PDA

View Full Version : Is the Hex AI really going to cut it?



ev1lb0b
11-12-2013, 07:32 PM
It seems a daunting task trying to code in all the decisions that a human player can make during the course of the game. Even with the limited card pool that Alpha is offering I've seem some complex decision making situations and have made mistakes....how can the AI possibly 'get it right'?

I assume that the current AI is just the first cut and has only the basics coded (dont attack with my 1/1 if the opponent has an untapped 2/2 etc) and it's buggy (playing Dingler on it's own creatures etc) so it's not any sort of benchmark at this stage but the PVE side of the game is arguably going to be much bigger than the PVP side so the AI is going to need to be 'on it's game' to challenge and (hopefully) utterly destroy us from time to time.

Am I worried about nothing?

Gwaer
11-12-2013, 07:39 PM
It's a daunting task to be sure. I really do think Chris Woods (Magical Creature) is up to the challenge.

Stok3d
11-12-2013, 07:40 PM
Am I worried about nothing?

Yes.

AI is low priority. Server Stability is highest as they want to get everyone into Alpha ASAP.
I believe the only true issue will be that the AI is too good and they will have to find ways to dumb it down.

ev1lb0b
11-12-2013, 07:50 PM
Yes.

AI is low priority. Server Stability is highest as they want to get everyone into Alpha ASAP.
I believe the only true issue will be that the AI is too good and they will have to find ways to dumb it down.

I am looking ahead 2-3 sets and multiple PVE experiences (dungeons/raids/keep battles etc) ahead of now....I have no expectation outside the basics for alpha.

Xenavire
11-12-2013, 07:54 PM
Yes.

AI is low priority. Server Stability is highest as they want to get everyone into Alpha ASAP.
I believe the only true issue will be that the AI is too good and they will have to find ways to dumb it down.

They need a chess masters mode, where they unleash the AI's full potential.

keroko
11-12-2013, 08:23 PM
I'm guessing the AI is set to be quite inept at the moment for a couple of reasons:

* - to allow new players who might be intimidated by playing another human to interact with the ai / test decks in a setting where the system doesn't immediately set about destroying them.

* - to allow for longer games, so to catch bugs from card interactions

* - self cast inopportunely etc. to test effects of cards miscast, opportunistically cast. It might use gap in current playtest data of its decks versus player submitted by considering 'holes' in its gathered testing criteria / datasets.

That last one's kind of tenuous - they'll no doubt make the ai play itself in simulation for the bulk of initial card testing interaction, debugging the results / jam ups seen in those 'games'.

just my guesses.

Zophie
11-12-2013, 11:12 PM
they'll no doubt make the ai play itself in simulation for the bulk of initial card testing interaction

This just makes me think of the Tic-Tac-Toe game at the end of the movie Wargames. :cool:

HyenaNipples
11-12-2013, 11:32 PM
According to Cory, the man they have doing the A.I. for Hex TCG is the dude who Danny Ocean would pick to make a TCG A.I. for some over-elaborate caper. 'Best in the world' is essentially what keeps getting tossed around as a descriptor.

Plus, Duels of the Planeswalkers has a decent A.I., and it has to work with cards much more complex than those in this base Hex set. So that is evidence enough that it is certainly possible to create strong A.I. for TCGs.

Eierdotter
11-13-2013, 02:48 AM
i do not see a problem in making the AI as good as we expect it to be. (in regards of making good decisions)

But i guess it would be hard to code something like "risk" into a AI.
short scenario:


i play a diamond deck and have 4 resource open with 3 card in hand. i am at 5 life with two exhausted Quick Strider (2/2, swiftstrike)

AI plays diamond and has an Aeronaut (3/3 steadfast, flight, lifedrain) 5 resource open with diamond aura in hand. only 4 life left.

should the AI attack, yes or no?
if he attacks:
- he could buff him with diamond aura to win the game
- i have 3 cards in hand, maybe a Repel, so he looses for shure

if he does not attack:
- i could attack and buff the blocked one with +1/+1 to kill the aeronaut (a scenario the AI has an answer to)
- i could attack and buff the unblocked one with a diamond aura if i draw a resource
- i could inner conflict the aeronaut and attack
- i could play a flying troop
- .... many options with 3 cards in hand

of course the AI could be programmed to know the players hand and base the decisions on that, but that is sort of lame.

Niedar
11-13-2013, 03:06 AM
According to Cory, the man they have doing the A.I. for Hex TCG is the dude who Danny Ocean would pick to make a TCG A.I. for some over-elaborate caper. 'Best in the world' is essentially what keeps getting tossed around as a descriptor.

Plus, Duels of the Planeswalkers has a decent A.I., and it has to work with cards much more complex than those in this base Hex set. So that is evidence enough that it is certainly possible to create strong A.I. for TCGs.


You really can not trust what Cory says when he talks about stuff like that. I am not trying to say he is untalented but really we don't need these types of exaggerations.

