Cheating, terrible AI

Here’s an example:

5 Likes

As you get better troops, ascend/trait them and become more experienced with creating teams with better synergy you will find counters for pretty much anything the AI can throw at you. The more troops you gain the more variety of teams you can create. For instance I have about 20 plus different PVP teams which I probably use on a daily basis, depending on the opposition defense and my best team with the right abilities to counter them.

Freeze is a great trait to start off with to try and prevent enemy cascades. Also try and have somebody like Gorgotha (probably not at level 80) or other troops that reduce skull damage in 1st slot, this will help if enemy hits a few lucky skull drops.

If everybody won easily from the very beginning there would be little incentive to level up and improve your troops/teams etc. I’m currently level 1166, but no matter your level, in the end all of us encounter a battle now and again where you literally can put the controller down and watch as the AI just demolishes you (I’m looking at you goblin teams!).

The AI has been tested multiple times it doesn’t cheat against the players.

The AI doesn’t really cheat. What you’re referring to is the RNG, which is modified. Apologists like to throw out things like recall bias and the gambler’s fallacy, but the fact is that people, myself included, HAVE tracked things with listed probabilities and determined the numbers to be off. There are “hidden” difficulty factors in things like PvP, and absent blatant troop stat manipulation, which would be obvious, we have RNG manipulation, which is not so obvious.

This goes back at least to 3.1, if not earlier. It’s considerably worse as the game has moved much more towards a pay to win model, so the overpowered troops you face end a match very quickly when the RNG wills it.

It’s lazy programming, but I get it. Look at racing games. No, they can’t mimic smart humans (at least not without time and money) so they instead rely on rubberbanding and infinite weapons. This game is the same - the AI isn’t any smarter, but the subtle (or not-so-subtle) probabilities on Fizzbang, or Kraken, or whoever make up the difference. It just so happens to be very infuriating.

It’s the equivalent of playing chess against an AI rated 1200 that gets two extra queens and calling it a 2000 rated player.

2 Likes

THIS is the sort of thing I’m talking about. The RNG (random number generator) does not appear random at all. I’m not sure if the AI gets a higher probility of skull drops (and dooms) but I’d not be surprised. But what I’ve noticed is the number of times a “random” event skews toward the computer.

Having leveled up (better) troops can make you care LESS about the false “random” luck of the computer, but the effect is still there.

I have a VERY small sample set at this point, but I believe it has gotten marginally better with the LASTEST (yesterday’s) update, but it is still obviously skewed.

I’m tempted to record like 20 hours of play and take notes, but I’ve things to do. It would take 20 hours of play (obviously) but about 40 hours to analize and annotate. Arguably, you could take a smaller sample size, but the smaller you go, the less confidence you can have with the result. I’d look at: number of skull drops in favor of each player (AI vs player), as well as instant win 3-4-5 matches from drops, and cases where the drops or reshuffle set the player vs AI up with a skull match (or 2 or 3) before turn ends.

The GOW devs claim there is no such cheating, but the provided evidence is to a report they did before it was even called GOW. I expect it has since changed. (probably to try to make more money though “lazy programming” (I agree, much easier to twiddle “randomness” than actually code a smarter AI or design better gameplay that is fair, but more difficult).

Thanks everyone for your comments and insight on this.

Amadan, once again, nails it!

Our AI and RNG is never skewed in favour of the AI. We have spoken about this at length in the past and also run various tests to show you the results. If necessary I can dig up all the old threads. :slight_smile:

4 Likes

I’ve reached level 220 and I’m done with this. Deleted from my system. Too much cheating. Routinely the computer gets excessive skulls/dooms or endlessly perfect cascades. This is just too stupid to continue with.

1 Like

We are sorry to hear about your experience @zbret. We have rigorously tested our AI to ensure that it doesn’t cheat against the player, and we do our best to ensure that it is balanced for our player experience. As such, the AI is less lucky overall than the player, especially at lower levels.

We will be sad to see you go, but we wish you the best for the future.

4 Likes

The rage is strong in this one…

1 Like

I think what leads to this perception is a nice maelstrom of factors that could be easily alleviated in a special event I’ve always wanted. Jar of Eyes has really made it worse for me.

