Cheating AI [Merged Thread]

DocT

I have done a short mathematical study on the statistical and probabilities of the Cascades that you're AI gets on a regular basis. it always seems to know exactly what play to make in order to get the best results to completely wipe your team out. for this to happen over and over again is statistically impossible if your AI is truly random. I suggest fixing it before you end up having some kind of class action lawsuit filed against your company. you don't have to be a math professor to calculate the odds of the Cascades that the AI gets on a regular basis. I suggest you reprogram and recode your AI to be truly random and not biased and cheating like it is now. I was a math major in college and I know that what I am seeing and what I have observed is statistically impossible based on true random drops. I highly suggest you fix it and I'm sure there are others out there that would agree with me 100%.

MarcusGraves

DocT wrote:

I suggest fixing it before you end up having some kind of class action lawsuit filed against your company.

LOL yeah okay, want a game with actual intentional and dev admited cheating AI go see Gems of War, but you're silly if you think you could actually sue over AI design in a game.

AI goes for it's best match based around strongest colors or if there's a 4/5 match on the board which they will ALWAYS take. I've been playing this for around 3 years and I can easily predict 90% of the time what move AI is going to make next on the board just by looking at it, YOU CAN ACTUALLY LEAD AI INTO BAD MOVES.

AI getting big cascades really is accidental or you as the player lead them into a potential cascade and didn't realize it, it's not what you wanna hear but other long time players are going to tell you the same things. Saying they're other people that will agree with you 100% is kinda a no duh just because there will always be people like that but it doesn't mean they're right either.

Also "short mathematical study" will lead people to ask for you to submit such data, which is weird for a math major to not give to back up their analysis.

DocT

Short study = 500 consecutive AI moves leading to cascades of 3 or greater. 153 times out of 500 times or nearly 1/3 of the time AI tile turned a match that cascaded supposed "random" creation of new tiles that were in the exact right place at the exact right time to cause this cascade of 3 or more times. There's 7 colors. That means that in order for that to happen, for each cascade, there's a 1 in 7 probability that the right color will be produced to complete the proper combination to continue the cascade. Just for it to happen ONCE. For it to happen 3 times in a row means it is making a 1 chance in 7 match 3 straight times 1/3rd of the time. I challenge you to pull 7 different cards from a deck and with them turned face down, pick the card you want from those 7 by guessing 1/3rd of the time. Let alone doing it 3 times in a row or more 1/3rd of the time. If you could do that well on a regular basis you'd be a bazillionaire going to Las Vegas. That doesn't even address the issues of how many times I've caught the AI cheating using powers it doesn't have the AP's to do or the spamming of team ups with no silver AP's at all. But you go along and keep thinking you're not playing a rigged game if you want if it makes you feel better.

We_are_Venom

MarcusGraves wrote:

DocT wrote:

I suggest fixing it before you end up having some kind of class action lawsuit filed against your company.

LOL yeah okay, want a game with actual intentional and dev admited cheating AI go see Gems of War, but you're silly if you think you could actually sue over AI design in a game.

AI goes for it's best match based around strongest colors or if there's a 4/5 match on the board which they will ALWAYS take. I've been playing this for around 3 years and I can easily predict 90% of the time what move AI is going to make next on the board just by looking at it, YOU CAN ACTUALLY LEAD AI INTO BAD MOVES.

AI getting big cascades really is accidental or you as the player lead them into a potential cascade and didn't realize it, it's not what you wanna hear but other long time players are going to tell you the same things. Saying they're other people that will agree with you 100% is kinda a no duh just because there will always be people like that but it doesn't mean they're right either.

Also "short mathematical study" will lead people to ask for you to submit such data, which is weird for a math major to not give to back up their analysis.

I've seen the computer make matches that didn't align with its "best interest" that ended up creating cascades of ludicrous amounts. For instance, Ares was the only named character on the team and skipped a 4 match green for a 4 match purple on the other side of the board that resulted in my storm going from 3k to dead just after I finished getting my blues for wind storm. This was today, and is not an isolated incident.
If the AI played as you described, it would "hoard" colors starting at its strongest and checking down, but it doesn't. Also today, Ragnarok had 11 green with 3 green match available. The AI skipped the green and immediately I knew I was screwed because it wouldn't skip it without reason. Cascade, 5 match black, 3 drop green, Venom stun, devour on 5 cap to kill and Rag ended the match with Godlike power.

