v0.04
last updated: Feb 06, 2005
Preamble
--------------
Borrowed from Sirian's post in the thread below
Improving the AI isn't just about making it more competitive. There is also making it behave more in line with the game's context.
AI is part of the experience. It's not just about winning... The AI impacts the "atmosphere". The more life the game maker can breathe into his or her design, the more engaging the game could be...The difficulty levels exist to hand the AI a handicap so it can continue to provide a fun experience even ...THE AI IMPACTS EVERYBODY. It's the partner, the opponent, the rival, the animation of gameplay.
Guiding Principles
--------------
1) No going back -The AI in Civ 4 should not regress back to a pre Civ 3 level.
2) The AI should at least be at good (in terms of overall performance) as the Civ 3 AI
3) Superficial Intelligence (see point 4 under Game Wide issues)
Game wide issues
----------------
1) Every game feature available to the human player should be available to the AI. And the AI should reasonably be able to use it (please, no SMAC style gameplay where there are lots of fun features that the AI can't handle) or more recently, C3C style issues such as lack of army usage, artillery usage etc.
2) Cheating to a minimum. See guiding principle #1. Civ3 AI cheats somewhat in that it knows locations of resources and your units. This is to help keep it reasonable competitive. Civ 4, if the AI is to cheat, should cheat in similar fashions. That is, cheats that are essentially heuristic shortcuts that grant AI some competitiveness in doing things that may simply be too abstract to program into an AI routine (clearly, settling on a piece of land with the expectation there may be a resource is probably a concept the AI can't grasp)
3) AI priority in expansion - Assuming the border/culture system returns, and I suspect it will (maybe it won't be called culture) then the AI should expand much more in a human fashion. That is, the top priorities would be to expand in areas that give synergy to its current empire configuration.
All to often, Civ3 AI expands (correctly) for the sake of expansion. There is a problem however. This sometimes leaves them open to sprawling swiss cheeze empires where one civilizaiton have several isolated pockets of cities. That's not to say such configurations is neccessarily wrong. But the AI doesn't distinguish been conquests that adds to its borders and extends its core outwards vs. taking cities half a continent away. To the AI, it would appear all cities are equal regardless of distance and 'fit' to its current empire configuration.
4) Intelligence. or at least the ability emulate what a human might do. I was so impressed with the way Sirian described his ideal AI (see Preamble) that I think he puts perfectly the little things myself and many others have wanted but never quite managed to put into words.
5) Forecasting. Use past Data to forecast and allow the AI to see trends and take action. Most basic flaw currently in the Civ3 AI is the AI superpower's inability to see threats coming from a middle of the pack human player who is slowly gathering in strength and knocking off rival after rival.
Human players play with a keyboard and mouse and can thus very easily see visually a lot of information, such as a rival's growth. The AI however remain oblivious and beat on each other senseless.
The idea is for the AI to recognize threats as they emerge. Once recognition is made, the difficulty setting may enter. On Chieftain, there may be a grace period before the AI will act on this threat on higher difficulty levels the AI will have less telorance for waiting. As always, this shouldn't just be geared to scoping out the human player (as this would break guiding principle #1 on regressing back to Civ2 level AI of ganging up). Giving the AI the ability to check up on all their rivals is integral part of the request.
Ultimately, no one wants a 'Big Blue' type AI. It's probably impratical. What would be interesting would be to throw in more layers of humanity into AI actions, in how it manages diplomacy, trade, war, peace and alliances. The Civ3 AI still feels very much a cold crunching machine. A good example is troop management. Rather than sending stacks of forces to be slaughtered by a human bombardment gauntlet or predictably pulling troops or ships that have been damaged, the AI should be able to asses the situation more broadly and potentially still send in damaged troops "sometimes" based on a broad set of variables that the developers can keep hidden. This would remove much of the predictability of an AI and introduce a much more human set of reasoning and introduce challenge not by being a superior tactician, but by simply being less predictable.
Unit Specific Issues
------------------
1) The Naval Factor - I was very impressed with the modifications to AI behavior vis-a-vis navies in C3C. However it still was very much limited by a date AI engine and had severe limitations... like the inability to hold on to beach heads. . The biggest problem right now is, once the massive landing happens, and the first few cities are taken, AI forces quickly fans out. Failing to secure their beachheads. Making it all too easy (even for the AI victim) to retake the beachheads and you end up with 20 turns of fighting over the same peice of land.
What I'd like to see is a much more intelligent AI that can distinguish between troops landing on beach heads and vs. troops moving over land. Perhaps if it decides to make amphibous landings, the game could mark the tiles it has decided to land on with a special flag that would indicate ' Amphibious Landing Here' this would in turn trigger a subsystem of AI would would take care of resupplying forces, reinforcing forces and fortifying existing forces at that location.
Governor Issues
-----------------
- City needs communicated to workers. I mean that if a city is drowning in food, the workers should know to mine more. If the city has lots of mountains/hills, the workers should know to irrigate around that city. Definitely non-trivial but feels like straight-forward calculations that the computer should be good at. (comment: I recall Soren's interview on Apolyton back in 2001 where he describes the AI being multilayered. A Leader AI for empire wide management, City AI for city management and Unit AI for unit management so something like this may already be implemented)
- City placement. They seemed to have this down in PTW and regressed in C3C. Hard algorithm to write, I know, but it should be better.
last updated: Feb 06, 2005
Preamble
--------------
Borrowed from Sirian's post in the thread below
Improving the AI isn't just about making it more competitive. There is also making it behave more in line with the game's context.
