A Random Comment

[Camate]

Ever experience a time crafting when you are at 90% and still manage to blow it up? Three times in a row?

I know I’ve experienced some unlucky situations like this and thought to myself “Oh c’mon! No way!” all the while thinking the RNG gods hate me. If you’ve ever had times when you thought the same and questioned whether there something wrong, you are in for an interesting read.

Grab a drink and snack, because it’s time for a mega-knowledge bomb from Hiroshi Minagawa.

Hello,

Since this was brought up in the UI forum, even though I’m not in charge of it I will go ahead and comment about this a general discussion point.

I checked up with our main programmer “K”, and for FFXIV it seems like we are using 3 famous random algorithms different depending on the application. Naturally, everything is processed server-side, and the random seed differs for each process.

So in response to the question if there is a bug with the coding that generates the same random number if you continuously press an action, the answer is that the system implemented is such that this type of problem will not occur.

There were older games where you could use certain patterns to your advantage to beat the game, but in recent days there aren’t any games like this anymore. Probably.

However, the rate (random) process in games from very long ago and up until now -- not only for FFXIV -- have continued to be doubted by players.

From my perspective it’s just a clear cut case of chance, but this alone may not be convincing enough, so I will try to explain about this a bit.

This discussion is not particular to FFXIV and it is more about why random number programs are easily doubted.
(Of course there have been a number of cases where there were bugs…)

So to start off, I’ll lay out the essential points quickly.

A. The standard random number programs used by FFXIV, and other modern games, generate practical “true random numbers”.
B. With a range of trial numbers that can be understood by a person, because it’s a true random number bias arises.
C. Since our brains are very adept in their capability to detect bias as abnormal or a singularity, we perceive true random numbers as abnormal.

I’ll explain about each point above individually.

First, I’ll start with the program for generating random numbers.

FFXIV and other modern games use a random number algorithm (properly known as a pseudorandom number generator), which was contrived by some really good mathematicians and computer science people, and it was tested for a number of fields and is used as a solid algorithm.

As an example, I’ll explain a little snippet about the periodicity of the random number algorithm known as Mersenne twister.
“The algorithm provides a super astronomical period of 2^19937 – 1 and 623 dimensional equidistribution up to 32 bits accuracy.”

Got it? No? That’s okay.

Random number algorithms are part of a field that was built by the results of pure mathematics. As long as we the users understand that the random numbers generated are correct and practical that is more than enough.

★Those of you who know about random number algorithms and examples of problems in the past that arose in games, you’re probably wondering about something I didn’t touch on above. I will add an explanation to the end of this post.


Next is about “true random numbers”.

When it comes to random numbers there are two different types with different distribution patterns: uniform random numbers and normal random numbers. As the number of trials increase, we start to converge on the ideal condition.

Conversely, if the number of trials (sampling number) is low, bias arises in random numbers.

As humans, when it comes to the number of trials in games and such that we are able to grasp, it’s normal for us to perceive bias in the random numbers.

In other words, the premise that it’s strange that there is bias in random numbers is mistaken.

Of course, with hundreds and thousands or billions of trials, we will approach a distribution that is uniformly random and the bias will disappear. Unfortunately, this is impossible for a human to actually observe and experience.

Due to this, every time you observe bias arising from a limited number of trials you can’t help but think that something’s wrong.

Finally, onto our cognitive fallacies.

Humans are able to instantly determine and intuit various things. It’s often said that our brains are really good at this but computers have trouble. On the other hand though, there is an aspect to this ability that is weak, and when this is exploited it opens us up to tricks and deceit.

The pattern of deviation in judgment and illogical interpretation is known as cognitive bias.

A leading example of where cognitive bias arises is the Gambler's fallacy.

This states that when a person observes multiple events over the course of time, they will begin to expect that the results in the future will be affected by what happened in the past, and will either feel convinced that the results were due to a cause-and-effect relationship, or have a feeling of strangeness.

When you flip a coin 5 times in a row, the rate in which you land either heads or tails for the 6th flip will be 50%. However, our gut feeling says “this time for sure!” and we modify this with an expected value, and as a result when it lands heads we think “No way!” or when it lands tails we think “Just as I thought!” This is cognitive bias.

As I stated in b) above, the feeling that something is strange even though it is not is our brains are trying to create a cause-and-effect relationship that we anticipate, but the rate behaves in a completely unrelated manner.

(I’m writing this in a definitive matter, but please be aware that I am regurgitating what I read from a book.)

Cognitive bias is similar to a bug in humans’ cognitive capability, but it seems like it’s a system that was acquired for our survival to decide things efficiently in times of emergency.

The whole topic of cognitive bias is really fascinating, so if you are interested I recommend reading
this book.
　
This pretty much sums up why we often think something is up with the probability we encounter in games.
　

Bonus info: An aspect I didn’t touch on earlier in regards to random number algorithms.

Amongst the random number algorithms, one method that is often adopted that has a high cost to efficiency ratio is the linear congruential method; however, it’s well known by implementers that there is an issue where large bias arises.