MarcusGraves

Article explaining game's AI: http://www.demiurgestudios.com/engineer ... -opponent/

You're free to not believe me, but if the game was rigged in the AI's favor then I (and many other players) would be losing almost all the time, and seeing how I'm not and able to predict what moves AI is going to make because I know AI's simple habits, I'm going to keep doing my chain winning and gathering my rewards.

If they were to tune the AI like in the 2nd half of the article, you'd be crying yourself to sleep every night, the AI even being able to recognize L shape 5 match would make you wanna punch yourself in the face repeatedly.

We_are_Venom

MarcusGraves wrote:

Article explaining game's AI: http://www.demiurgestudios.com/engineer ... -opponent/

You're free to not believe me, but if the game was rigged in the AI's favor then I (and many other players) would be losing almost all the time, and seeing how I'm not and able to predict what moves AI is going to make because I know AI's simple habits, I'm going to keep doing my chain winning and gathering my rewards.

If they were to tune the AI like in the 2nd half of the article, you'd be crying yourself to sleep every night, the AI even being able to recognize L shape 5 match would make you wanna punch yourself in the face repeatedly.

In my opinion, it would be precisely because of the shortcomings of the AI they would skew tile drops in favor of the computer team. Also, it seems silly to use a source from those whom created the game, and the accusation is levied, as evidence that there is nothing to what is being said.

JVReal

We_are_Venom wrote:

I've seen the computer make matches that didn't align with its "best interest" that ended up creating cascades of ludicrous amounts. For instance, Ares was the only named character on the team and skipped a 4 match green for a 4 match purple on the other side of the board that resulted in my storm going from 3k to dead just after I finished getting my blues for wind storm. This was today, and is not an isolated incident.
If the AI played as you described, it would "hoard" colors starting at its strongest and checking down, but it doesn't. Also today, Ragnarok had 11 green with 3 green match available. The AI skipped the green and immediately I knew I was screwed because it wouldn't skip it without reason. Cascade, 5 match black, 3 drop green, Venom stun, devour on 5 cap to kill and Rag ended the match with Godlike power.

If the AI had to choose between two different Match 4's, it may have chose which of the colors would trigger an ability or has the highest match damage. If Ares was the only board mover, Purple is not his strongest color, but perhaps it was the strongest color overall and the strongest color of one of the goons. Perhaps it would give the goon enough AP to do one of it's cheap CD tile drops.

I too was surprised to see Ares not match a yellow or green when I was afraid he would match away my IM40 CD tile, and instead go for a purple... but then I saw he was with a goon that may have had that strongest color.

The AI doesn't care that one is a Hero character and two are Goons, it just knows here is the team, here are the match values and here are the ability costs (and will only execute them in order they are listed).

DocT wrote:

Short study = 500 consecutive AI moves leading to cascades of 3 or greater. 153 times out of 500 times or nearly 1/3 of the time AI tile turned a match that cascaded supposed "random" creation of new tiles that were in the exact right place at the exact right time to cause this cascade of 3 or more times. There's 7 colors. That means that in order for that to happen, for each cascade, there's a 1 in 7 probability that the right color will be produced to complete the proper combination to continue the cascade. Just for it to happen ONCE. For it to happen 3 times in a row means it is making a 1 chance in 7 match 3 straight times 1/3rd of the time. I challenge you to pull 7 different cards from a deck and with them turned face down, pick the card you want from those 7 by guessing 1/3rd of the time. Let alone doing it 3 times in a row or more 1/3rd of the time. If you could do that well on a regular basis you'd be a bazillionaire going to Las Vegas. That doesn't even address the issues of how many times I've caught the AI cheating using powers it doesn't have the AP's to do or the spamming of team ups with no silver AP's at all. But you go along and keep thinking you're not playing a rigged game if you want if it makes you feel better.

Did you also track your own moves? Your own cascades? Every move changes the layout of the board, not just the AI's move. Every tile replacing your move impacts the AI's move and board layout. Did you track how many times your move set up a match 3 or match 4 that led to a cascade? Did you check and track the AI AP after every turn or during each turn? The AI doesn't add the AP to the pool until the end, so if the AI needed 5 Blue for Daken and his turn started with 4, had SW who generated a blue tile that created a blue Match 3... if you could pause the game, the display will still say 4 AP, but internally it would know it has 7 Blue, and will then use the blue ability despite the display saying 4. The AI doesn't stop to think so the AI doesn't refresh the AP display in between moves like it does for the human player.

A one sided study, proves nothing. Spend the next 500 battles tracking every move, yours and theirs, ap gain, ability use, which abilities, tile conversion due to abilities, charged tiles, etc. There are too many variable out on that 64 tile board for you to track only the AI and say that it cheats.

