I want smarter opponents, not higher level opponents

Unknown

There's a particular group of powers which involve selecting a tile (e.g. Take Aim, Deceptive Tactics, Gravity Warp, etc...), and the computer just sucks at taking advantage of them because it seemingly selects random tiles. If the player has special tiles on the board, the computer's Moonstone should be smart enough to select a location for gravity warp that will match them away. Hawkeye and Bullseye should at the very least place in a position to create an immediate match. BWGS and MNMagneto should try to make match 5's, but no L's allowed.

I know there's an argument the AI should be weak or else it's too frustrating for players. I'm NOT suggesting the AI needs to be smarter about making matches to produce cascades, or anything like that. Just a couple simple if-then restrictions on when to fire off abilities. This would constitute a minor defensive buff for a couple of weak characters.

Unknown

Sorry should have posted this under suggestions.

Unknown

I agree. I love when Hawkeye spawns a critical tile and then doesn't use it.

Unknown

They already tested this and said it was too frustrating.

Unknown

MattyMattMatt wrote:

They already tested this and said it was too frustrating.

Yeah I read that too, but the quote was vague on the details of the AI and how they conducted their tests, so it's not clear to me that minor improvements in AI must necessarily lead to a more frustrating experience. I'm just saying it seems wrong to me that defensive teams can't use all their powers effectively.

_RiO_

MattyMattMatt wrote:

They already tested this and said it was too frustrating.

Tested how? By retaining the same overleveled opponents and enabling them to use skills?
Yeah ofcourse that would come out as a negative.

Pwuz_

The only other AI they could get was Deep Blue, the chess computer.

Because of his ability to predict future moves of the player, and cascades, Deep Blue typically won all matches in 3.5 moves. They were able to bring that average up to 4.6 moves, but that required giving Deep Blue Yelena & Bagman for 2 of their 3 characters for all matches. The play testers said they felt very emasculated to have their Max Level Buffed Sentry, Buffed Daken, Buffed Hood Team from the lightning round beaten by a Rental Featured Character, Level 40 Yelena, & Level 94 Bagman.

Apparently the middle ground is hard to code for. Use a random number generator every turn to decide if Deep Blue or Derp Blue makes the move?

Unknown

With better AI, you could keep the opponent levels lower and still feel difficulty when attacking them. Right now it feels like you're playing a system who is, for the lack of a better word, stupid. It sort of takes away from the satisfaction of winning battles, eg like you're winning a game against a child.

Unfortunately, they have to pander to the mass market, so I doubt this will ever change, unless they specifically have events where the AI changes.

Unknown

Pwuz_ wrote:

The only other AI they could get was Deep Blue, the chess computer.

Because of his ability to predict future moves of the player, and cascades, Deep Blue typically won all matches in 3.5 moves. They were able to bring that average up to 4.6 moves, but that required giving Deep Blue Yelena & Bagman for 2 of their 3 characters for all matches. The play testers said they felt very emasculated to have their Max Level Buffed Sentry, Buffed Daken, Buffed Hood Team from the lightning round beaten by a Rental Featured Character, Level 40 Yelena, & Level 94 Bagman.

Apparently the middle ground is hard to code for. Use a random number generator every turn to decide if Deep Blue or Derp Blue makes the move?

The modern middle ground for this problem would have the AI purposely playing badly until you're ahead by an amount that it cannot comfortably catch up. For example I think in Chess programs they spot the human a rook and a bishop (and always play white) before they remotely play seriously because that's the difference between the strongest human player and the strongest computer program. Of course if they did that people would just be complaining about how it's cheap the AI suddenly gets 'lucky' after the player had a great start.

From a programming point of view most of the ability usage in MPQ is quite trivial. An ability like Take Aim should never be less than a match 5 when used by a computer if they wanted to. They'd be able to calculate the outcome of moves like Twin Pistols, Surgical Strike, or Polarized Force every single time and there are just too much nasty things you can do when you know the outcome of moves like that. The scaling doesn't quite work there because being able to know stuff like this usually will allow you to do quasi infinite combos so you'd indeed likely to see a case where The Hood did Twin Pistols 4 times back to back because he knows exactly what will cascade into what every single time and you'd just die terribly unless he's 100 levels lower than you! I'm assuming you have to lift the 'can't use same move twice' restriction if you're going to even pretend the AI is smarter.