Why Players Think the AI Cheats Part Deux: Hyper Fighting Championship Edition


#1

All right. Let’s warm up. Roll your eyes, nice and big. Hold it. You were going to do it anyway so let’s get it out of the way.

This week’s bounty and raid were fairly miserable for me, marked with a lot of losses out of nowhere. When I start losing, I start paying more attention. I want to see what’s going wrong and adjust around it. I started noticing some patterns in the AI that are… curious, at best. I think the intentions are good, but they also sow the seeds of player suspicion.

Imagine, if you will, a board with an obvious 3-skull match and an obvious 3-green match. They’re on opposite ends of the board and won’t disrupt the other if taken. No obvious cascades anywhere. Which would you expect to see the CPU take? My bet’s on the skull match. It should know if it takes the green match, I get the skull match.

(Are there situations where you might let the skull go? Sure. Maybe you’ve got Rock Solid and maybe you’ve got a troop that needs 3 green to get online. We’re not in that situation. This is just a plain choice between two matches.)

I’m finding the CPU skips skull matches an awful lot. When you do that, there’s always some small chance that the gems you displace cause a cascade + free turn to fall. It’s rare. It doesn’t happen more often than it should for the CPU, in my opinion. But here’s the thing: when it happens after the CPU appears to make a “stupid” move, I’m way more likely to cry unfair. It makes it look like the CPU had psychic knowledge of what would fall and knew the green match would lead to a free turn so it could get green mana AND a skull match.

It’s weird when, say, it takes a vertical 3-match underneath an obvious skull match. Why didn’t it take the skulls? 90% of the time I get the turn back and I’m “whew”. 10% of the time a skull falls out of the sky and I’m mad.

Why’s the CPU ignoring obvious skull matches? I think the devs might’ve done this because if it plays “too smart” players won’t be happy to lose more. But as I’ve outlined above, when it skips an obvious move and turns out better for it, a player’s natural inclination is to think something fishy’s going on.

The CPU shouldn’t be missing really obvious matches. It’s not just skull matches. I’ve seen it miss obvious 4 or 5 matches on occasion. I don’t think it’s psychically predicting skyfall: more often than not the only result is I laugh at how dumb it is. But one last time: when the CPU has really good luck after making a really stupid move it sticks out. Especially when you’re forced to play a lot of matches in game modes where you have comparably bad troops like Bounty. My games have been longer and I’ve seen a lot more of the CPU’s moves lately, and it’s just… weird.


#2

so youre saying it cheats :smiley:


#3

Originally the AI always had to take skull matches, but this led to skull baiting and being able to get out of trouble all too easily. They changed it, such that it was a percentage chance, which I think is different with different troops.

The same is true for whether a troop casts a spell, although I think they smartened-up the AI (conditional percentages) for casting converters such as Valk, Tai-pan, Gimlet etc where there is a higher chance of them not casting if there is not a certain number of the two target colours being present on the board.

Having said that, in 1 Trophy matches my sense is that the AI opts for skull matches at a higher frequency than in 3 trophy or GW matches.


#4

Slypenslyde shows why this is a no-win scenario for the devs. He knows that by letting the AI avoid skull matches, the devs made the game easier, on average. But the drawback is that when the AI wins despite making a dumb move, players feel even more stung.

It’s tough to be a game dev. I think I’ll stick to theoretical astrophysics.


#5

I guess you mobile / steam players dont know, but mr strange wrote the unity ai decision code ages ago when consoles gems first launched.

Later on, mobile / steam got the same ai, but i thought that was over a year ago.

Ai decisions are made through a random though weighted formula of possible moves.


#6

Obviously we didn’t know otherwise these threads wouldn’t keep popping up. I didn’t know it for certain, but seeing the occasional “mis-play” by AI I had suspicions of that.

On a side bar i’d definitely noticed for some time that the AI was less dumb about taking skull matches when entangled among other things.


#7

I play on PS4 and I have always noticed the ai will always go for the skulls and the 4 in a rows when I’m playing pvp. When it comes to things like raid, bounties and rescues sometimes the ai will miss the obvious matches but as the levels go up there are less and less opportunities for 4 in a row. The worst I think, however is the assistance (I have never turned it off so it is my fault anyway) where it ‘recommends’ your next move. Sometimes it will lead into a huge move for the ai if I don’t think through the consequences of taking the the ai’s suggestion. Again, that is totally my own fault for not turning off the assistance.


