[Book Review] Mike Cook's "Next Level: Making Games That Make Themselves" (Patreon)

Mike Cook is a cool guy! I first became aware of him back around 2015 when I was following him on Twitter and attended PROCJAM; a jam he founded where people were encouraged to "make stuff that makes other stuff" - sometimes games, sometimes not.

He also created ANGELINA - a game-designing AI (before the current wave of technocapital-fueled enshittification of every hope we had for the future of such technologies) and was the author of the NO-ONE IS GOING TO BUY YOUR GAME manifesto from last month's Manifesto Jam; a document which many scratched their heads at, perplexed, while anyone who's ever tried teaching game development to anyone aged 18-25 agreed with... with a grim smile and a sagely and handsome nod.

I like Mike's work, so I was really excited to get into his new book NEXT LEVEL for the upcoming episode of the Game Developers' Library.

You'll get mine and Tom's thoughts on the book in a week or so when the podcast is ready, but my notes and reflections are below... my notes on the GDL books are usually paywalled for folks in my 2 highest tiers of patronage, but I thought I'd make this one free as a bit of a sneak-peek. Also, thanks to the publishers for providing Tom and I with review copies - this is the first time we've reviewed a new new book, and I appreciate the opportunity!

The overall structure of Next Level makes it very readable, and although I'd say it works fine as a primer on the concept of procedural generation more generally, it's also the sort of pop-science writing you could probably gift to a family member who is generally into games, or who works in tech and is loosely interested in how games work but doesn't make them.

For most of the book we follow a structure where a concept from math and/or computing is introduced and we're shown how it can be used creatively. Then, the following chapter uses a famous historical game as a methodological case study in that use of procedural generation. By the end, the book breaks from this structure a little, but for the first 2/3 of the book, this structure makes it exceptionally easy to blitz through quickly.

In Chapter 1 we learn some basic principles of chance in dice and card games. This is the stuff you probably already know about probability bell-curves when summing the results of multiple dice, or how poker players judge odds, and the tarot's function as an early procedural generator. There's no direct mention-by-name of apophenia, here but it's a concept which comes up in Chapter 1 and will show up in later chapters where we're discussing how games like Dwarf Fortress and Caves of Qud encourage players to make connections where there probably aren't really any.

In the second chapter we're given a case study on Elite (1984) and Rogue (1980) with slightly different foci. Elite gives us a way of thinking about seed numbers (in this case using the tribbonaci sequence to generate a massive galaxy of planets) where Rogue is more of a spatial game which allows us to think about algorithms to generate rooms and halls, such as the drunkard's walk. I found it kind of delightful that this algorithm (which I've probably arrived at from first principles more than once) even had a name.

In Chapters 3 and 4 we're introduced to digital equivalents of the Burroughs/Bowie cut up method, whereby game elements are assembled not entirely from scratch via algorithm(s), but starting with chunks of authored content. Our test game for this methodology is - as you might've predicted - the Spelunky cave generator. As with Elite and Rogue, Cook doesn't skimp on technical detail here in terms of explaining how Spelunky's procgen works... but he still does it in service to this broader view of the role of randomness in human creativity, rather than some unique function of computers per se. If you've already read Derek Yu's Spelunky - which Tom and I covered on our first episode together - then there's not much new in this chapter. But I'm glad that its here, not least because it's easy for a goofily-presented cartoony platformer might get lost in games history compared to some of these more lofty sandbox games.

Chapters 5 and 6 are about types of noise generation, which are introduced thoroughly with a lot of great diagrams, then explored in the context of Minecraft. There's a lot of stuff I already knew here by name, but hadn't quite grasped the application. For example, Cook talks about how one "tier" in the images produced by a noise generator could be isolated and treated as "rivers" - something I'd never personally considered because I always assumed rivers were created as a path rather than a boundary which just happens to resemble a path. Again, the emphasis here is on the relationship between the generator and the actual human(s) using it - a process of tweaking, interpretation, and application of imagery and metaphor to the generator outputs to make something which is suitable to the immersive-fantasy aspect of a game (or also, in Minecraft's case, it's geological economy).

