The AI
Comments
-
ElvaanStrider said:@UweTellkampf
I reread your excellent OP and you say the AI will not withhold cards. I have witnessed this multiple times.
In fact, I cannot get 100% on the first Story chapter block because the AI stops playing cards after I kill 4 creatures (need 6 on 1.2). I kill them, destroy all Supports and it sits there with 6x fully charged cards.0 -
Ok, I wasn't really going to do this, because i would absolutely love to work on the AI myself, but the future of the game is more important to me.
You can't implement all the AI in a separate block and then give it a bunch of cards.
You need _more_ information included in the cards.
The current example is things like Skylander's Shot, it has a flag that, unfortunately, tells the AI it's a "buff" and should be played on it's own cards. The information on the card is wrong, not the AI, the AI is being fed dodgy data.
But the AI needs far _more_ data for this stuff.
You set the AI up to know whether it needs "life", "creature destruction", "a creature", "damage", "card draw" etc. It can have priorities based on it's current situation.
The AI is general. The information on the cards adds the specific. Sure there'll be some special effects, but the _majority_ of cards can be added by setting the AI to take basic information.
Cycling becomes possible because it becomes, "None of the existing cards meet my priorities (which are creature destruction), I'd rather _anything_ else, so I will cycle this.", but things like "Faith of the devoted" can be taken into account because if you have it on the board then when the AI asks the card "What can -you- do?", it can say "I can summon a 2/2 zombie and draw a card, but at the moment I will also do 5 damage and gain 5 life".
It's similar to java interfaces. The cards advertise what they can do and the AI just has to put it all together. The main work on the AI is general, most cards can be added quickly, a few things take a bit of effort.
Adding cycling would be pretty easy, though the threshold of when to cycle becomes murky. (You don't want to do 1 turn kills, but would the AI actually work out that it could do that itself? I'd love to know.)
Energy... That's just a new resource and pretty easy for the AI to adjust to.
Working on this would be my dream job.
(Feel free to contact me.)
5 -
I personally would love to see, for card and ability handling at least, something akin to [redacted]'s Gambit system. We could even get something like an AI tournament were the AIs can fight in the background and players get rewards based on performance.2
-
I like Greg the way he is. Why does everyone want to change Greg?1
-
Corn Noodles said:I like Greg the way he is. Why does everyone want to change Greg?
Personally the main things that _really_ break my illusion are when he kills his own creatures for no reason...
I just want the biggest problems gone, but doing it right is better than just tweaking things.
(Difficulty levels based on a rating system would actually be cool, honestly, but I don't care too much about it.)
I care about longevity, about being able to play this after I retire.
The problem is that _long_ term it's probably not maintainable the way it is. If Magic and Puzzle Quest keep going (and I _want_ them to more than anything) then there will be tons of new sets to add and things to do. The current system is set up so it takes far far too much work to add new sets. The tweaks have to be done directly to Greg, like stitching on new body parts, and his Frankenstein stitches are showing.
New sets should be more like a new set of clothes than extra limbs.
1 -
Kinesia said:Corn Noodles said:I like Greg the way he is. Why does everyone want to change Greg?
Personally the main things that _really_ break my illusion are when he kills his own creatures for no reason...
I just want the biggest problems gone, but doing it right is better than just tweaking things.
(Difficulty levels based on a rating system would actually be cool, honestly, but I don't care too much about it.)
I care about longevity, about being able to play this after I retire.
The problem is that _long_ term it's probably not maintainable the way it is. If Magic and Puzzle Quest keep going (and I _want_ them to more than anything) then there will be tons of new sets to add and things to do. The current system is set up so it takes far far too much work to add new sets. The tweaks have to be done directly to Greg, like stitching on new body parts, and his Frankenstein stitches are showing.
New sets should be more like a new set of clothes than extra limbs.1 -
Corn Noodles said:I like Greg the way he is. Why does everyone want to change Greg?
4 -
The current system is set up so it takes far far too much work to add new sets. The tweaks have to be done directly to Greg, like stitching on new body parts, and his Frankenstein stitches are showing.
Actions themselves have implementations but those are written once, and then can be applied to any card. The reason you see issues with Skywhaler's Shot and similar is because the card data is wrong.