After you track that 500 battles, play 500 more making the same decisions the AI does. Using the logic in the link provided above by someone else, play with that logic, and see how you do compared to the AI.

Every tile only has 7 choices of what it could be, and I believe the TU tiles (the silver ones) have a lesser rate of drop than the rest. It can't change the position of the other tiles, but know that a vertical clear and the dropping tiles have a better chance of generating another match (and a subsequent cascade) than a Horizontal clear (from my own observations). This is probably because a vertical clear has an entire row of randomly generated tiles drop in where the other ones were, with possible matches on either side. The Horizontal row disappears, the tiles drop down and the new row appears at the top with only one side matchable.

I think you would have to recalculate the odds every turn because the board is a new configuration every time. Do you credit the cascade or lack of cascade to the shifted tiles as a result of your move, or to the newly generated tiles as a result of your move?

AI cascades are ridiculous and do seem to happen often, but we do have our own fair share of cascades. If the AI and player had to flip a coin at the beginning of each match to see who goes first... would the cascades change? Would they even out?

I look at it like this: The AI plays with constant rules, gets constant results. We do not play constant, we play randomly. We play variably, we will sacrifice a move the AI would take for a future objective or for an L-shaped match 5. Our results may vary. We are the random variable that works against us.

Finally, there is no true random when it comes to a computer program. Only a program that tries its best to simulate randomness. The RNG that this game uses is a bit faulty and many many people would agree with you that the RNG could use some tweaking. When it can drop 7 black AP down a column, that's a problem no matter who it favors.

Another experiment you could do based on your own illustration, is take 7 cards, shuffle them, draw a card, and notate it. Put it back in and shuffle again, draw a card again. Track that 500 times and see how many times you draw the same card in a row. That would only track 1 tile location on a board of 64 instances of the same thing. To account for the reduced odds of TU tiles, you'd have to only insert card #7 something like 3 out of 5 times you shuffle.

Nightglider1

DocT wrote:

I have done a short mathematical study on the statistical and probabilities of the Cascades that you're AI gets on a regular basis. it always seems to know exactly what play to make in order to get the best results to completely wipe your team out. for this to happen over and over again is statistically impossible if your AI is truly random. I suggest fixing it before you end up having some kind of class action lawsuit filed against your company. you don't have to be a math professor to calculate the odds of the Cascades that the AI gets on a regular basis. I suggest you reprogram and recode your AI to be truly random and not biased and cheating like it is now. I was a math major in college and I know that what I am seeing and what I have observed is statistically impossible based on true random drops. I highly suggest you fix it and I'm sure there are others out there that would agree with me 100%.

AHAHAHAHAHAHAHAHAHAHAHAHA! Thanks for the best laugh I've had in a long time.

wirius

You may have been a math major, but I doubt you took statistics. You're either an amazing troll, or need to do something actually worthwhile with your life to fulfill your need to be special and significant in the world.

dsds

This is what really frustrates me. When people go into personal attacks, they should be banned. Do not allow them to continue to do it and then lock the thread. Because this only ends up being buried once it is locked.

Neither side can provide actual proof to support their side. However one side is using personal attacks. And I think that counts for cyber bullying.

Don't get me wrong, I don't think it is entirely in the AIs favor either. But I do see a lot of cascades going on for me and the AI and it's abnormal considering the odds. I am no statistics major, but doesn't the probability get exponentially higher each additional consecutive cascade? This is if a tile drops and matches existing tiles on the board. If it dropped three of the same color on consecutive cascade, the odds of that happening are even more slim.

Yes, each drop is a new probability and doesn't count the previous ones. But a 4 consecutive cascade should be quite rare, yet we see them all the time. Again I've seen them on my turn too. But the frequency of these crazy cascades for either me or the ai is ridiculously high. If someone can give formula and calculate the odds of a having 4 consecutive cascades. It's in the hundreds for a tile drop to match existing tiles. Then if a drop of 3 identical colors, the number would be even higher.

PeterGibbons316

dsds wrote:

This is what really frustrates me. When people go into personal attacks, they should be banned. Do not allow them to continue to do it and then lock the thread. Because this only ends up being buried once it is locked.

Neither side can provide actual proof to support their side. However one side is using personal attacks. And I think that counts for cyber bullying.