I've actually been dipping back into bedrock MC a lot recently and it's been wild to see how much the world generation algorithms have changed. No longer do I need to dig down and then boringly strip-mine square lattices until I hit the resources I need, because the cavern generators have become so much richer and its far more common to find diamonds and other minerals while simply exploring.

But at the end of the Minecraft section Cook also draws attention to this - how a generator-driven game like Minecraft, which is basically a live service more than a static media object - can lead players to lament previous versions. And while I've found myself enjoying where Minecraft has gone over the past few years (bearing in mind I was playing it before the first commercial version) it's also understandable that people might long for a more simple cube world. That is, after all, the original promise of the game, and the richer it becomes in its variety of mobs and resources, its graphical fidelity, and the interwovenness of its biomes - the more it appears to simulate a "real" fantasy world, and the less it wears its generation on its clunky sleeve.

We now get a chapter on simulation - what does it mean for a player to explore a virtual world which isn't only designed around them as the hero-solipsist-protagonist - but where agentic NPCs do stuff when you're not looking. This leads to chapters on Dwarf Fortress and Caves of Qud - which have made me want to go back and play both, despite not really having the time at the moment [thanks mike]. I'm glad Cook put these case studies side-by-side, because it allows him to look at very different applications of the same sort of technologies and how they are used to achieve different artistic goals.

While we do get the explanation of one new PGCtechnique during the chapter on CoQ (Wave Function Collapse) the discussion is more about how these two games vary in their use of simulation-based PGC histories Dwarf Fortress is different every time, whereas CoQ generates a world with a consistent starting area (and history in that area) and then a different world outside of it (with some of the same broad political structures and historical realities). Dwarf Fortress generates a history which is a "real" set of events (many of which are free for the player to interpret to infer their own relationships into: see Apophenia above). CoQ instead generates histories which are always meant to be seen as authored and flawed re-tellings, with some written by hand and others generated when the player starts a new world.

I really appreciate Cook's way of putting these two titles next to each other and comparing them. It really hits at the central thesis of this book (and of basically a lot of what I've seen Mike make or write since the first PROCJAM I attended in 2015). Yes, computers are cool and can do cool things. But the coolness of the things is a product of how they are used. They are tools. Not in the current Gen-AI cop-out way of saying "it's just a tool". But in the "let's compare Caves of Qud and Dwarf Fortress" way of saying "it's just a tool". A pure technologist couldn't write this book; someone who was only interested in games as a test of a type of technology would not see these nuances. Cook takes apart Caves and Fortress and is able to show us how one configuration of fantasy and systems is able to produce a cosy and whimsical world of pragmatic toil and drunken frivolity, while another creates a harsher, enigmatic, alien science-fantasy world which is deliberately uninviting and, to some degree, always unknowable.

I'm still working my way through the last two chapters, having fallen behind on podcast work what with My Game Having a Deadline Now and Also Writing My Own Book Which is Due in 9 Months. We get a very measured discussion of Gen-AI (including whether that's the right name for it) which addresses all of the issues with it in a way which doesn't come across as scolding (then again, I might be biased). It's impressive that Cook is able to find a working example which brings the topic back to games which also highlighting all of the issues with machine-learning inputs and outputs; a brief recounting of how AI Dungeon was made, and how (to this day) it still lacks many of the content-output controls of other chatbots, despite working in fundamentally the same way.

It's probably worth noting - given the name of the page you're on - that all of the games discussed in the book are (or at least started as) indie productions. And maybe that's testament to the risk aversion which comes with larger studios. I've often heard roguelike devs say that one of the reasons they like working with procgen is that they never get bored of testing their own game. For those of us who are engaging with these long experimental projects with no guarantee of a paycheck at the other end, maybe there's a lot of wisdom in trying to - like David Bowie using William Burrough's cut-up-poetry methods - find the intersection between "method which creates something cool for others" and "method which is fun to actually do, regardless of outcome". As indie devs, we continue to Fuck About... and occasionally, if we are very lucky, we also get to Find Out.

Mike's book is a must-read for anyone who's making a roguelike or PGC-heavy game - although it's more of a treatise in creative philosophy of those tools than a "how to" guide on any of them. For anyone else, it's going to be a great piece of "history by topic", and - as I said at the top - a good stocking-filler gift for the creative nerds in your families.

-JBW