First: we’re encouraged to make teams that either explode most of the board, generate a large number of gems in specific colors, or convert gems from one color to another. When there are 6 possible matching colors, the odds of cascade aren’t as high. A converter aims to drop that to 5 possible matching colors, and generators like Divine Protector can push it down to 2. Also, many popular teams summon a storm.

So when you cast something like Divine Ishbaala’s ability, you usually get a free turn and a board that’s mostly void of skulls, yellow, or purple. Now only 3 colors are left on the board. If you then cast Infernus, some gems are exploded. You’ve got a pretty good chance of a cascade there, but if you don’t get one, you just passed that luck to the enemy. It’s worse if you trigger a fire storm: now the board has effectively 3 colors and you’ve tilted the RNG to create one of them more frequently.

So without going into a lot of other examples: we’re encouraged to create teams that create very cascade-prone boards. If our luck fails, we pass that cascade-heavy board to our opponent. This leads to a lot of stories like, “I fell behind, made a lucky match, cast Ishbaala, cast Infernus, then of COURSE the CPU got 6 cascades, 2 free turns, and I was left with only Ishbaala.” If your luck is even a little bad, this is probably the most statistically likely thing to happen.

I have always felt the antidote is to create the opposite badfeel situation. I want a mode without free turns and a limit on cascades. In this mode, if either player reaches 3 cascades, the board is cleared and the turn is given to the next player. No free turns is what it says: not even goblins get a free turn in this mode. Storms should also not be allowed to happen. Exploders and generators would have a cap on how many gems they could effect.

This would be a slower, more grindy mode, but also removes a large portion of luck from the game. I think if a lot of people played it, they’d decide they don’t mind free turns and cascades as much. Or, they’d come to the “no free turns” mode for a break if they get frustrated.

Anyway, my main points are:

  • I think whether or not the RNG is fair, the way “good GoW teams” operate creates situations that don’t feel fair when they unfold.
  • I wish we had a mode where many of those factors were mitigated.
4 Likes

In the grander aspects of things, I don’t mind converters so much, at least you can “plan” for them. Say you use Valkyrie and there is one blue and 3 yellows in a straight line, you fire her off and you get an extra turn, and you plan your moves from there.

Exploders and color creators (like DP) lose all that. Then, it’s just about whatever falls in you or the AI’s lap. And, with things like storms, it can either go very good or very bad.

All in all, I do think there is too much RNG in this game. I feel true strategy went out the window a long time ago. I don’t have a problem with (planned) extra turns*, like I mentioned, as they still take strategy and I like converters, overfall. The other stuff, though… can get quite messy.

*when I say planned extra turns, I don’t include Goblins, by the way. I think Goblins, as they are, are ridiculous and I don’t like them. I only use them when I have to, like that one Raid week we had for their kingdom.

1 Like

I don’t wanna brag, but I cheat way more than the AI does.

1 Like

I’m also level 80 (i read somewhere people like it when they can relate) and I think it’s not that weird for the AI to have some advantage cause you’re playing vs an AI…

He/She/It doesn’t really have a brain like you do. Even with a few matches in his/her/it’s favor you should still be able to win 80-90% of your battles easily.

Wouldn’t be fun if they didn’t put up a good fight.

1 Like

The biggest problem with the AI and RNG is streakiness, in my opinion. If you are looking only at raw percentages over time, it looks like the AI is “balanced” properly. For example, maybe you ran 50,000 trials and got 25,000 successes with gobchompers. That looks balanced on the surface.

But then, if during those 50,000 trials, you have 3,000+ instances of a fail-streak of 5+ occurring, then something is extremely wrong. Because there should only be a 3% chance of getting the same 50/50 result 5 times in a row.

And that’s usually what, in my opinion, usually causes players to perceive the AI as cheating. Or at least is a major contributor to it. It’s not just that the player is sensitive to a bad luck result occurring; but that WHEN bad luck does occur, it comes in clusters most of the time.

You don’t lose a match because the AI got a statistically average cascade 2 times during a match. You lose a match because the AI got 4 consecutive sets of 3 skulls falling into the same spot on the board and instantly matching to each other for 4 separate free attacks. (12 1/6 chance events occurring consecutively should have astronomically low odds. So low that I should never have seen this happen more than once in a month, but it usually happens weekly at least.) Or because you failed with your gobcompers 5 times in a row, even though while there is only a 3% chance of that happening, statistically, it reliably happens around 10% of the time.

