??? Crazy, but setting it to 0 has made the ai much more aggressive. I especially enjoyed it when I saw the Japanese attack the US mainland (seeing as my map is 120x240 at first it would only attack it's closest enemies), sending troop transport after transport to the mainland. I have also seen it use the aircraft carriers(but not often, still working on that).There still are some quirky ai behavioral problems going on but hopefully fixable.
Also I have set the num_city_defenders to different integers depending on the aip warsetting. I have randomized the aip files to switch on a random basis. I have created 14 seperate aip files just based on different styles of war to conduct and how many city defenders the ai will prioritize. Anyway, those are the major changes I can think of, sorry about some of the spellings for the priorities, I'm at work right now and I am going by memory.

Darryl states in other letters that increasing the max_eval numbers seems to get the AI to attack more, perhaps because it evaluates more cities and thus finds more opportunities for success.