AI Enhancement Tool - first test results

[{CH}ArticleBot]

As we wrote in previous news article, we made some progress with AI Enhancement Tool programming. We decided to put the tool to test.

To do that we chose one of the maps selected for World Championship 2016 Heroes 4 online - W16 El Dorado. We let the AI play this map twice for two months. First playthrough was AS IS (AI-1), and second playthrough AI was aided by scripts from AI Enhancement Tool (AI-2).

H4MG AI panel settings:



Then we compared results, and want to share them with you.


 AI starting area:



 

AI without scripts did not claim resources just next to it's hometown even after two months:



 

The best AI-1 hero after 2 months development is a level 6 Paladin with 1 artifact.



 

Compare it with AI-2. It is a level 17 General with 9 artifacts:



 

Of course he is not using them efficiently, and his army is not the best in his own kingdom, and his skill selection could be better, but he will put a much better fight than AI-1. It may seem that all the differecne between lvl 6 and lvl 17 can be explained by initial 9 extra levels for AI-2. But number of experience points between lvl 1 - lvl 6 and lvl 10 - lvl 17 is not nearly equal.

 

Thieves Guild for AI-1. Green army had a single level 1 hero with 7 black dragons as a point of reference for the strength of the other AI armies:



 

Thieves Guild for AI-2. Same Green army here. For some reason Blue AI fares much better than before:



 

 Kingdom Overview with armies and income for AI-1:



 

Same screen for AI-2. Much stronger economy, bigger reserves and more numerous armies:



 

Conclusion.

We have to remember that this is not really a better AI, it is just helping poor AI in hope that it can use our help and fare better. And this is the result of a single experiment. Despite the fact, that AI did not choose the best possible skills, did not use all the artifacts, did not raise an army nearly strong enough to defeat human player, was not aggressive enough to start an offensive, etc, etc, we can safely conclude that AI did not waste the help it was offered. It became stronger, which means that the whole idea behind AI Enhancement Tool is viable, and players can get more fun playing against AI on existing maps.

If you have ideas that can help develop better scripts for AI, we want to see them in comments below this article or in AI Enhancement Tool forum thread.
Read full article...

[iLiVeInAbOx05]

Thieves Guild for AI-2. Same Green army here. For some reason Blue AI fares much better than before:

I would just like to note that the blue army is also AI, so it also received the same AI enhancing events and scripting as the red army.

I'd also like to note that when I ran this test with the entire map visible for the AI, the AI made even more progress.

[Taro]

Nice! Is this Eq or standard version? And on what difficulty AI was playing? And one more thing: if AI has only allies (no enemies) it plays slower than normal. Red and blue are enemies or allies?

[Karmakeld]

Well it's great to see it actually makes a difference in AI's Progress, if they get a bit of help early on. And nice work with the Enhancement Tool.

[Pol]

Because I consider AI the only drawback of H4 (the same goes for H3 although it's a bit better) this is a great news. Eagerly awaiting upcoming Equilibris release, hope that it will follow at some point.

[iLiVeInAbOx05]

Nice! Is this Eq or standard version? And on what difficulty AI was playing? And one more thing: if AI has only allies (no enemies) it plays slower than normal. Red and blue are enemies or allies?

This was using WoW version on Champion difficulty, and red and blue players are enemies. On lower difficulties the AI progresses quite fast, so there may be some things to address there.

[Romanov77]

Thank you guys for all the efforts you are making!
*high-five*

[Karmakeld]

This was using WoW version on Champion difficulty, and red and blue players are enemies. On lower difficulties the AI progresses quite fast, so there may be some things to address there.

Hopefully this will change when you get to the reduction of neutrals on encounter. I suspect that is the reason to faster progress on lower difficulties.

[iLiVeInAbOx05]

Hopefully this will change when you get to the reduction of neutrals on encounter. I suspect that is the reason to faster progress on lower difficulties.

Neutrals being reduced on encounter is already in place and occurring

The reason for the faster progress is the smaller neutral stack sizes while the AI still has all the other buffs (gold and resources so AI can build and recruit more). Then the AI is confident it can steamroll all the neutrals AND the neutrals get reduced so AI doesn't really lose much of anything. Novice is the new Champion!

[Karmakeld]

Well in some cases I believe Novice is also the old Champion - as I've seen cases (mentioned to you), where AI on Novice did progress faster than on Champion. Anywho, have you run a test, where you compared the AI with Enhancement buffs on champion with standard, non buffed novice. In essence I believe the tool was intended for higher difficulty Levels, so buffed Champion vs non buffed Novice would be a more interesting result than buffed vs. buffed.
Regarding skills and artifacts, aren't skills just accepted by the first choice offered? (meaning no selection). Now that Namerutan is back, will he continue with his artifact script? 'coz that could help improve AI's choices.

[Namerutan]

Now that Namerutan is back, will he continue with his artifact script? 'coz that could help improve AI's choices.

My mind is becoming old. Please, give me some clues as reminder...

[iLiVeInAbOx05]

It was Patfx who was working on the artifact sorting script

I'll comment more later, just wanted to post that real quick

[iLiVeInAbOx05]

Anywho, have you run a test, where you compared the AI with Enhancement buffs on champion with standard, non buffed novice. In essence I believe the tool was intended for higher difficulty Levels, so buffed Champion vs non buffed Novice would be a more interesting result than buffed vs. buffed.

I haven't, but it's pretty easy to just hit end turn a bunch and watch the AI

Regarding skills and artifacts, aren't skills just accepted by the first choice offered? (meaning no selection). Now that Namerutan is back, will he continue with his artifact script? 'coz that could help improve AI's choices.

I'm not sure how selecting skills is handled by the AI. If you have suggestions for some good starting skills for AI heroes, please post them in the AI enhancement thread

[Taro]

On novice even if AI progress faster, you can rush it very easily. That is not the case on champion. So buffed novice is not the same as buffed champion, because you will have to fight these large neutral armies, while AI will defeat them rather easily and faster than you. So it's fine as it is.

[Karmakeld]

It was Patfx who was working on the artifact sorting script

I'll comment more later, just wanted to post that real quick

I know this is old, but did this artifact script ever get finished?

[iLiVeInAbOx05]

If it did, I never received it from Pat.

[Karmakeld]

okay, I could hear if he ever got to start working on it or not.. thanks

[iLiVeInAbOx05]

If time permits, I may go back and revisit that thread to remember what the script logic was for the artifact sorting. Using the H4MG to script is so much faster than the actual game editor.