[dogscoff]

I think the best way to improve SE4's AI is to meaningfully increase the complexity of the AI's decision-making processes. The more complex a system is and the more variables it responds to, the harder it is to predict its response to the stimuli you give it.

Make those responses meaningful rather than random, and you'll have a pretty intelligent looking AI. See the thread a while back which discussed "nethack", a game which has undergone this process for the Last 20 years or so, and is now (apparently) frighteningly clever.

As an analogy, consider the combat modifiers. At the base of it, you have a simple calculation: 100% chance of hitting, and then deduct 10% for each square of range.
Too simple. Too predictable. So the system was made more complex: modifiers were added for hull size, racial bonuses, your sensors & enemy ECM, your experience, their experience, war shrines... Now, in order to predict the outcome of a ship firing on another, you have to know all those different variables. We have acheived unpredictability- except that the combat modifier system is pretty much transparent, with all the to-hit percentages readily available at the click of an icon.

Imagine if it wasn't though. Imagine that you couldn't see your chance to hit, or the modifier you get from experience or sensors or ECM or anything. You would have an unpredictable system, but the important thing is that it would be meaningful unpredicatbility (good) rather than completely random unpredictability (bad). Why meaningful? Because the unpredictable elements are governed by meaningful decisions. (ie sensors increase hit rates, ECM decrease them) The trick to unpredictability is making these decisions so numerous and based on so many different things that you can't hope to accurately predict their combined result.

And that is the way forward with the AI. For example:
in early Versions of SE4, the AI would completely ignore minefelds and throw ship after ship into them. Then MM improved the system slightly and made it so that when the AI first encounters mines, it immediately researches minesweepers. However, this is a simple and predictable equation that can be exploited by a human - drop one single mine and watch the AI waste research and resources developping minesweepers. What is needed now is to make the system more complex, so that it becomes less and less predictable: the AI could look at the importance of the mine-laying empire's territory when deciding whether or not to invest in minesweepers. If it's some backwater neutral, don't bother. Does it already have the pre-requisite techs? Maybe also have the decision affected by the urgency of other research commitments and the possibility of other solutions (making peace, developping armour, expanding in a different direction.) By the time all this was added, you'd have absolutely no idea how the AI will respond if you drop a mine in its path. You will know, however, that it will make a pretty sensible decision.

Balancing these decisions won't be easy, but that's what modders are for. As these levels of complexity are added in, bit by bit, patch by patch, the AI will get exponentially smarter and smarter.

This seems to be more or less what MM is gradually doing now with SE4, although I get the feeling (and this is just my hunch, I have no real evidence) that SEIV patching is rolling towards a halt, concentrating more now on fixing bugs for a final, bug-free Version of SEV and Version 1.00 of SEV.

The only trouble with this method of development is that it's a very lengthy process, and that we have only one programmer. We'll get there eventually (Space Empire XII?), but patience will be required...