AI is part of the experience. It's not just about winning... The AI impacts the "atmosphere". The more life the game maker can breathe into his or her design, the more engaging the game could be...The difficulty levels exist to hand the AI a handicap so it can continue to provide a fun experience even ...THE AI IMPACTS EVERYBODY. It's the partner, the opponent, the rival, the animation of gameplay.
Guiding Principles
--------------
1) No going back -The AI in Civ 4 should not regress back to a pre Civ 3 level.
2) The AI should at least be at good (in terms of overall performance) as the Civ 3 AI
3) Superficial Intelligence (see point 4 under Game Wide issues)
Game wide issues
----------------
1) Every game feature available to the human player should be available to the AI. And the AI should reasonably be able to use it (please, no SMAC style gameplay where there are lots of fun features that the AI can't handle) or more recently, C3C style issues such as lack of army usage, artillery usage etc.
2) Cheating to a minimum. See guiding principle #1. Civ3 AI cheats somewhat in that it knows locations of resources and your units. This is to help keep it reasonable competitive. Civ 4, if the AI is to cheat, should cheat in similar fashions. That is, cheats that are essentially heuristic shortcuts that grant AI some competitiveness in doing things that may simply be too abstract to program into an AI routine (clearly, settling on a piece of land with the expectation there may be a resource is probably a concept the AI can't grasp)
3) AI priority in expansion - Assuming the border/culture system returns, and I suspect it will (maybe it won't be called culture) then the AI should expand much more in a human fashion. That is, the top priorities would be to expand in areas that give synergy to its current empire configuration.
All to often, Civ3 AI expands (correctly) for the sake of expansion. There is a problem however. This sometimes leaves them open to sprawling swiss cheeze empires where one civilizaiton have several isolated pockets of cities. That's not to say such configurations is neccessarily wrong. But the AI doesn't distinguish been conquests that adds to its borders and extends its core outwards vs. taking cities half a continent away. To the AI, it would appear all cities are equal regardless of distance and 'fit' to its current empire configuration.
4) Intelligence. or at least the ability emulate what a human might do. I was so impressed with the way Sirian described his ideal AI (see Preamble) that I think he puts perfectly the little things myself and many others have wanted but never quite managed to put into words.
5) Forecasting. Use past Data to forecast and allow the AI to see trends and take action. Most basic flaw currently in the Civ3 AI is the AI superpower's inability to see threats coming from a middle of the pack human player who is slowly gathering in strength and knocking off rival after rival.
Human players play with a keyboard and mouse and can thus very easily see visually a lot of information, such as a rival's growth. The AI however remain oblivious and beat on each other senseless.
The idea is for the AI to recognize threats as they emerge. Once recognition is made, the difficulty setting may enter. On Chieftain, there may be a grace period before the AI will act on this threat on higher difficulty levels the AI will have less telorance for waiting. As always, this shouldn't just be geared to scoping out the human player (as this would break guiding principle #1 on regressing back to Civ2 level AI of ganging up). Giving the AI the ability to check up on all their rivals is integral part of the request.
Ultimately, no one wants a 'Big Blue' type AI. It's probably impratical. What would be interesting would be to throw in more layers of humanity into AI actions, in how it manages diplomacy, trade, war, peace and alliances. The Civ3 AI still feels very much a cold crunching machine. A good example is troop management. Rather than sending stacks of forces to be slaughtered by a human bombardment gauntlet or predictably pulling troops or ships that have been damaged, the AI should be able to asses the situation more broadly and potentially still send in damaged troops "sometimes" based on a broad set of variables that the developers can keep hidden. This would remove much of the predictability of an AI and introduce a much more human set of reasoning and introduce challenge not by being a superior tactician, but by simply being less predictable.
Unit Specific Issues
------------------
1) The Naval Factor - I was very impressed with the modifications to AI behavior vis-a-vis navies in C3C. However it still was very much limited by a date AI engine and had severe limitations... like the inability to hold on to beach heads. . The biggest problem right now is, once the massive landing happens, and the first few cities are taken, AI forces quickly fans out. Failing to secure their beachheads. Making it all too easy (even for the AI victim) to retake the beachheads and you end up with 20 turns of fighting over the same peice of land.
What I'd like to see is a much more intelligent AI that can distinguish between troops landing on beach heads and vs. troops moving over land. Perhaps if it decides to make amphibous landings, the game could mark the tiles it has decided to land on with a special flag that would indicate ' Amphibious Landing Here' this would in turn trigger a subsystem of AI would would take care of resupplying forces, reinforcing forces and fortifying existing forces at that location.
Governor Issues
-----------------
- City needs communicated to workers. I mean that if a city is drowning in food, the workers should know to mine more. If the city has lots of mountains/hills, the workers should know to irrigate around that city. Definitely non-trivial but feels like straight-forward calculations that the computer should be good at. (comment: I recall Soren's interview on Apolyton back in 2001 where he describes the AI being multilayered. A Leader AI for empire wide management, City AI for city management and Unit AI for unit management so something like this may already be implemented)
- City placement. They seemed to have this down in PTW and regressed in C3C. Hard algorithm to write, I know, but it should be better.
