The Quest Mode

[gohongohon]

I know it has been a while since the last post (but i love the fact that im James bond deck ) but i have been playing every version of the game and the quest mode. It has been really fun and i like it a lot but... (lol i hate when people do stuff like that). Playing it when you get a pack a game is just bad and boring. But the thing is playing it on very hard mode is a little silly. I have got to the point were it just seem silly to play without jace 2.0. Also all my decks look about the same RUB control. With jace milling is easy to get 100-200 per match which means you can get some upgrades and stuff. With agro (which with the pool you get from very hard is just bad) you just get the win bonus which is what 20 per match and we get a pack every other match just makes it really hard to build. Not sure if im the only one that plays very hard mode but i would think the only decks are being played is control. I mean says you go w/g agro you have what at most to start with 2 removal spells (that is being generous) you cant be the flyer decks or the wall deck or the forest walker deck and all the other agro decks are just plan better. well sorry about the rant and stuff i guess i will say some ways to make very hard mode better.

I would make winning with creatures worth more maybe like 60-80 per round. You could make the store have more items in it (yeah i know that it gets more and more the farther you go) but in the start is rough unless you hit some common removal in the store. Maybe give very hard a modifier with the gold it does get so like, easy is 100% normal 120% hard 150% very hard 200% something like that. Or atleast you could make it worth something to play very hard. Say more commons and uncommons to start with. Maybe quest quicker something im not sure. Why not more lands (not basic)

Other idea is just not having different modes.

[timmermac]

I'm actually finding that I need to handicap myself in Very Hard mode in fantasy. I've taken to playing with my entire collection, as if I play with a mono-colored or 2-colored deck, I'll usually monkeystomp the computer. This is after I've collected the pet/wall, of course.

[gohongohon]

i agree later on you are right but not at the start considering you cant make a one color deck in the beginning. At best you get something like 18-19 of one color and half are not playable. That and i dont like to lose which is really hard to do lol. I just want it to be worth playing agro is all i guess. Really the only way to make the game harder is by making the ai a lot better.

Not to sound mean or anything but is the ai limited by the programmers skill or something else. Like say the programmer is an avg player and made the ai. Would you say if the programmer was of really high skill and made the ai. The ai would be alot harder?? Sorry not trying to sound mean but i dont know how programming ai .

[timmermac]

Actually, at the start, you can make it extremely easy on yourself. You can pretty much pick 2 colors to start with, sell everything in the other colors, and buy plant/pet and make it really hard for the computer to beat you. The only reason that I'm 8-23 so far in my current quest is that I consciously decided to play with every card until I get to either 51 wins or enough components of the various decks that I want to build to put one of them together. I'm only selling cards if I see something in the card shop that I need for one of my decks or a card that I win has cumulative upkeep.

[Rob Cashwalker]

The program as a whole was started by mtgrares, who has said numerous times that he has only played a handful of actual games and doesn't own very many cards at all.

I've been playing since Mirrodin, I've played at FNM and pre-release tournaments, drafted a couple times with some of the guys from the store and I'm teaching it to my daughter. I'm not a great player, but I do alright.

The AI is limited more by architecture than anything else. It is not capable of "strategy".
There is individual card AI that in some cases tries to look at the game state and make an "educated guess" about playing the spell or ability. A few of these AI decisions specifically look for other cards in play to attempt to take advantage of "combos".
There is combat AI that tries to make semi-intelligent combat decisions.
However, the AI isn't able to take advantage of instant-speed reactions. (Counterspells are a total hack, and the method isn't applicable to other spells)

On the other hand, the AI often comes up with an amazing play every once in a while.

mtgrares has often posted about advanced AI concepts he would love to try to apply to Forge, however it would practically be useless to do so. It would require significant architectural changes not to mention what it might do to the already sluggish performance in certain game states.

The AI decks for quest mode are specifically designed to take advantage of what the AI is good at - aggressive combat-oriented decks. Their difficulty comes from the card choices, not from better decisions.

[Chris H.]

Thank you Rob for fielding this question. I like your response. There is a couple of areas where I would like to add some additional input.

Forge is a non-commercial app and we do not have a company investing millions of dollars to help fund the project. The AI for the old Shandalar game was limited and the AI for the new version that has hit the markets recently also has a limited AI. Forge is not the only mtg app with a non-optimal AI.

I agree that it would not be practical to implement advanced AI concepts. During one of our brainstorming sessions we started to go a little overboard. Rob brought back down to earth by stating we should take small baby steps forward.

There are a few non-combat oriented decks for the AI to use in quest mode. Counterspells, bounce type spells, Black Vise and a few others. The AI will play a few of these types of decks and the change of pace will sometimes surprise me.

