If AI has 50% to hit between 2 targets and one of them has “Dying Wish” … hit the other 
(or I’ll keep spanking every boss with my Unseven 
)
Constructive Criticism Thread: Improving A.I. 😎
1 Like
They play the basic decks. I think most thoughts on improvement in this thread are directed at the boss battles.
1 Like
I wish the bosses would have a interesting mechanic which gives them a edge but just works combined with general smart behaviour…
Unfourtanetly bossbattles are just opponents with MASSIVE unfair advantages which are just beatable because they are braindead:(
Imagine the heretic in the hands of a good player…
3 Likes
Lul train a large neural network
2 Likes
Hmm you know, I don’t think they’ve updated the AI for a while. It appears to me that the AI doesn’t understand how new or changed cards work. Furthermore, I believe that the AI has no idea what “value” and value trading is. For example cards like eclipse or Taygete are great to use against AI since they seem to be ignoring their effect and care only about stats. Same can be said for dying wish minions. Perhaps the AI first “scans” the board for threats and if it finds one it tries to kill it if possible or move away from it or simply ignore it completely. What interests me the most is what exactly is a “threat” according to the AI. Is it a big minion? Is it the only minion on the board? Is it a minion with a specific keyword (ranged)? Anyways maybe the biggest problem with the AI is that it doesn’t have a plan. Yes the bosses have an unfair advantage but do they know that? Do they have a certain game plan or do they just wait for the player to take action so that they can react? Unfortunately I have no knowledge of how the AI actually works so most of these questions will have to be answered by someone that does.
Maybe that someone is you 
2 Likes
The problem is actually, how do you define ‘value’ in a way where such planning becomes trivial?
2 Likes
When you stand behind someone and don’t move, you should backstab for free damage.
(Instead of standing doing nothing like a clingy thingy.)
2 Likes
It’s entirely possible to improve the AI but feature different difficulties where the AI ignores or implements different logic. See League of Legends. It took a long time, but after enough public demand they updated the simple AI and implemented different challenge ratings.
Also, thumbs up to @kelvindagr8 for the xkcd comic. I love that web comic will all my heart (it helps I’m about to graduate with a BS in Computer Engineering).
1 Like
Kinda off topic here but I think it would be cool to play slightly toned down versions of the bosses against our friends
1 Like
What if…
You play as a boss against another player playing as boss. Both players use their boss decks.
3 Likes
…xkcd is the best. I especially love the What If? section.
So, do you have any specific solutions and what do you think the decision making code would look like? Maybe some one could construct a map of possible choices. A simple one, because an exhaustive one would be massive.
Allow bosses to take our entire decks and hand into account
1 Like
A.I will probably never be smart,The road to improvement is cheating
-
Unfair cards like 1 mana 5/5 with celerity or 0 mana draw 5 cards for the boss.
-
Multiple forms we already seen boss with two forms but switching forms at different points for different reasons would make it hard for example ever time you heal the boss turns into Mechazor for a turn.
The honest truth is we don’t need the A.I to be smart.We need the Boss to provide a challenge .The biggest problem right now is that they are trying to provide content to all level of players at same time.The boss should come with 2 modes Easy and Hard.Once you remove that fact the New player and bad players have to beat it.You can put in more unfair stuff .The bosses in dark souls or Zelda aren’t smart the provide a challenge with a gimmick. The AI could be better but is good enough at present to provide a challenge
1 Like
abstrusegoose was better.
1 Like
It’s entirely possible to improve to the AI without giving them card advantage. Simply adjusting their priorities and using assigned priority values to events to create flexible turns (instead the the rigid turn order they follow now) would create tremendous change.
The AI also had a horrible habit of killing minions it can rather than trying to avoid or focus greater threats. They too easily let you manipulate their turn with well placed fodder, when you know what they’re looking for.
1 Like
Yes, it’s obvious that the AI can be improved. People who say it’s impossible or not worth had nothing to do with AI coding. Better AI > overpowered cards new players have problems beating, IMO.
The most obvious changes are:
- the movement should be done after finding correct attack targets. Targets should be based on different types of priorities - primarily lethal, then at least some of dangerous targets that can be killed (e.g. Shadowdancer, FWM, Kelaino… - this could be a simple ordered list), then the most logical trades (at least the most basic conditions like AI minion attack > max[player’s minion health]). Right now bosses move their minions first and often they block their own lethals or don’t attack the minion they moved towards (Sarlac mainly), which is terrible.
- similarly, spell order often doesn’t make too much sense. I don’t see the point of giving a boss Ethereal Blades when they clearly don’t know what to do with that card - in my case the boss casted it last on a minion that wasn’t in reach of any of my characters, not to mention the boss was on the opposite side…
- at least some keywords should be implemented, for example like people mentioned minions with Backstab should go for backstabs if possible
3 Likes
I’d like to see:
- An open API.
- An AI ladder that bots are permitted to queue for (humans can queue too, if you WANT to play vs bots) which gives no rewards per game played but which DOES give rewards for being top at the end of the month.
- After ~6 months of that, a supported AI tournament (with a decent cash prize if a sponsor can be found), with the stipulation that all entrant’s code be licenced to CPG for free for, say, 2 years for the limited purpose of driving the in-game Boss AI.
And then the problem will be solved.
1 Like
Lol, I don’t think the argument was ever that it is impossible. Rather, it is that making such an AI is at the very least, non-trivial.
2 Likes
Anyone else have ideas?
2 Likes
This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.