Shannon V. OKeets
From: Honolulu, Hawaii
If anyone here has played the Strategic command series from the Battlefront gaming company please let me know.
When the German AI in Strategic command attacked Russia after destroying Poland and France you had to play your best game without mistakes to stop them. I could not until I found that one famous stop gap that hurts all computer AI's. A trick within a trick.
While Germany was defeating France I [Russia] built a string of forts along rivers in front of Moscow from Rostov in the south to Vologda in the north, by the time the Panzers reached the line of forts I had been able to put a large amount of artillery just behind the forts. For the first time in the game the AI struggled with the situation. Attacking piece meal instead of concentrating in one crucial area with a lot of its units.
I personally thought the AI was very adequate for a game constructed five years ago. Their Pacific game fell short with the AI IMO. I found that given enough power the AI could be very strong even against better than average players.
This sounds like Strategic Command Global Conflict, for which I've only played the demo. The Original SC has a fairly limited AI, which is hard-coded with certain grand strategies. The ability of the AI varies according to the level of the game - tactically, it can be quite lethal; operationally it can be competent; strategically, rather inept. It constantly wastes build points moving units back and forth via SR. It doesn't really know how to handle unusual situations, such as an Allied invasion of Italy in 1939/40. It throws away the German u-boats early by keeping them in range of British airpower.
Some of us are old enough to remember the original SSG (Strategic Studies Group) that published some very good early computer games, such as Reach for the Stars and Carriers at War. Roger Keating, the designer known for his excellent AI's, made the very important point that most game AI's fail because they are tacked on at the end. Per Keating, proper game design requires designing the AI first. This serves to guide the designer, so the game code takes the AI into account from day one. If the designer realizes that he needs several new variables for the AI only after the code is 90% complete, that's too late. In designing the AI first, the game designer also learns what features should be excluded because they can't be properly handled by an AI. Keating also made the point that converting an existing boardgame to computer is the most difficult situation for a game designer - not only are the game rules already written, but they were originally designed for humans, not machines.
I want MWIF to have an AI. I want it have a competent AI that doesn't cheat. I'm keeping my expectations low not because I believe Steve and others won't put in the necessary effort (they will), but because I recognize that designing this game's AI will be incredibly difficult. I also expect, like the original SC, the AI will be better at the lower tactical levels than the higher strategic levels.
I expect the reverse.
WIF has some phases that are clearly strategic (e.g., production planning and production). Programming those won't be much of a problem because the search space is relatively small. Likewise decisions in the DOW phase are pretty straight forward, assuming that strategic plans have been defined for that purpose - just select one and stay with it unless something drastic happens.
Operational decisions could be difficult to make (e.g., where to send reinforcements, how to allocate air and naval units around the world and within a theater.
But tactical decisions can be very difficult since they are mostly related to the hex grid and sea areas. Geography is difficult for an AI Opponent. Players glance at a map and know where threats and opportunities exist (assuming the player is any good at the game). But for an AIO, examining the map is tedious and potentially a source of errors. Tactical decisions also involve sequencing actions. While the AIO can do an exhaustive search, elapsed time is a consideration. Again, the hex grid opens up an enormous range of possibilities. Comparing WIF to chess, chess is a closed ended search space with few moves available to either side. WIF has hundreds of units on the map and a world of hexes to explore. For example, there are 64 squares on a chessboard, and 83 sea areas in MWIF.
Perfection is an elusive goal.