In fact, there was a game that mistook the use of this function, and as the sample size grew larger the more bias increased. Meaning that just because it’s a well-known method doesn’t necessarily mean that it’s proof of it being accurate.

Additionally, even if a random number sequence is generated properly, depending on how the application uses this value, ultimately there may be cases where strange patterns arise.

So with all of this said, this is not one big post about the evidence of the accuracy of XIV’s various probability calculations. As I mentioned that the beginning of this post, the latter portion of this long comment is mainly just chitchat, but in response to the question as to whether there is something wrong with the probability in XIV, the answer is no, there is nothing wrong. 　

[Parz3val]

True RNG with regards to Atma is utterly inequitable. One person spends 10 hours, gets all 12; another 50 hours, has only 1. SE please change. Discuss.

[Evangela]

fix Atma if you have time to fix something else.

[MasamuneBranford]

Would be nice if SE came forth and said what the Atma drop percentage was - whether Gold, Silver, Bronze has an impact - etc.

[Parz3val]

Would be nice if SE came forth and said what the Atma drop percentage was - whether Gold, Silver, Bronze has an impact - etc.

They've stated it's pure RNG. I know friends who've spent 60+ hours & are only a couple or so in.

SE, PLEASE - here forward nerf Relic Chains two steps back from current (e.g. Atma should've been eased up at Novus release; Animus should at next release)

[IndigoHawk]

Human use "better than safe sorry" and "common sense" statistics, not "A causes B".

When an airplane crashes, people avoid air travel because "better safe than sorry". Yet air travel is the safest form of travel (much safer than cars) and one plane crashing doesn't change that. However, people feel safer driving.

Similarly, people think real RNG is unfair. If they flip a coin once and don't get heads, it's ok. Twice and they start getting unhappy. 5 times and they believe the system is unfair because of "common sense" ... when actually 5 tails in a row is reasonable.

One way to handle this is to make systems that aren't random, but instead cater to people's poor grasp on causality. If you flip a coin once and don't get heads, increase the chance of getting heads by 50% (so 50/50 first time, 75/25 next time, 87.5/12.5, etc). Do that each time until the person wins. They will believe the result is random and feel it was fair because they won.

[LyndisEcliar]

Thank you Camate and Hiroshi Minagawa, that was a terrific read. I love that you also recommended us a terrific book to read, in fact I just ordered a copy for myself. I look forward to reading more insightful comments like these from all the staff at some point.

[Genz]

Minagawa's posts are always awesome.

On topic : http://www.youtube.com/watch?feature...RJE-oOY#t=1102
Cold, mathematical randomness is perceived weirdly by players.

[Welsper59]

I appreciate the response on the matter, and while it really might just be a fallacy, consistent and predictable outcomes really do still raise a flag. When your norm is to fail more than succeed, regardless of the fact that it is significant odds in your favor (70% or greater success chance), and has been occurring over thousands of instances from any given individual, it might still be worth looking in to. I would not question the system in place if it weren't for the fact that the opposite does not hold up... ever.

If I can somewhat consistently fail something with a 70% chance of success 8 times in a row, essentially meaning the 30% of failure won out, shouldn't it be possible that I successfully do something with a 30% success chance 8 times in a row? As anyone who plays this game will tell, that will never happen... ever. The game seems to favor failure regardless of your chances of success. There are so many occasions throughout my mining in this game since launch that a 95% success chance has yielded 3/4, 4/5, and 5/6 fails. They're always in batches, be it in a row or over a larger pool of attempts. This isn't even including the rather consistent occurrences of 1/4 failures on nearly every node with a 90% or higher success chance. Why is it common and predictable that I fail a 45% success chance to meld 8+ times in a row, meaning the remaining 55% won out all those times, yet I will almost never successfully gather something with 55% success rate as a common and predictable outcome?

Maybe the different algorithms are used multiple times for each result when you craft/meld/desynth/gather, maybe it's just my (and many others') imagination on the matter, or maybe its an incorrect display of success chance (as I'm going to assume that is a separate tool from the actual RNG systems used). Either way, it just seems way too strange that for something that is true RNG, outcomes regardless of percentages become predictable in occurrence. Mind you, I'm not referring to the likelihood that after 20 attempts in failure that the next should be success, but rather the fact that it's predictable and common that I got there in the first place so often. This may just be coming from a recent frustration of failing 18 times in a row to meld something with a 45% success chance.

[Lexxuk]

A computer cannot do random (on it's own), that's why it has pseudorandom in the name, it makes you *think* it's random when really it isn't, it's possible to ensure that the results you get from a pseudorandom number generator are exactly the same every time (i.e. you start the generator, write down the first 1,000,000,000 numbers, restart it and the same 1,000,000,000 numbers come up), which is a bit like having the same lottery numbers this week, next week, the week after and the week after that.

If you wanted true randomness you could use the keyboard/mouse/controller movements of a user to generate a random value which is then applied, so, I've no idea where I'm going with this except that SE's random isn't actually.... random, it's a fixed system of numbers which can be generated again and again and again.