Well obviously this cannot be done with anything as complex as spring however my proposel is to let the community build the Intelligence.
Isnt there a way to allow players to record their game to be stored in some main server upon the completion of a game?
I play a very good 1v1 BA game vs someone else and i wish to let the AI learn form my experiance meaning base his BO and reactions on what i did in certain situations.

Isnt it possible to do that?It can be done in stages.
AS i see it there are a few types of units.
1)Attcking turrets like the llt hlt guardian etc..
2)Resource structures like mexes fusions advanced solars etc..
3)Buildings structures like labs and nano turrets.
4)artillery units
5)assault units
6)bombers
7)fighters

Why cant the player sort each unit and building manually and ssend it to the ai.
Each unit type would have certain rules to guide its use by the ai.

In short allow the AI to mimic players by allowing players to reocrd their plays thus eliminating bad behaviours.

it's called "uploading replays to a replay site" and could be done automatically by an autohost or something.

anyway, NN's are completely unsuitable for RTSes due to the time they take to learn anything at all, especially when you consider the number of inputs an RTS has.

Though you could probably get something cobbled together if you made modular AI buttons for it to press for each (large) sector like ATTACK, DFENS, ECON, RAID, BOMB. That along with heavy processing code to simplify the incoming data. You might be able to make a net that outperforms many humans, but only on the southern plains of DSD.

viewtopic.php?f=21&t=16797&p=320350#p320350

If you read a bit further u'll see the proposal about neural nets. Though not completely on this level, I will be gathering statistics and depending on the results an AI may be able to benefit from it.

Neural Netowrks arent a singular solution, they require pre and post processing and real world solutions usually make use of several other AI algorithms. Essentially they are just function approximators/optimizers. So given Inputs of what the composition of the enemy force is, it could learn what the best counter force is, but they wont control the whole decision making process.