Don't get me wrong, I don't think it is entirely in the AIs favor either. But I do see a lot of cascades going on for me and the AI and it's abnormal considering the odds. I am no statistics major, but doesn't the probability get exponentially higher each additional consecutive cascade? This is if a tile drops and matches existing tiles on the board. If it dropped three of the same color on consecutive cascade, the odds of that happening are even more slim.

Yes, each drop is a new probability and doesn't count the previous ones. But a 4 consecutive cascade should be quite rare, yet we see them all the time. Again I've seen them on my turn too. But the frequency of these crazy cascades for either me or the ai is ridiculously high. If someone can give formula and calculate the odds of a having 4 consecutive cascades. It's in the hundreds for a tile drop to match existing tiles. Then if a drop of 3 identical colors, the number would be even higher.

The problem here is that no one ever considers their own impact on the game. People subject to a statistically significantly high number of cascades are bad players. This isn't meant to be a personal attack, but sometimes the truth hurts. I regularly make moves that trigger 4 or more cascades intentionally - because I can see them on the board. This isn't uncommon, or a sign of good luck, it's just smart play. The opposite can also be done to reduce the risk of AI cascades based on the state of the board and what tiles could potentially drop into the board. Here's a board example....










Here I can take the red match in the middle, which is more likely to result in the AI getting a purple match-5 which will necessarily be followed by a yellow match 4 and who knows what else. OR I can take the blue match right next to it resulting in a red match 3, purple match 3, black match 3, and green match 3....just from what is visible to me on the board. Same board - same test sample, but two completely different results based on play style.

udonomefoo

PeterGibbons316 wrote:

Here I can take the red match in the middle, which is more likely to result in the AI getting a purple match-5 which will necessarily be followed by a yellow match 4 and who knows what else. OR I can take the blue match right next to it resulting in a red match 3, purple match 3, black match 3, and green match 3....just from what is visible to me on the board. Same board - same test sample, but two completely different results based on play style.

I love gambling on that vertical match 5 drop. I'll take it almost every time. And I get burned by it a lot. I'm ok with that.

PeterGibbons316

udonomefoo wrote:

PeterGibbons316 wrote:

Here I can take the red match in the middle, which is more likely to result in the AI getting a purple match-5 which will necessarily be followed by a yellow match 4 and who knows what else. OR I can take the blue match right next to it resulting in a red match 3, purple match 3, black match 3, and green match 3....just from what is visible to me on the board. Same board - same test sample, but two completely different results based on play style.

I love gambling on that vertical match 5 drop. I'll take it almost every time. And I get burned by it a lot. I'm ok with that.

That's fine.....as long as you don't come and complain about cheating AI after the fact.

For the record, if it is on the top row, or 3rd row down you should take it, if it is in the second row as shown you should not. I'm speaking generally here, of course there are other situations where it might make sense to gamble, or in a DDQ4 where you are in a hard denial mode you might just take only the purple match-3 on the left to avoid the possibility of the AI matching the 2 on the right.

Uthgarprime

Here's the thing, 3, 4 and 5 sets fall too often for each tile to be generated randomly. Instead I would hazard a guess that there is a calculation done based on the available space. Before any tiles drop a percentage is rolled. I don't know what the numbers would be so I will make some up. Lets say if 1 to 80 is rolled x number of random tiles drop. 81 to 90 a match 3 drops. 91 to 98 a match 4 and 99 or 100 a match 5 drops. So I don't think each tile is generated randomly every time. Hopefully this sort of made sense. Cause statistically 5 of the same drops way to often in this game for each tile to be generated randomly. Whether the AI cheats or not is another question but this theory may explain some of the "non randomness" of things.

PeterGibbons316

Uthgarprime wrote:

Here's the thing, 3, 4 and 5 sets fall too often for each tile to be generated randomly. Instead I would hazard a guess that there is a calculation done based on the available space. Before any tiles drop a percentage is rolled. I don't know what the numbers would be so I will make some up. Lets say if 1 to 80 is rolled x number of random tiles drop. 81 to 90 a match 3 drops. 91 to 98 a match 4 and 99 or 100 a match 5 drops. So I don't think each tile is generated randomly every time. Hopefully this sort of made sense. Cause statistically 5 of the same drops way to often in this game for each tile to be generated randomly. Whether the AI cheats or not is another question but this theory may explain some of the "non randomness" of things.

Can you please quantify "way too often"?

Based on the data you've collected exactly how many times out of 10,000 were 5 tiles of the same color dropped in an empty row or column? A developer has said "Each new tile is selected at random, individually" and I'm just wondering what evidence you have that would suggest that statement is not true.