[Rob Cashwalker]

Another point or two on the AI-

A vast majority of cards are hard-coded. Many effects have been copied from existing cards or written twice with two different approaches. (even when written by the same person) This can make the computer seem dumb when it can handle one card better than another, even when it's the same basic effect.
To this end, our general approach is to systematically replace as many of the common effects with a keyword handler. Then the AI for each effect becomes centralized. While building the keyword handler, I usually try implement a better AI or at least more flexible one than was originally used in the hard-coded card.

Given the architecture, the single biggest leap we could take for the AI would be to enable it to react to the game at instant speed. Currently each SpellAbility has a "canPlayAI" method which assumes that it is always the computer's turn during a main phase. (sorcery speed)
I see no reason we couldn't flesh out a list of "canPlayAI_{action or phase/step}" methods, like "canPlayAI_InResponse" or "canPlayAI_MyAttack".
Then, as part of resolving any spell, the game checks all instants in hand and abilities on the battlefield for "canPlayAI_InResponse". The AI for a counterspell would look at the spell on the stack and the gamestate in general, possibly returning true. Then the game would execute its resolve.
Likewise, during combat, the game could check for Giant Growth's "canPlayAI_MyBlock" and find out that one of its creatures is blocked, and Giant Growth would save it.
The AI would still be making relatively primitive decisions, but with the AI able to react to your plays, it will become more challenging.

[gohongohon]

I hope i do not sound ungrateful or anything. I was just trying to found out how it worked. I'm not a programmer or anything so i don't know how easy it would be to make happen but here was a few little things i think could help the agro decks be better.

Not sure if it has changed but it seems that if you have 2 2/2 first strikers they will attack with there 4/4.

It also cant seem to know how to win most of the time outside just having more attacks then your health. So like they have 5 2/2 and i have 2 4/4 and at 2 life it will attack with everything. But say i was at 8 life it will not attack me because i have 4/4's but it cant see if it just attacks all out 2 of his would die putting me at 2 and next turn he would win if i dont put a blocker.

In the sliver deck (and most decks) it will skip the first main phase to attack and then play his slivers 2nd main phase. Also the use of wild mongrel has gotten better but could use some kinda coding where if it is targeted by black removal (terror and the like) to change to black and live.

Also if you could find a way to let the ai know about Pestilent Kathari it just wrecks them lol.

This might be my lack of ai knowledge (which is zero) but i thought madden has an ai that can learn from there players that play them. I know that EA has billions of dollars but is there something that is like that but for cheap or something. I know there are draft apps on the net you can play and the drafters get better though you drafting more and it learns off of what you do (tcg player)

How does the AI read a card like overrun. Like what makes it trigger using it. Say when it has 4 creatures or will it say do X = my life + Y = the defenses of my creatures Z = the attack power after overrun. So if Z > X + Y use it or if Z < X + Y but Z > X use it. That i would guess be easy to do but what makes it hard is say Z < X + Y and Z < X still use it if you can see one turn in the future. So say he is at 20 and the over run deals 15 (numbers are just an example) and he has more attack's then opponent has blockers and could kill him next turn if nothing changes. Like 5 2/2's are left and he is left with 2 4/4's.

To me it seems that the AI cant see the next turn ahead like say a chess ai can do. Sorry that is a lot of text i hope it makes since and i really like that you guys are active. Any ways back to trying to stay undefeated.

[gohongohon]

I'm not sure you care ( but i think you do) but i think i would be willing to note everything that the computer is doing wrong if you wanted maybe i could make a post about my day of playing and stuff that could might make the ai better.

[gohongohon]

Here is what i say form one match with the MONO WHITE DECK (not sure the name)

1. Played Savannah Lion before playing the Soul Warden in the same turn.

2. Kept gaining life with Ajani Goldmane when he had 2 creatures on the board and i had zero blockers (it seems to go right for the token which in the end made him lose because i took it with mind control lol)

3. Used Elspeth to make 1/1 tokens when giving something flying and 3/3 would deal more damage. In an agro deck (without 2 or more Anthoms in play) Making something fly will net more damage the 1/1 token. The token is more for control mode not attack mode (not always). OR when you need the blocker to protect form the counter attack.

4. Used Armageddon when i have 1 Signet 1 Oblix's To his Mox. With his 4/3 loin to my Hand of Cruelty Plus Jace the Mind when he had not land in hand. I think Geddon is way to hard for a computer to use.

5. With the new M11 blocking rules (i think we use it) it is not picking the right order if we are not it seems to deal damage to the block i set first and then second. Example he attack with Bull Ceradon (5/5) i block with Hand of Cruelty first and then his Angel of Mercy (3/3 flyer) he kills nothing... my guess is he dealt 3 damage to the Hand and 2 to the Angel. Way to fix would be if it cant deal killing damage to both have it deal to the one with the highest one. I would guess it would be hard to evalate which card is more of a threat like say Dark Confidant or Gray Orge

Im not sure if something like this would help and if not i will not bother you with it anymore

[timmermac]

I don't have enough authority 'round here for my $.02 to make a difference, but I certainly don't mind this much. I occasionally notice the computer doing funky things, especially with decks that I've made. Take, for instance, my Burning Elf decks. They have 4 copies of Beastmaster Ascension. The computer will cast one and not attack with creatures. The way I built the deck, a human player shouldn't win unless they have a very good deck. Same goes with my Sliver deck, which isn't included with the game, but most of the decklist is somewhere in the forum.