#8

Last big changes to AI I believe was introduced on consoles dec 2016 in patch 2.1.5
They had some posts from that period explaining how the ai worked

AI and Defense Team Improvements

  • Defending teams can now be customized to better suit your preferred play style
  • All players can pick preferences for their Defense Team’s gem matches
  • Test your defense teams from the defense tab in the PvP menu
  • AI will pick more appropriate board matches according to 4 or 5 or a kind, mana which your team still requires, or to deny any opponents from gaining mana
  • AI Troop Spells should cast at more appropriate times

#9

I’d buy the idea that defense and events have a different AI. I don’t think I see the same kind of behaviors when I’m playing PvP/GW.

But I’m not sure because while a 12+ turn game in Bounty can be fairly common, in most PvP/GW setups I’m trying to win before turn 8. So I’d also buy that I simply don’t see enough turns in the average PvP game to set up these scenarios.


#10

They behave differently, for now (Caution, AI is the wrong term here, as UKResistance correctly points out has not changed in 2+ years and has been emphasized and pointed out many times by the devs on the forums).

That’s because PvP still runs on the original vanilla system that is highly likely scheduled for replacement.

Given the expected arena/PvP revamp that may or may not be coming in 4.3 (should be mid-late March per normal quarterly patch schedules), I would highly expect that arena and PvP would adopt the new “invisible difficulty system” used in events and would be the standard gamemode moving forward. If so, I’m heading for the blast shelter before the bombs start falling on the forums.

In reference to your original post, most of that has been covered by others already in this thread. That said, as I’ve said before, the AI isn’t “broken” or “cheating” at all. Rather, the AI that hasn’t changed since 2016 is being artificially assisted by the invisible artificial difficulty system used in events (aka “the luck factor”). The “dumbness” of the AI becomes irrelevant when the difficulty system is artificially dropping in exactly what the AI needs to match, or circumvents the AI behavior entirely by dropping in entire sets of match 4s and 5s. Doubly confounding is when the AI makes a seemingly nonsensical move (likely in a nod to simulating “human error”, only to result in a 6+ chain of matches which frequently includes skyfalling skull matches out of nowhere.

Anecdotally, there appears to be a “safely valve” to protect newer accounts from this behavior that is toggled off in Raid/Delves at level 100. Unsure about Invasion/Bounty mode cut-off points, as troops don’t scale to 500 in those modes.

Easy enough to visually observe though, if one stalls through the early levels of each mode and avoids clearing each match quickly. The skyfall behavior at enemy level 20 is drastically different from the skyfall behavior at enemy level 200.


#11

Yeah I want to make clear here because people keep mentioning it: I don’t think the AI is cheating at all. To cheat, it would have to have some foreknowledge of what’s dropping.

What I’m saying is there’s some other mechanic that makes the CPU decide to “not see” some matches we’d generally expect to be the “best”. The obvious intent of that mechanic is if the CPU is too eager to take skull matches you can skull bait and for some reason that’s bad.

When it makes that wrong move, sometimes excellent things happen anyway. For it to be cheating, it would have to be making the wrong move because it will get a better result. I see it, more often than not, make a dumb move and get dumb results.

But psychologically, when you see an opponent do something really stupid then win, you don’t feel good. So this is on my list of “reasons GoW seems to intentionally work to make me feel bad”. Random luck is sure to reward bad moves every now and then. That doesn’t make it a feel-good moment that should be part of the game.

I still don’t know how much I buy this “luck factor” thing, but if I sidestep that I do agree that I feel like it’s a completely different game on turn 12-15 than turns 1-5.


#12

Ive notic3d that when I’m low on sigil the valraven is much more apt to depart quickly. Sometimes I can get past this and get the bugger but frustratingly sometimes I’m forced to wait or spend g3ms to proceed. ESPECIALLY when closing in on a major orb or vault key.
I’ve also noticed similar things to the OP especially concerning passing on easy match3s with resulting blessings … And many other little things. Especially extra turns that were not earned occasionally.
But for me, a refugee from the horror show that is LGOH… These are minor things overall. I think that the game is superb. I can spend if I want and the money goes so much further than… Ugh I won’t even say the name again.
All that said… I have read the other thr3ad recently about this and I kind of agree that maybe there is some kind of occasional thing…but if there is I believe it’s mainly to gently encourage spending? (Mainly the valraven exampl3)
If it’s actually som3 malfunction I feel based on what I’ve seen since October that the devs here are much better at being Not Asses than … Ugh you get the point.


