Requests for AI improvements
Report wrong Card behavior to get it fixed.
PLEASE ADD SAVEGAMES TO YOUR TOPIC !
PLEASE ADD SAVEGAMES TO YOUR TOPIC !
Moderators: BAgate, drool66, Aswan jaguar, gmzombie, stassy, CCGHQ Admins
Re: Requests for AI improvements
by Korath » 11 Jan 2014, 14:29
Still don't understand the AI combat algorithms nearly well enough to do so yet - it doesn't do the same save-the-game-state, make-a-random-move, assess-the-resulting-game-state, restore-the-previous-game-state speculation that it does for casting. The interface we exploited for deathtouch was originally meant for Thicket Basilisk and Abomination (which trigger on blocking and not combat damage), and so it's not at all surprising that the AI doesn't make an exception for first strike.
-
Korath - DEVELOPER
- Posts: 3708
- Joined: 02 Jun 2013, 05:57
- Has thanked: 496 times
- Been thanked: 1108 times
Re: Requests for AI improvements
by HarlequinCasts » 11 Jan 2014, 16:21
AI Attack Logic
I noticed the AI will only ever attack you based on how much life you have. In practice, its rule regarding this is fine in almost all cases but it seems a little odd. It doesn't necessarily need to be changed, but you can decide.
If the total power of the AI's creatures would take greater than or equal to 50 turns to kill you, the AI will simply not attack. For example in the first screenshot, the AI has 1 Raging Goblin and I have 50 life (it skips its attack step). In the second screenshot, the AI has 3 Raging Goblin s and I have 150 life (it skips its attack again). If in either case, I change my life total to 49 or to 147, the AI will suddenly start attacking.
I only mention this because in most cases it seems strictly sub-optimal for the AI not to attack, simply because you always want to chip away at your opponents life because you never know what you or your opponent will draw. (For example if I had 50 life and a Test of Endurance and the AI had a Raging Goblin, it should definitely attack!)
I do get that in some cases, attacking is pointless (like if the player has some arbitrarily large amount of life), but this seems like a weird AI restriction if not only for the corner case with Test of Endurance I mentioned.
I noticed the AI will only ever attack you based on how much life you have. In practice, its rule regarding this is fine in almost all cases but it seems a little odd. It doesn't necessarily need to be changed, but you can decide.
If the total power of the AI's creatures would take greater than or equal to 50 turns to kill you, the AI will simply not attack. For example in the first screenshot, the AI has 1 Raging Goblin and I have 50 life (it skips its attack step). In the second screenshot, the AI has 3 Raging Goblin s and I have 150 life (it skips its attack again). If in either case, I change my life total to 49 or to 147, the AI will suddenly start attacking.
I only mention this because in most cases it seems strictly sub-optimal for the AI not to attack, simply because you always want to chip away at your opponents life because you never know what you or your opponent will draw. (For example if I had 50 life and a Test of Endurance and the AI had a Raging Goblin, it should definitely attack!)
I do get that in some cases, attacking is pointless (like if the player has some arbitrarily large amount of life), but this seems like a weird AI restriction if not only for the corner case with Test of Endurance I mentioned.
Last edited by HarlequinCasts on 27 Jan 2014, 20:31, edited 1 time in total.
-
HarlequinCasts - Posts: 922
- Joined: 07 May 2013, 14:33
- Has thanked: 68 times
- Been thanked: 30 times
Re: Requests for AI improvements
by stassy » 16 Jan 2014, 18:08
When player lose a flip coin triggered from Karplusan Minotaur, AI damage itself or its own creature...



- stassy
- Moderator
- Posts: 5274
- Joined: 25 Feb 2009, 07:06
- Has thanked: 471 times
- Been thanked: 337 times
Re: Requests for AI improvements
by HarlequinCasts » 06 Feb 2014, 21:15
We've been using this thread for so long, I want to know if we can clarify what sorts of things are even relevant to post here? Is this a wish-list, or is this a place where obvious AI bugs are fixed? Or somewhere in between?
For the record, no previous posts are prompting me to ask this. I just so often see the AI do tiny things like cast Firebreathing or Pattern of Rebirth on my creatures for no reason, that I wonder if I should report them.
For the record, no previous posts are prompting me to ask this. I just so often see the AI do tiny things like cast Firebreathing or Pattern of Rebirth on my creatures for no reason, that I wonder if I should report them.
-
HarlequinCasts - Posts: 922
- Joined: 07 May 2013, 14:33
- Has thanked: 68 times
- Been thanked: 30 times
Re: Requests for AI improvements
by stassy » 07 Feb 2014, 03:42
yup, those kind of issue that make you facepalm 