BlackRoger
11-13-2013, 03:32 AM
Im not worried about the pve expirience too much. as was said duels of the plainswalkers had a decent ai and I suspect they put much less effort in then hex will.
The keep defense might be more problematic though, since they cant test those decks if they dont know what these will be.
Dont expect the Ai to play your combo decks properly.

keroko
11-13-2013, 06:38 AM
...of course the AI could be programmed to know the players hand and base the decisions on that, but that is sort of lame.

perhaps better would be AI that cosiders past games on the system (they are recorded), and extracts heuristics on the likely behavior of a given player, deck 'type' as seen by cards played and likely other deck components (much like we humans can guess what we're up against pretty quickly when seeing it in play).

I'd like to see generated nemesis AI, which would plague you in PvE (with exceptional rewards if defeated) - that know's your past moves, your likeliehood to bluff / hold back land pretending its a counter etc etc.

They could potentially do a great deal in this space; they could build dragons - have players defeat them, and then rebirth them iteratively stronger with each new life based on collected data.

Shadowelf
11-13-2013, 06:49 AM
AI in Alpha is meant to serve as a deck testing tool; they know its far for perfect

http://www.youtube.com/watch?v=cUWPaB5bScU check 0:23:50

hex_colin
11-13-2013, 11:01 AM
You really can not trust what Cory says when he talks about stuff like that. I am not trying to say he is untalented but really we don't need these types of exaggerations.

I think you'd change your mind if you ever actually met Chris Woods. ;)

Xenavire
11-13-2013, 11:15 AM
I think you'd change your mind if you ever actually met Chris Woods. ;)

I wish I could - they really need to come to EU next year. I dont really give 2 hoots which con or event, but I want my fair chance to meet them.

Stok3d
11-13-2013, 12:10 PM
I think you'd change your mind if you ever actually met Chris Woods. ;)

My thoughts exactly.

DreamPuppet
11-13-2013, 08:18 PM
I'll admit that i'm a bit worried about the AI. If they're going to have over 100 unique PvE encounters (1v1 dungeons to 3v1 raids) at launch how are they going to make them all work properly when the PvP AI right now (i know it's alpha) is as stupid as it is? (using murder on it's own things, using dingler on it's own things... Just to name a couple of the more flagrant things i've seen)

As someone else mentioned, duels of the plainswalkers AI is pretty good but even that one makes some mistakes and they have what less then 10?? true AI encounters to get working in the game and a few others where all the AI does is play the same card 20 times in a row which doesn't need much intelligence.

Of course PvPers only care about server stability since they'll never face AI opponents.

Skirovik
11-14-2013, 01:11 AM
perhaps better would be AI that cosiders past games on the system (they are recorded), and extracts heuristics on the likely behavior of a given player, deck 'type' as seen by cards played and likely other deck components (much like we humans can guess what we're up against pretty quickly when seeing it in play).

I'd like to see generated nemesis AI, which would plague you in PvE (with exceptional rewards if defeated) - that know's your past moves, your likeliehood to bluff / hold back land pretending its a counter etc etc.

They could potentially do a great deal in this space; they could build dragons - have players defeat them, and then rebirth them iteratively stronger with each new life based on collected data.

I really love this idea! :D

keldrin
11-14-2013, 05:44 AM
perhaps better would be AI that cosiders past games on the system (they are recorded), and extracts heuristics on the likely behavior of a given player, deck 'type' as seen by cards played and likely other deck components (much like we humans can guess what we're up against pretty quickly when seeing it in play).

I'd like to see generated nemesis AI, which would plague you in PvE (with exceptional rewards if defeated) - that know's your past moves, your likeliehood to bluff / hold back land pretending its a counter etc etc.

They could potentially do a great deal in this space; they could build dragons - have players defeat them, and then rebirth them iteratively stronger with each new life based on collected data.
That would be kind of like a chess master, using a chess database to prepare for a specific game against another master.
A little strange though, since the past data is dependent on the cards you draw. And will the computer be able to know if a deck has been altered. Meaning it knows you added and removed certain cards.
If I have played nothing but shin'hare decks. Then challenge the computer with a elf deck. Will the AI be expecting me to play shin'hare cards the whole game.

Shadowelf
11-14-2013, 05:57 AM
That would be kind of like a chess master, using a chess database to prepare for a specific game against another master.
A little strange though, since the past data is dependent on the cards you draw. And will the computer be able to know if a deck has been altered. Meaning it knows you added and removed certain cards.
If I have played nothing but shin'hare decks. Then challenge the computer with a elf deck. Will the AI be expecting me to play shin'hare cards the whole game.

They have said that sideboarding will be allowed on a dungeon by dungeon basis; generally dungeons that try to screw with your deck won't allow the use of a sideboard

http://www.thehexvault.com/blog/2013/09/02/dragoncon-2013-hextcg-pve-design-panel/ 0:56:20

There will some vilains that if you face and fail to kill in a dungeon, they will come back and be prepared to face your deck (the AI will be able to read what's in your deck after the first encounter and act accordingly)

http://www.thehexvault.com/blog/2013/09/02/dragoncon-2013-hextcg-pve-design-panel/ check 0:48:45

jgsugden
11-14-2013, 09:58 AM
The placeholder AI is going to be nothing like the AI the system uses for PvE. I imagine that will have AI that is designed to work with the particualr deck. And for situations where it is not, like the keep defense, they'll likely make vast improvements over time (many of them coming before release). You're worried about nothing at this point.