Uthgarprime

Its not hard to figure out, look at the number of times per game that a 5 or 4 set of tiles falls from the sky. I have no exact numbers but each time a row or column falls, the game generates the tiles. If there is a 1 in 7 chance of a certain color falling that means its very small odds that the same color would fall 5 times in a row. While all them are "rolled" separately there is still a probability of a given color and whether or not multiples of the same color will fall in a row. Like a flip of a coin, its a 50% chance a given side comes up, but its a smaller chance that the same side shows up twice in a row even though its the same odds of each side each time. now multiply that out by 7 sides and 5 flips in a row. It should be 1 drop every several thousand that a set of 5 drops but I usually see it once a day and there are not thousands of drops for me in a given day cause I don't play that much. And remember its only the drops that a set of 5 could even fall in cause its usually just 3 tiles. I don't know the exact math, but it just doesn't add up using basic math, I am sure someone will come on here and tell me I don't know math or statistics like they usually do but it shouldn't be that often.

Uthgarprime

And just because D3 says its so doesn't make it true. Like when they said they upped the hp of all the non played characters to make them more viable. less then a handful were more viable cause their powers were still bad. They really just upped the hit points to make the games take longer so our characters would take more damage thus making us buy more health packs.

PeterGibbons316

Uthgarprime wrote:

I am sure someone will come on here and tell me I don't know math

No need.....

Uthgarprime wrote:

I don't know the exact math

Don't you think there is a problem when you admit to not understanding what should be happening, but then accuse the game of cheating based solely on a feeling of what you think should happen?

I'm no statistician, but I am an engineer, and I do have a pretty firm grasp on math and statistics, so I'll help you out......

There is roughly a 15% chance that any given tile will be a particular color, and about a 10% chance it will be a TU (this coming from that article I linked above, and it's obvious to me that TU tiles appear slightly less frequently than colored tiles). That means that for any given 5 tiles to be red for example you have 0.15^5, and for any 5 tiles to be of the same color you have that same number times the number of colors or 0.15^5*6 and if you add in the TUs you get 0.15^5*6+0.1^5 which comes to 0.00047. So those are the odds that any 5 new tiles will match. Now each time you drop a new row or column there are 4 different locations where a match-5 could reside. So for a vertical drop you could have a match-5 that lands at the very bottom, you could have say a black tile at the bottom and then 5 reds, or 2 blacks and then 5 reds, or 3 blacks and a match-5 at the top. So you multiply that number by 4 to get the odds of a match-5 appearing anywhere in the column/row to give you 0.00186. Also, a match-6/7/8 would also trigger a new column/crit so you add those odds in (which don't change it by much since it's even harder to match more than 5) bringing your total odds to 0.00192 or about once for every 500 new rows/columns. How frequently you see new rows/columns depends largely on your play style, but once per day definitely doesn't seem outside the realm of possibility.

The odds that a match-4 or better appears? better than 1 out of 100 drops. A match-3 in a new column? about every 10 times. A match-3 will appear in the 3 spaces that were just cleared about every 50 times - not counting surrounding tiles.

So that's the math with a random distribution, and based on my play it seems pretty reasonable to me. If you are going to suggest otherwise you will need to provide data that shows these kinds of tile drops happening significantly more frequently.

EDIT: If my math is wrong someone PLEASE let me know.

PeterGibbons316

So I ran through the DDQ today and it was pretty quick - 10 minutes, and I counted 30 match-4/5s that cleared a row/column. So that's 3/minute. At 1 match-5 drop for every 500 you would expect to see it happen about once every 3 hours of play. Also, I was thinking that if you make a match-5, you create a critical tile, meaning that you only need 4 of the same color to fall into place, not 5 - this would obviously increase the frequency, but to determine how much you would need to know how many match-5s come for each match-4 which is harder to get a good estimate on.

Uthgarprime

So tired of auto kill cascades, why can't I kill level 250 plus ai characters in about 3 turns. I mean its so fun to lose before you even start the match, Why don't more companies do that since its clearly the best feature of the game. 2 times already today my game was over before I got 10 turns one time and 3 turns another. AI auto cascaded both times the 3 turn one killed IF the 3rd turn just with cascades, the next game OML was killed on turn 10 cause the auto cascade gave the ai enough ap to fire off all its high damage powers in one turn. I ask again how is this in anyway fun? No one at demiurge has ever answered this question cause they like to frustrate players. Oh and I know I have never won 2 games in a row against level 250 ai's that quickly.

udonomefoo