Fiddling with the attack/defense knobs in the .ini

[lujo]

I very much like these being there and I've had a wonderful experience of playing against a totaly different AI just now. Not necessarily smarter, but definitely different, and it was doing some things I watched with my mouth open in awe as I thought "Oh, so you CAN do that you bastard!".

They'll take feedback.

Now I reported something related to this allready and it got a "not a bug", and while it's "not a bug" it's affecting performance a lot. The AI is too eager to attack with 0 powered creatures, or creatures which will just survive combat even without any fiddling with the .ini it seems. This, sadly, ends up as "the AI will effectively just tap it's dudes", and the kind of dudes it'll tap are the 0/4 or 1/3 dudes whom it needs for defense.

I got the AI to attack more and chump-block less, and it was a pleasant and strange experience where the AI was suddenly pressuring me and attacking into trades, but it seemed to attack when it had no chance of either dealing damage or killing anything in combat. That's kind of the spillover from the "eager to attack with whatever will survive attacking". It seems to be able to calculate what that is even at about -84 "reluctant to attack" but attacking just to have everybody survive and do no damage AND kill nothing says something is missing.

It was also a bit frustrating that it had to become suicidaly "brave" to just attack with it's critters. On the other hand, I've kinda built the decks for an insanely paranoidly defensive AI. Feels weird that the dials even go any further in that direction, I have to see what happens if I turn them that way.

I'll run some tests and post some results, and anyone who does or has adjusted their attack/defense modifiers in the .ini is wellcome to, too.

The main thing is that this kind of setting will always suck on a global basis whatever it is. If each enemy had it's own overriding stats like these, so that you could tweak it's style as you build it's deck, that would be very interesting. Considering that the difficulty setting seems to override the thinking time as set by the shandalar.ini, I wonder if it's possible to hook up behavior-pattern overrides to individual enemies? That last bit might be just nonsense, but who knows?

[lujo]

So here's some feedback on this thing, not much but possibly meaningful and an explanation for why it's difficult to test this.

Getting the AI to be aggressive is relatively easy by using the "Human isn't likely to block" knob. This is a problem, because whether this should be a factor at all depends on the deck being played, and it's also a problem because this makes the AI aggressive in a stupid way. It will attack with little guys into big guys thinking his guys won't be blocked, and it'll achieve nothing.

If you dial the thing to "Human will block" this makes the AI extremely reluctant to attack (like, seemingly even more so than usual, if that's even possible). Which, again, is a problem, because what the AI isn't any good with is attacking into trades. "Human will block" dialed up to 100% should not be a semi-binary "makes you attack or makes you not attack" thing, it's really supposed to initiate an assessment of who'll end up with more stuff on the board post-combat and that ought to determine whether to attack and with what.

IDK what equation the AI is using, but right now I'm not sure it can be made to play remotely right with the tools we have.

And also, the obvious problem with what I've just said is - even if you could make the AI think the human will block, and it then decides to attack into a trade, the human can "trick it" by not blocking, leaving it tapped out and open to attack. The thing to note there is that right now the "getting tricked" scenario is the ONLY thing the AI will "settle for". It'll only attack if it thinks it's dudes won't get blocked and/or die, which more or less means that it'll either attack madly to achieve nothing (noone will die but noone will do anything to the enemy), attack madly to die for no gain (because it was falsely assured it won't be blocked) or attack madly into the worst scenario (human just tanks the damage with life total and counterattacks ftw).

The reason it's somewhat extremely difficult to test this, and it's unlikely to really get much more feedback than this IMO, is that the idea of having universal knobs like these is... I lack a word which doesn't unintentionaly translate to an insult... "uncompatible with properly and moderately sensibly played magic the gathering as assumed by folks who make the cards and set the mana prices for the effects and divided the color pie and color design philosophy" is what I would want to say.

If I narrow down the card pool and effect pool to some fraction of what there is, I can make X number of decks for the ultra-conservative default knob settings. But then all decks will play essentially the same, because the global parameters narrow down the design space (especially when you factor in bugs, misimplementations, lacking mechanics, rules discrepancies, etc etc). So much so that making even remotely functional decks in certain colors is madeningly difficult and only really possible due to outlier cards which the color only ever got 1 or 2 of in 20 years.

But if I adjust the valves to make the AI madly aggressive in order to suit, say, red decks better, I would have to then make decks conforming to a new set of arbitrary rules (a different portion of the wide design space which MtG consists of), and I couldn't apply those to other colors or decks made for the default system. If I made some strategies work, I'd make others collapse, and I'd have to make a set of 50 decks for every setting. This is an insane proposition, making one deckpack with all decks being functional, for this game right now and probably for the forseeable future is something a complete fool would think possible. I basically fiddle with it out of masochism.

The basic thing is that MtG doesn't have "optimal" global settings and that it's always been designed with "perfect imbalance" in mind (often in order to make players need to buy more than one deck or collect cards out of more than one color and thus buy more cards). So you can't have the same AI playing red and blue decks or even have the same AI play two different decks in the same color.

This is true even if the decks look completely similar. Like "Burn" and "Sligh", both being very fast agressive red decks, except the first is effectively creatureless and just throws stuff at the enemy head, while the second uses burn as removal to clear the way for it's cheap beatdown creatures and then after turn 3-4 throws burn at the opponent's life total. These decks can't be played by the same AI, and the AI which suits one would make the other complete garbage and vice versa. The first one will always throw Lightning Bolt at your head and if it doesn't it's completely misusing it, and the second will always clear the path with Lightning Bolt and if it doesn't it's completely misusing it (until turn 4-5 anyway, but one misused Lightning Bolt and the deck lost the match).

This makes global settings of any kind both pointless and difficult to test. I could probably find the settings that would make a particular type of red deck functional as opposed to disfunctional.* This would take time and effort. It would also turn some number of cards specifically designed for that type of deck from "might as well not be implemented for all practical purposes" to "I can use these in AI decks". In case of the color red (and also green) this can mean huge swaths of cards. But then I'd have to turn these settings back to default because this would make some other color disfunctional, which means it's literally a waste of time.

Not in the ever-popular "waste of MY time, as determined by me, because I don't wanna" but literally, because nobody can do anything with any findings or results unless there's a system where you can set these for individual opponents. Then it would make sense to figure out what settings work best for what kind of deck, assign proper knob configuration to particular opponents (lots of things to consider there, too, as how much starting life a deck has is immensely important), and then give those enemies the right kind of deck to go with their configuration (or vice versa).

*(Or rather, with the settings currently available I'd have to do the opposite, see if there's any red strategies / cards madly agressive enough to work with the current possible aggro settings. I would say this is time consuming but I'm not entirely sure most folks who thow that phrase around actually have the correct idea about just how time consuming something like this would be.)