It really doesn’t matter if the numbers “balance out” in the long run when streaks occur so far outside of the normal realm of probability that events with extraordinarily low odds of happening are downright common due to the RNG’s habit of picking the same result several times in a row.

2 Likes

This thread is edging dangerously close to Terry Pratchett “one in a million shots succeed 90% of the time” territory.

5 Likes

Yes. It does. RNG balancing over the long term is what statistics is all about. You want structured randomness where 50% happens 50 in 100 times. That’s just not how the game works. The game has actual randomness, which has to be observed on the long term.

1 Like

If I made a coin flip program and it consistently delivered results of 50 heads in a row, then 50 tails in a row, then it is fundamentally broken. an RNG that does not follow statistical probability for streaks as well as singular events is a bad RNG.

3 Likes

It has been proven by a resident stats poster that the RNG has a tendency towards the streakiness you are describing. But I’d save my breath. The bulk of the forums has decided that so long as “in more games than I will ever play” it balances, everything is perfect, because numbers and facts are scary. You’ll have better luck convincing people universal healthcare is a good idea.

I really enjoyed that person’s stats posts and they stopped shortly after:

  • People complained TDS seemed streaky.
  • He proved TDS was indeed streaky.
  • People argued loudly he was dumb and his sample set was too low.
  • The devs nerfed TDS, citing “it’s too streaky” while also denying the statistical analysis.
  • Everyone continued insisting “it’s not streaky at all, if the devs agreed they’d do something about it”.
  • If you point out they did do something, that’s fake news from the Deep State and you’re an idiot.
2 Likes

There are certain industry standards and established methods to evaluate general quality of pRNG engines. One example is a quite popular set of Diehard tests Diehard tests - Wikipedia and there are a few other established by the regulators of gambling industry. Altering the random selection process of the gambling game might be perceived as at least borderline grey zone. Thus, in general, anything that is claimed to be random should indeed be random or as close to random as reasonably possible without inflating the costs.

I can elaborate a bit on the subject. For example, to get a license to operate a gaming site, RNG has been certified by an independent third-party. The third-party testers will analzse the source code and run tests (usually some Diehard variation) to ensure that generator is random. Some poker sites may have details of the certification of their RNG : https://www.pokerstars.com/poker/rng/

In Nevada, there is a specific law that regulates quality of RNG: https://gaming.nv.gov/modules/showdocument.aspx?documentid=3450

There is a lot of info out there on the subject, it is very explored area with lots of statistical science on the subject.

In general, GoW client uses Mersenne-Twister pRNG. It is the same pRNG as used in most recent versions of MS-Excel starting from 2010. Mersenne Twister - Wikipedia

It is kind of OK-ish with some external support, which is mandatory, not too good and not too bad, kind of passes most of the tests and fails in some of them but is generally considered as acceptable. IDK what the server uses for its pRNG. When I did some testing, I used the most simple Runs test ( Wald–Wolfowitz runs test - Wikipedia ) which is a very basic standard test that does not require primary output of the generator. It is a very powerful test, so, if something is not random according to it, it is most certainly not random according to other more complex tests.

It should be noted that GoW generally “masks” its random in many cases due to additional perturbations in the output numbers which sometimes makes it hard to actually test randomness. So, the only reliable way to it seems to create a simulation and then compare simulated numbers with actual numbers. Sometimes, some types of transformations are so complex that even if the primary output is random, final results of the transformation might appear as non-random due to multiple interference from multiple sources. Since even if one source of interference is non-random, that may create some non-random events that won’t pass the test even though the majority of the input is random.

There are also methods to evaluate streakiness. Excessive streakiness will make pRNG fail the randomness tests. So, some streakiness is expected and acceptable.

Sorry, it is a bit too general and too long, well, or maybe too short. :slight_smile:

7 Likes

It does not make me feel more confident in the game’s RNG to know that it uses a pre-built RNG as a base which is “kind of okayish” in testing, then further modifies it to be even less random…