[gohongohon]

it is not matter of having a good deck. the thing with Beastmaster Ascension is compared to Overrun or its ilk is the overrun is just better. Really Beastmaster Ascension is just a bad Anthem that takes a long time to get online and is bad when you have no creatures (like both cards) Say on turn 5 you have 5 creatures and tokens and you have beastmaster online that is really not better then just using overrun to kill them on the spot. It also makes you do silly things just to get tokens like make bad attacks just to get token i could not see how the computer could use the card right. SO say you have BA on the board with zero tokens on it with say 3 creatures but my 2 blockers will kill 2 of your attackers. Would you attack just to get 3 tokens... I really think it is to hard of a card for the computer to use and really for most people to use. If i were to make a deck it would with nest envaders and the 3/3 and 2 tokens and awakening zone. But still in that deck i would rather play overrun or overwhelming stampede.

[Chris H.]

The AI used by chess, Madden Football or a draft app is not necessarily similar to the type of AI that a MTG app would use. I am not an expert in AI code but I think that MTG apps tend to use a "hinting" system to give the computer an idea of what to do.

The forge AI assumes that it is best to summon creatures during the main 2 phase. In many cases this may be the right thing to do. But summoning "Lords" should be done in the main 1 phase. I guess that "hinting" systems work well for many of the decisions that are made by the AI. But situations may come up where this "hinting" system causes the computer to do something stupid.

At this point someone may add in additional code to handle this "unexpected outcome". Someone recently mentioned an oddity in the way the AI uses Crib Swap. I have looked at the code for this card. There is some AI code for this spell but it may not be optimal. I will take another look at this code and I may come up with a better "hint" on how the computer should use this spell.

[timmermac]

it is not matter of having a good deck. the thing with Beastmaster Ascension is compared to Overrun or its ilk is the overrun is just better. Really Beastmaster Ascension is just a bad Anthem that takes a long time to get online and is bad when you have no creatures (like both cards) Say on turn 5 you have 5 creatures and tokens and you have beastmaster online that is really not better then just using overrun to kill them on the spot. It also makes you do silly things just to get tokens like make bad attacks just to get token i could not see how the computer could use the card right. SO say you have BA on the board with zero tokens on it with say 3 creatures but my 2 blockers will kill 2 of your attackers. Would you attack just to get 3 tokens... I really think it is to hard of a card for the computer to use and really for most people to use. If i were to make a deck it would with nest envaders and the 3/3 and 2 tokens and awakening zone. But still in that deck i would rather play overrun or overwhelming stampede.

You bring up a valid point. However, Overrun is a 1-time only effect, and the Ascension gives your creatures the +5/+5 permanently. There's also another card in the second version of the deck that speeds thing up considerably. I'm just going to pop out the decklist for both of the Burning Elf decks.

4x Priest of Titania
4x Skyshroud Elf
4x Llanowar Elves
4x Elvish Champion

4x Gaea's Anthem
4x Glorious Anthem
4x Blaze
4x Fireball
4x Lightning Bolt
4x Beastmaster Ascension
4x Coat of Arms

4x Savannah
4x Plateau
4x Taiga
4x Stomping Ground
4x Temple Garden

The second version of the deck replaces the Glorious Anthem with 4x Doubling Season. I may create a third version that replaces the Plateaus with 4x Gaea's Cradle here at some point.

[gohongohon]

Here is how i make a deck. First i will point out the problem with this deck.

4x Priest of Titania (Best elf)
4x Skyshroud Elf (What is the point is it because you can tap for red... your mana is fine with out it)
4x Llanowar Elves (Fine elf)
4x Elvish Champion (Weak lord if we are looking for ramp Elvish Archdruid is way better)

4x Gaea's Anthem (this is really for all 16 of the anthem effects including Coat of Arms and beastmaster. in the current version we are playing 16 creatures so... that is way to few bodies and all the bodies are really too weak for BA to work without dieing. So it would seem you will lose atleast 3 attack to get it going. Also you could not attack until you have a few lords and anthem which by that time the opp. will have his game plan it will not matter.)
4x Glorious Anthem
4x Blaze ( I will group the 2 X spells in one that is way to many X spells 3 should be enough)
4x Fireball
4x Lightning Bolt
4x Beastmaster Ascension
4x Coat of Arms

4x Savannah
4x Plateau
4x Taiga
4x Stomping Ground
4x Temple Garden

With a tribe like Elves Anthems are not needed because of there lords so if i was going to play a Elf Ball deck i would drop the balls (lol) and work with BA in Elves here would be my list nm that i will build a Elf Monument deck and a BA deck If you guys want to use them for the game then that is cool .