Over the years, something I've seen over and over again is people observing that human intuition of odds vs how odds works in game contexts significantly diverge. When mathematicians are involved in discussions in particular, the default tendency is to frame humans as being bad at properly understanding the true meaning of the math; when mathematicians are speaking in relation to video games especially, I've seen repeatedly mathematicians responding to players complaining that the RNG is 'unfair' by doing controlled tests to determine whether the RNG outputs produce outcomes in line with the stated odds. ie the mathematician sees '30% chance to happen', and does a few hundred tests to see if the thing does in fact occur around 30% of the time. Typically whebn mathematicians do this, they conclude that the odds are fair and that players are just whiny crybabies. Like, to be clear, I've yet to see a mathematician be so overtly demeaning, but the actual point I'm making is that they start from the belief that the mismatch between game design and human expectation is human expectation being simply incorrect, and the extent to which they humor such complaints is to double-check whether the math does what the math says it's supposed to do. In my view, this is Missing The Point on innumerable levels; this is in fact a topic I've touched on here and there over the years, but I've always struggled to construct a complete essay on the topic that clearly covers all the layers in the way I want. Hopefully this page will at last be the success. ----------------------------------------------- So first I'm going to revisit and expand on a point I've alluded to a few times on Vigaroe: Casinos, gambling, everything in the vicinity of those. Early video game design tended to draw its sense of how to approach randomness from this familiar framework. I suspect most people doing so didn't think of what they were doing as 'I am using a casino as my guideline', but fundamentally I've been clear for a very long time that this was functionally so, regardless of whatever the internal experience might have been. Now, there's two major facets to this that I've attempted to describe before, but I've only partially articulated one of them: The House Always Wins. Remember how I said players complain about 'unfair' RNG? Yeah. A casino is built to be unfair RNG. This is fundamental: any business model centered on gambling has to favor the business somehow, or it will inevitably cease to be a functioning business. Which fundamentally means that imitating a casino's approach to randomness means you're baking 'unfairness' into your game's metaphorical DNA; the math doing what the math says it's doing doesn't mean the setup isn't unfair. It can mean the math doing exactly what it's supposed to do is unfair to the player. Thankfully, as 'roguelites' have risen enormously in popularity there's been an attendant reduction in casino-style usage of randomness, replacing it with approaches to randomness that are less fundamentally hostile to the player. This trend is in fact part of what's helped catalyze my understanding of how the framework to the randomness matters in whether the result is fair or not, as I got to viscerally experience that old-style randomness tends to fundamentally anger me whereas new-style randomness often helps keep me interested in game by virtue of the unpredictable experience demanding I adjust my strategy in response to variable conditions so winning once doesn't mean I'vbe 'solved the game'. The other facet I took even longer to come to verbal grips with is this: what I'm going to obnoxiously call 'real' randomness is a very different creature from 'game theory' randomness, and the human expectations around randomness are correctly-adapted to 'real' randomness. Let me give a concrete example. Say you have a barrel that you have been told has 99 apples and one orange. For the purposes of this exercise, we will say you are terribly allergic to apples and cannot risk eating even a single bite, but oranges are fine, oranges are great even. You are hungry, and really don't want to go seeking food out elsewhere. If you blindly reach into this barrel in an attempt to sate your hunger, we can say you have a 1 in 100 chance of actually getting the orange you need. The odds are against you! Ah, but here's the thing: once you pull out one of those apples you can't stand, what is a person likely to do? Set the unwanted apple aside and reach in again. This perfectly normal human behavior results in the odds on your second attempt being slightly higher; a 1 in 99 chance of getting what you want. And every time you iterate it, your odds get better, and in fact they get better increasingly faster; without getting overly technical, just blindly reaching in over and over again like this is extremely unlikely to require 100 attempts to get the orange you want. Furthermore, it does in fact have a clear maximum fail rate: you can only miss the orange 99 times. If you fail 99 times, your hundreth time is a guaranteed success. If you're familiar with common fallacies among gamblers, the above might put you in mind of one: the concept of being 'due', where a gambler believes that failing many times in a row means they are ever more likely to succeed on future attempts. This is indeed a fallacy when it comes to a roll of the dice or a flip of the coin or various other forms of randomness common to gambling, but when it comes to navigating a physical environment where you have partial information that can be abstracted into random terms? Yes, you are 'due' to find the thing if you keep searching. The only way for this to be untrue is if your information is bad somehow; in our barrel example, if you get to the bottom and still no orange exists, the correct belief is not 'I was unlucky', but is more likely to be 'the person who told me the contents of this barrel lied to me'. In my view, this is what human intuition on 'randomness' is heavily adapted to: navigation and future planning of a physical environment where only partial information is available but other parts can be inferred to some extent from what is currently known. This is a completely different type of randomness from 'let's roll a dice twenty times'; a D20 is under no obligation to land on any particular facing. If you roll it a thousand times, it's unlikely it will never land on 20 in that streak, but it's perfectly possible.