AI Scenarios - concept and evolution

[el cid again]

RHS - which is a mod for pbem use - came up with the idea of scenarios for AI players.

The first idea was to do a Japan Enhansed Scenario (a term used theoretically by Joe Wilkerson years ago)
which - because Japan was stronger - might give AI more to play with - and therefore be more of a challenge.
While this no doubt is a valid idea - the more AI has the tougher it is - it did not address the problems AI has
with WITP in general nor with RHS modifications in particular. So while the scenario first tried - EOS - worked -
it failed as an AI scenario. AI was worse than in any other mod or in stock - because it was bewildered by interior
river systems and other RHS features. Similarly, it could not deal with the idea of Active Russians not at war.

The next step was to modify EOS to get rid of things that AI could not address. In particular, it went back to stock
for rivers - there are simply no interior river systems or ports or units on them - and the Russians are passive - just
like stock. This worked better - but not well enough - and AI still wastes too much effort on silly things - and does not
do things that need to be done much of the time.

The next step was to program the units to "tell" AI what to do with them - assign missions, targets, etc as appropriate.
This kind of works too - and kind of not: it works for SOME missions on SOME units - but if you assign a horizontal bomber unit
to ASW - never mind that is it is real job - AI will not let it keep that mission.

Things worked well enough to regard AI scenarios as training devices - and players wanted a version more like CVO - which is a popular basic scenario. So RHSCAIO was born - it is simply CVO with AI helping modifications as described above.

The original AIO scenario is based on EOS - and that means it is focused on a Central Pacific offensive by Japan. But AI does not understand this - so the extra power of Japan is squandered and just left for the Allies to mop up. Players wanted a scenario that used these assets better - so MAIO was born: modified AIO is focused on the South Pacific - and AI is a bit less confused by that.

[el cid again]

A couple of problems that might be correctable remain: one of these is that AI attacks Kamen in the USSR - in spite of passive Russians.
This may be corrected by relocating and renaming - reverting to the stock town of Oglahamensk - which does not suffer the same fate. The 7.9 edition of AI scenarios will incorporate this possible fix - but ONLY for AI scenarios - human scenarios retain Kamen.

[herwin]

Do an AI scenario based on a limited war. Japan just takes the SRA, the Philippines, and a bunch of minor locations. It's conservative, so it should be easier for AI to manage. The Allies have to win by the end of 1943.

[el cid again]

Another thing is that there are AI modules which are more programmable if you don't use the entire map area. These probably make
the AI far more "intelligent" - you get to select things like basing priorities and so on. This is a good idea - but doing it would require a lot of time - and so I think it is a project for the AE world - assuming we get AE to work as expected. Since a lot of other problems will be addressed - in particular getting rid of hard code - and a lot more slots will exist - it might result in a very fine thing. Sort of like a UV type scenario in the WITP system.

[el cid again]

Another problem reported is that Japanese invasion groups in MAIO run into mines which were laid by Japanese ships. I am not able to make sense of this - no submarine or minelayer is set to lay any mines - but in case it happens under AI control that a minelayer entering a foreign port lays mines - I removed the ML from the task forces. They sit in port waiting for AI to come up with a job for them - if it ever does that.

[el cid again]

A concept that has not gone anywhere is to write a routine that reissues the orders for the AI side. This involved three steps:

1) Create a first turn with proper programming for the Japanese side - and a player starts using that turn instead of from scratch.
That worked - but there is no current turn to start with

2) Use a Guided AI service - sending in turns so we could pretend to be AI - and gather data about how AI messed up. Only a couple of turns were sent in - so we never got the data.

3) Write a special utility program that used the data to write orders. A player would then play a game as if it were face to face - but would run the utility for the other side. The intention was to do this for Japan - but if it worked it could also have been done for the Allies.

This concept is suspended until we see convert over to AE standards. I estimate we might return to it in about a year - but again - only if we get enough turn data to work out what the utility needs to do.