"Oatmeal is Cheap: A Fundamental Theorem for Procedural Generators" by Younès Rabii

By studying the Kolmogorov complexity of procedural generators, we end up with a fundamental theorem for programs that can generate things. There's a deep link between the knowledge we encode in a generator and the complexity of what it can output. In the end, this result helps us understand the pragmatic decisions designers have to take when developing games like Minecraft, Borderlands or Tea Garden. The talk is illustrated with interactive examples that the audience can play with.
Younès Rabii
Game Designer + Generative AI Researcher
@pyrofoux
Younès Rabii is a PhD researcher in game AI and the award-winning designer of games such as SUPER IS HOT, Tea Garden and Neurocracy. Their work has been previously exhibited in the French embassies and international festivals. Their current research is focused on building AI techniques that can help designers create new forms of play and storytelling. Younès' work has two main goals: (1) resist the systemic oppression within the AI and game industries; (2) create games that can surprise players and designers alike.
----
Recorded Sept 22, 2023 at Strange Loop 2023 in St. Louis, MO.
thestrangeloop.com

[@blaiseutube]

I happen to be baking oat bread, listening to this lecture.

[@jorgemonasterio8361]

Just outstanding.
The graph view of the theorem was so great.

[@Debrugger]

Great talk, the playable graph is just awesome

[@everybot-it]

Insightful!!

[@donaldhobson8873]

This is another case of dubious interpretation of the maths.
What we care about is the in practice visual interest. This is a function of how the human brain processes information.
Sufficiently rich processes, like running randomly generated turing machines / cellular autonoma can produce lots of rich patterns from little code.

[@Debrugger]

Not a fan of your dismissive tone, and not sure what your point is. A random turing machine is not an interesting minecraft world is it? The idea of their paper is to formalize the space between a checkerboard and noise, both of which are rarely what you want when generating eg, game levels even though one has lots of patterns and the other no "repetitiveness" in the information theory sense.
Simple cellular automata as you say are usually too close to the former category, patterns are too similar and not interesting enough for humans.