#13

You said it yourself: you noticed more. It doesn’t have anything to do with the Valraven chances changing, but when you know you’re low on sigils, you’re paying more attention to whether or not it flees. When you look more closely for something, you’ll notice it more often.

This is called “confirmation bias” - peoples’ brains tend to pay more attention to things that they want to see, and remember things in a way that validates what they believed happened.


#14

The AI in normal pvp is different. When the devs introduced every other mode, there was a different AI attached.

The different AI is dense as concrete.

Yesterday or two days ago, the AI I was against was given a 3-match and a 4-match of skulls. It could have done both of them, of course. It chose the 3-match. I was in delve level 470, the boss room, meaning that I was Titan - and I had a barrier up. If it had chosen the way that got it both skull matches, it would have taken off half my hero’s health or worse. But no.

Regardless of the fact that this turned out to be very much in my favor, I don’t like seeing that! That’s the sort of nonsense that makes one feel bad for winning. So when this kind of crap happens and we lose, well …

Actually no, feeling bad for winning is even worse.

Put simply, high level delves and bounties and so on make me feel like I’m vsing lvl500 Mythic Ogres.


#15

Yeah this is roundabout my point.

I’d rather face an AI that makes consistently optimal moves (based on visible information). I don’t care if that means it’s susceptible to skull-baiting. I find that more pleasant and strategic than seeing the AI occasionally stab itself in the face.

I’m the dumb meatbag, I’m the one who should be self-face-stabbing.


#16

Careful with that one, optimal moves will make the game fiendishly difficult. Technically, the AI could evaluate each possible move in the blink of an eye, calculating deep cascade results and probabilities for favorable gems to drop in, then pick the best one. As a human, you’d have to contemplate each of your moves for several minutes to not get taken advantage of, and you’d still come out worse. I don’t think that’s the game most of us want to play.


#17

Go play the snore fest known as Chess.


#18

The vast majority of board states consist of a choice between two matches. When I’m in the state where there are 6-10 matches I already do slow down to try and figure out which is the “best”, using the same kind of “Well what might drop in” heuristic the CPU would have to use.

You’re vastly overestimating what an AI with no foreknowledge can predict, just like @UKresistance is making a false equivalence between my imaginary game and Chess. In Chess, you start with the same pieces and they don’t randomly get replaced every move. A CPU is readily able to chart millions of possible future game states and choose the most optimal. For GoW, since the board changes every move, that gets intractable very quickly. (In some other match-3 games, the board has more predictable patterns, but they aren’t head-to-head so it’s moot.)

In GoW, when I’m deciding between 2 3-matches, I’m trying to guess what 3 random gems will be selected from a set of 8 possibilities. This isn’t hopelessly difficult, but it’s also true I’m going to be wrong more often than right.

That’s why humans take the skull match in this situation. Statistically speaking, it’s as likely your 3 gems will be hurtful as helpful, so it’s best to take the move that does what you want predictably. It takes an excessively strained explanation to say, “Take the non-skull match, because your upper row has 4 purple gems separated by a gap and you have a 12.5% chance of a 5-match if you do this”. I still don’t think most players would agree “12.5% chance of 5 match” vs. “100% chance of skull damage” is a fair choice.


#19

The ai cant evaluate what move is better, it has no idea what the board looks like except for how many of each gem, it cant tell your troop health, it can only target randomly. It randomly makes a move, though the choices are weighted.
This was all explained by the devs in late 2016,and i cant search for ancient posts on my mobile.

And each troop has its own probability weights. Which i found surprising.


#20

Part Deux!?
More like Part 2000000 and DEUX!

I am both shocked and not surprised at all that these debates still rage today.
I will say what I have always said: Skull Baiting days were better. It added a “skill” element that “Human Players” could actually utilize to their advantage. I won many a battle in my early noob days to much overpowered AI teams, because I could bait them and buy myself enough time to get that final spell off.

Oh and hello Forums! :wink: