Good game, but....the AI is very bad.

[Gandalf Parker]

Originally posted by proteus:
I guess my first post told you about the AI weaknesses:

This has been done often in other threads, and better than I can do it, but lets go over these again....

1. The AI is massing weak troops. Why? I see hella lot of light infantry and cheap troops. Pointless, they are toast.

There is only 1 AI and it plays the same in small or large games, low indepts or high, low magic or high, and any nation. LI are not pointless. They are only pointless to certain nations in certain games. If you think their purpose is to beat up your hvy troops then play some more.

If you REALLY feel this can be improved then show us a forumula. What type of troop, when, how many, what percentage of the formula? Write us the AIs code so we all can test it to see if it holds up in more different game situations than the one now.

If you check out the threads I mentioned then you would see that there is alot that players can do on this even if they are not coders. Do a search for...
SUGGESTION: AI castle building algorithm
SUGGESTION: AI troop building algorithm

2. The AI is making weird strategic decisions. It is moving around with strong commanders without a point, and without a decent army.

Like much of this, this is an opinion stated as a fact. If you are going to work with quantity over quality then random movement is best. It holds up against human players much better than smart movements.

3. The AI is not equipping its heroes properly, sometimes not at all.

Another area where a small effort on your part to come up with a formula would have kept you from making statements like how easy these would be to fix. Get specific and lets test it.

4. The AI is not protecting its heroes/commanders.

Probably true. I think it does a better job when they are at home, and probably when they are first sent out. But eventually the flak troops die off and you are left with just the big ones. Especially bad if thats the AIs god.

Im not sure how well the path-finding works for something like a "go home" plan. Especially since the AI does not store information well for multi-turn actions so each turn it would have to re-decide that the piece needs to go home.

Maybe a "stop and build army if less than 100" command or something so it can boost up with whatever indepts are at that province but that would end up with "crappy stupid armies" again.

5. The AI isnt using the mid/late game spells very well.
Same problem as equipment. Come up with a formula. One of the reasons that AI's in other games can do better is that the choices are less and more obvious. The AI does cast spells, it does select what it casts, youjust think it could be improved. But would your improvement work for all nations in all games without alot of "if this and if that"? (and how well would you take it when somoene says your way is stupid and broken?)

6. The AI wont summon too many "heavyweight monsters". -> I wonder what the heck the AI is doing with its magic income.

Since its a turn-based PBEM game its hard to write in multiple turn planning. Saving up for things is difficult to write. Thats true of things like castles, temples, major spells, major summonings, major equipment. If an AI has 50 gems and needs to decide what to do with them then it will decide to make 5 castings of a 10 gem summoning instead of save up for a 75 gem summoning. To decide any different would have it saving up gems early in the game when it should have been doing small summonings.

7. The AI's battlefield tactics is kinda laughable as well.
Actually Ive seen alot of improvement in this. It now uses more formations and positioning on the battlefield. Unless I use scouts to watch their battles as they approach me I can have my strategys ruined by the formations they use. Of course such surprises are easy to beat if you know they are coming but thats alittle hard to fix in an AI.

[ May 09, 2004, 14:59: Message edited by: Gandalf Parker ]

[proteus]

Thanks for your detailed reply Mr. Parker.
I think that you are right in many things, we should come up with detailed ideas. Propably the devs know about these AI probelms as well...
Castle & Troop building algorythm is what we propably need first of all.
My idea is, that the devs should code the AI to build LOT Less LI units, concentrate on summons and HI & HC.
Maybe this is a problem because of the few AI fortresses. No fortress -> no advanced units.
IMHO first of all the AI castle building should be fixed. Than it may build HI & HC over LI.
If not, than the devs should implent troop build algorythms.
I will post more ideas soon.

[MStavros]

Yeap, Im gonna make a mod like that. I think it will be interesting to see, that what will happen with the AI, if all LI will be taken out.


Oh guys a question! Can we remove/disable spells????

[PDF]

Originally posted by proteus:
Thanks for your detailed reply Mr. Parker.
I think that you are right in many things, we should come up with detailed ideas. Propably the devs know about these AI probelms as well...
Castle & Troop building algorythm is what we propably need first of all.
My idea is, that the devs should code the AI to build LOT Less LI units, concentrate on summons and HI & HC.
Maybe this is a problem because of the few AI fortresses. No fortress -> no advanced units.
IMHO first of all the AI castle building should be fixed. Than it may build HI & HC over LI.
If not, than the devs should implent troop build algorythms.
I will post more ideas soon.

Yes, I proposed the same kind of mod some time ago, but in a more complicated manner (change cost.res of units to "help" AI to not buid crap). But a straight take-out-the-LI approach is certainly simpler. You will still have AI builds a lot of indy militias , but at least national troops should be better...

Gandalf,
Your defense of LI is good, but really LI is useless to AI, as it plays it as if it were HI ...
A human player can do *some* things with LI because he'll use it purposefully (patrol, front line fodder, mass enchantments, sneakiness...), that's not the case with the AI, so it should rather not build any !
(edit) : Why are you asking ppl to give "formulas" for AI ? Users complain about a feature they think "bad", they don't *have to* give a solution...
(Should be your techie-oriented thinking )

[ May 10, 2004, 08:45: Message edited by: PDF ]

[proteus]

Originally posted by PDF:

(edit) : Why are you asking ppl to give "formulas" for AI ? Users complain about a feature they think "bad", they don't *have to* give a solution...
(Should be your techie-oriented thinking )

Hehe yup, I agree with this sorta. Anyways Gandalf propably meant that we should help the devs with that. However they know that whats wrong with the AI, I guess they can solve those problems with coding/testing.
I will try to help them, to give them examples if I can, but I dont know that it will help at all or not.

[Gandalf Parker]

I dont think its the techie in me so much as its the diplomat in me. I could give specific examples if I knew what areas you felt you were expert at.

If someone reads you a paper that they think is good enough to get published but you dont think so, then there are alot of response you can give.

You can say it sucks, and walk off.

You can say it that in your opinion it needs something more, and walk off.

You can say it could use some improvement and at the same time give specifics.

You could give specifics and make some possible suggestions of wording changes.

Of course the easy thing is to say it sucks and walk off. If it doesnt change you can say "I told you" over and over and over. Choosing a different way of approaching it is probably as simple as how good a friend they are and whether or not you really want them to look at what you are saying. This is even more true if what you are pointing at are some lame jokes that the writer doesnt feel is a major point to the paper at all.

I do not consider myself a programmer. But I have worked with programmers and I know that you can often meet them half-way (usually called psuedo-code) then they will see the answer and convert it to real coding. It doesnt take a programmer to talk in language which trys to cover many possibilities using "if this" "if that" "but not this" although having been a parent can help

[PDF]

Gandalf,
I agree that complains are more productive if possible solutions are given...
But as long as the complaint itself is argumented ("it sucks" being *not* documented, "AI builds lot of crap LI" is better ) I find it mildly offensive to request people to find solutions ("Ah you think AI is crap ! So now tell me how can you do better yourself" ?)
The problem is not that players should fix the AI, and that's the "build AI" is feeble is no debate... Now it's to IW to state if they will either try to improve it, or else externalize the parameters to allow players to mod it (techiespeech back )