- stassy
- Moderator
- Posts: 5274
- Joined: 25 Feb 2009, 07:06
- Has thanked: 471 times
- Been thanked: 337 times
Re: Requests for AI improvements
by BAgate » 12 Feb 2014, 11:21
The AI just cast a Pact of Negation on me with only 3 mana available, so he lost next turn. Is there any limit on the pacts that would prevent this?
Working on: housekeeping and archived reports
Re: Requests for AI improvements
by Korath » 12 Feb 2014, 11:35
Yes. Not enough information to repro.
-
Korath - DEVELOPER
- Posts: 3708
- Joined: 02 Jun 2013, 05:57
- Has thanked: 496 times
- Been thanked: 1108 times
Re: Requests for AI improvements
by BAgate » 12 Feb 2014, 12:45
What info do you need to repro?
Working on: housekeeping and archived reports
Re: Requests for AI improvements
by Korath » 12 Feb 2014, 13:03
-
Korath - DEVELOPER
- Posts: 3708
- Joined: 02 Jun 2013, 05:57
- Has thanked: 496 times
- Been thanked: 1108 times
Re: Requests for AI improvements
by BAgate » 12 Feb 2014, 23:35
Huh, I didn't think it was that scary a situation. If I remember, he was 20 life I was 13, he had an arcbound ravager and a 2/2 creature with 2 islands and a mox sapphire in play. I had no creatures and cast an Ardakar Valkarye (which I can never remember how to spell), which he countered with the pact.
Working on: housekeeping and archived reports
Re: Requests for AI improvements
by stassy » 13 Feb 2014, 04:23
So I guess the AI analyze opponent bf 1st before his own situation, since Adarkar Valkyrie AI base value is set to 100, which is the scariest threat for him.
- stassy
- Moderator
- Posts: 5274
- Joined: 25 Feb 2009, 07:06
- Has thanked: 471 times
- Been thanked: 337 times
Re: Requests for AI improvements
by Aquillion » 16 Feb 2014, 11:29
Is it a known issue that the AI doesn't understand flanking? It seems like it will reliably chump-block flanking creatures using creatures that would have traded off with or killed it if it weren't for flanking (often with stuff that will have its toughness reduced to 0 the moment it's assigned as a blocker), in situations where there's no need for it to do so.
The AI also doesn't understand how the Warrior en-Kor ability works (it will try to trade off attackers or blockers with it without understanding that I'll just redirect the damage elsewhere), but I suppose that might be a bit beyond it.
Finally, it doesn't seem to understand how visible activated boosters like Angelic Page work -- it will make attacks and blocks that are obviously disadvantageous for it on account of those.
Oh, and it doesn't understand how cards that drop -1/-1 counters (like Serrated Biskelion and Serrated Arrows) work -- facing a card like that, it will reliably block a creature with defenders who have exactly enough power to kill it, even if it has an extra defender who could be thrown into the block without losing it. Then I'll drop a -1/-1 counter on one of the defenders and they'll all die uselessly.
(I've always assumed these issues are well-known and unsolvable, but I noticed they're not in the thread and figured it couldn't hurt to mention them.)
The AI also doesn't understand how the Warrior en-Kor ability works (it will try to trade off attackers or blockers with it without understanding that I'll just redirect the damage elsewhere), but I suppose that might be a bit beyond it.
Finally, it doesn't seem to understand how visible activated boosters like Angelic Page work -- it will make attacks and blocks that are obviously disadvantageous for it on account of those.
Oh, and it doesn't understand how cards that drop -1/-1 counters (like Serrated Biskelion and Serrated Arrows) work -- facing a card like that, it will reliably block a creature with defenders who have exactly enough power to kill it, even if it has an extra defender who could be thrown into the block without losing it. Then I'll drop a -1/-1 counter on one of the defenders and they'll all die uselessly.
(I've always assumed these issues are well-known and unsolvable, but I noticed they're not in the thread and figured it couldn't hurt to mention them.)
Re: Requests for AI improvements
by stassy » 16 Feb 2014, 11:47
Well, seems to be AI has now a slight more chance to focus on creature with more AIB now, tough we might need more specific situational cases in order to be sure of that :

The Platinum Angel got Disenchanted 1st, but the Colossus of Sardia (which has an AIB of 20) got targeted after so not the best example...

The Platinum Angel got Disenchanted 1st, but the Colossus of Sardia (which has an AIB of 20) got targeted after so not the best example...
- stassy
- Moderator
- Posts: 5274
- Joined: 25 Feb 2009, 07:06
- Has thanked: 471 times
- Been thanked: 337 times
Re: Requests for AI improvements
by Aquillion » 17 Feb 2014, 08:12
The AI doesn't understand how Stampeding Serow works, and will cheerfully cast it with no other green creatures out over and over again, even to the point of discarding other cards it could have played as its hand fills up.
Re: Requests for AI improvements
by HarlequinCasts » 17 Feb 2014, 11:24
Sadly I've seen no increase in the AI's desire to prevent itself from losing to my beloved Battle of Wits decks.stassy wrote:Currently its AI base value is set to 40 (aka lowest priority for enchantment), so let's try to set it to 100 to see if it does anything...
Here you can see it passed on using a Counterspell and then passed on using Disenchant before my next turn win.
-
HarlequinCasts - Posts: 922
- Joined: 07 May 2013, 14:33
- Has thanked: 68 times
- Been thanked: 30 times
Who is online
Users browsing this forum: No registered users and 52 guests