Part B ("the tweaks have to be done...") is incorrect. The AI is not modified in any way unless a new board mechanic is added (an overload consideration was added in KLD, as it was a brand new board mechanic - note that embalm/eternalize are simply activation gems and thus already have a consideration).
The issue isn't that they have to keep adding to the AI. It's that the original implementation of the AI is too simplistic - and in some cases (see: match consideration for match 5s & supports) is flat out wrong. What they have in place is a good framework, but is currently insufficient.
0 -
There's a universall problem in coding. For a game like this there are layers of code built and balanced on how the previous layer works. Once you change something so fundamental, it can and usually causes bugs and unexpected results down the line.
I've seen many games "break" from tweaking the AI too much and development spends all their time fixing and not creating. Game goes stale from lack of content and dies.0 -
I agree that Greg could play a little better. Match better gem's and priorize cards and loyality effects. But for spell targets it's very tricky. I remember a Online Game of "Paper" Magic an XBox (2009 or 2012 iteration), so very limited card pool, but I won by a silly move: target my own creature with DD wich results in instant win through mirror damage to opponents (don't remember creature). This is what AI could not. Or should. Leave creative play for the players, to catch a reward or move to abetter board position. But yes:destroy his own creature without benefit, happens very often with Greg, should be avoided.
0 -
Poor D3. I bet when they started selling this infinitely expandable game, it didn't occur to them they might have to write some new code every now and again.
0 -
shteev said:Poor D3. I bet when they started selling this infinitely expandable game, it didn't occur to them they might have to write some new code every now and again.
When you have a money hat covering your eyes, it's hard to see anything else.
0 -
shteev said:Poor D3. I bet when they started selling this infinitely expandable game, it didn't occur to them they might have to write some new code every now and again.
Adding content, building on what is already there, is a lot easier and what you want in their position.
----------
I'm not saying leave the AI as it is, but changing it already has Greg acting funny because of the trickle-down effect I described. Changes can be made, but they have to be careful is all.0 -
@octal9 I agree, I just think from the current behaviour we see too much that seems to be code driven rather than just data on the cards... Actually, my words don't driectly match my thinking, I need to detail it more.
Like with Skywhalers... You shouldn't be telling the AI whether a card is "good" or "bad" to play on your own stuff. You should be just saying "This destroys a creature" and the AI should be able to work out that mostly you destroy the other players creatures but there are edge cases where you destroy you own.
Instead of having to explicitly set every destruction effect the AI should be able to work out whether it's a good idea to destroy it's own stuff or not.
The data on the card should be about what it can do and what it can target, but the AI should be deciding whether it's appropriate to cast on it's own thing.
The current breakdown in responsibility isn't right.0 -
Those edge cases where you kill your own creatures would be very "intelligent" choices.
Many human players have yet to unlock these decision tree branches properly.1 -
The AI does it currently sometimes, but whether it's deliberately or coincidental I don't know...
0
Categories
- All Categories
- 44.8K Marvel Puzzle Quest
- 1.5K MPQ News and Announcements
- 20.2K MPQ General Discussion
- 3K MPQ Tips and Guides
- 2K MPQ Character Discussion
- 171 MPQ Supports Discussion
- 2.5K MPQ Events, Tournaments, and Missions
- 2.8K MPQ Alliances
- 6.3K MPQ Suggestions and Feedback
- 6.2K MPQ Bugs and Technical Issues
- 13.6K Magic: The Gathering - Puzzle Quest
- 503 MtGPQ News & Announcements
- 5.4K MtGPQ General Discussion
- 99 MtGPQ Tips & Guides
- 421 MtGPQ Deck Strategy & Planeswalker Discussion
- 298 MtGPQ Events
- 60 MtGPQ Coalitions
- 1.2K MtGPQ Suggestions & Feedback
- 5.6K MtGPQ Bugs & Technical Issues
- 548 Other 505 Go Inc. Games
- 21 Puzzle Quest: The Legend Returns
- 5 Adventure Gnome
- 6 Word Designer: Country Home
- 381 Other Games
- 142 General Discussion
- 239 Off Topic
- 7 505 Go Inc. Forum Rules
- 7 Forum Rules and Site Announcements