Mike Cook
@mtrc.bsky.social
AI, creativity and procedural generation researcher. No, not that kind of AI. I write and make games and take photographs of cities. Senior Lecturer at King's College London.
A prototype for a much larger system
http://www.possibilityspace.org | he/they
A prototype for a much larger system
http://www.possibilityspace.org | he/they
They also present an optimistic view of legal power balance, too (like Herndon and Dryhurst's The Call). Arc Raiders is published by Nexon, who make billions in revenue every year, and the law surrounding these contracts has barely been tested in court. What would a violation of it even mean?
November 25, 2025 at 1:19 PM
They also present an optimistic view of legal power balance, too (like Herndon and Dryhurst's The Call). Arc Raiders is published by Nexon, who make billions in revenue every year, and the law surrounding these contracts has barely been tested in court. What would a violation of it even mean?
It's also notable that this technique is largely used in professions that are already highly exploitative - voice acting, modelling, music. These are people who already struggle to find good work, and can't easily advocate for better deals. So there's not a lot of negotiation going on.
November 25, 2025 at 1:16 PM
It's also notable that this technique is largely used in professions that are already highly exploitative - voice acting, modelling, music. These are people who already struggle to find good work, and can't easily advocate for better deals. So there's not a lot of negotiation going on.
Mostly companies do this for two reasons: because it helps deflect PR (which is why you get industry spokespeople pointing out how the devs "did it right") and because it insulates them legally. Adobe literally puts "commercially-safe" on its adverts because that's what actually matters.
November 25, 2025 at 1:11 PM
Mostly companies do this for two reasons: because it helps deflect PR (which is why you get industry spokespeople pointing out how the devs "did it right") and because it insulates them legally. Adobe literally puts "commercially-safe" on its adverts because that's what actually matters.
I saw some chat about Arc Raiders' use of AI voice actors recently - I'm probably not going to rerecord the talk I gave at EXAG but I thought I'd drop in a few slides about the idea that you can make generative AI ethical by getting people to sign a piece of paper to say you can use their data.
November 25, 2025 at 1:09 PM
I saw some chat about Arc Raiders' use of AI voice actors recently - I'm probably not going to rerecord the talk I gave at EXAG but I thought I'd drop in a few slides about the idea that you can make generative AI ethical by getting people to sign a piece of paper to say you can use their data.
I've made/contributed to six new things at least this year, which feels great and is probably a new personal record for making (and finishing) stuff. I think I will release one more thing before the year end, fingers crossed.
November 24, 2025 at 11:44 PM
I've made/contributed to six new things at least this year, which feels great and is probably a new personal record for making (and finishing) stuff. I think I will release one more thing before the year end, fingers crossed.
It’s gonna be awkward if something kicks off in the Mediterranean in the next week but that’s a risk I’m willing to take.
November 24, 2025 at 1:57 AM
It’s gonna be awkward if something kicks off in the Mediterranean in the next week but that’s a risk I’m willing to take.
people are always like oh isn't javascript funny, it thinks a string is a number, meanwhile lua appears to have been designed by six of the world's most powerful and nefarious evil wizards to specifically torment humankind (me)
November 22, 2025 at 2:25 PM
people are always like oh isn't javascript funny, it thinks a string is a number, meanwhile lua appears to have been designed by six of the world's most powerful and nefarious evil wizards to specifically torment humankind (me)
Well Tom, good news for those who aren't sure what to do: I'm going to be recording a little series of videos this weekend to guide people through making their first generator in Tracery next week, in five quick lunch-hour-sized sessions - no experience required!
November 20, 2025 at 4:03 PM
Well Tom, good news for those who aren't sure what to do: I'm going to be recording a little series of videos this weekend to guide people through making their first generator in Tracery next week, in five quick lunch-hour-sized sessions - no experience required!
November 19, 2025 at 5:47 PM
Joon I have some bad news about my inner circle.
November 19, 2025 at 4:29 PM
Joon I have some bad news about my inner circle.
Find these videos, and more, here: www.possibilityspace.org/pixie/
And find the paper, again, here: ojs.aaai.org/index.php/AI...
🧚
And find the paper, again, here: ojs.aaai.org/index.php/AI...
🧚
November 14, 2025 at 6:40 PM
Find these videos, and more, here: www.possibilityspace.org/pixie/
And find the paper, again, here: ojs.aaai.org/index.php/AI...
🧚
And find the paper, again, here: ojs.aaai.org/index.php/AI...
🧚
We can do more nuanced things too! For this, I set up Pixie's goal so it was to make the player move around a lot, but not make the game much harder in the process. It created this spawned of very slow-moving obstacles, so they're easy to dodge but make you run around a lot more.
November 14, 2025 at 6:36 PM
We can do more nuanced things too! For this, I set up Pixie's goal so it was to make the player move around a lot, but not make the game much harder in the process. It created this spawned of very slow-moving obstacles, so they're easy to dodge but make you run around a lot more.
This can result in quite... unusual suggestions. Part of the fun with working with Pixie was seeing this kind of thing. All the examples in the paper are run on games I didn't write, so it's all new codebases Pixie wasn't designed for!
November 14, 2025 at 6:35 PM
This can result in quite... unusual suggestions. Part of the fun with working with Pixie was seeing this kind of thing. All the examples in the paper are run on games I didn't write, so it's all new codebases Pixie wasn't designed for!
So for example, here's an implementation of Breakout. We'd like to make a powerup to make it easier for the player to win. After a while, Pixie came up with this - the ball gets bigger on each bounce. Watch what happens... a very broken idea, but one that's kind of fun and unexpected.
November 14, 2025 at 6:34 PM
So for example, here's an implementation of Breakout. We'd like to make a powerup to make it easier for the player to win. After a while, Pixie came up with this - the ball gets bigger on each bounce. Watch what happens... a very broken idea, but one that's kind of fun and unexpected.
Pixie requires just a little bit of guidance to get going: some annotations on your code, an example game scene it can load to test, and a goal to aim for (we express this as a numeric goal, like maximising the player's score). Then it tries lots of ideas to see what works!
November 14, 2025 at 6:33 PM
Pixie requires just a little bit of guidance to get going: some annotations on your code, an example game scene it can load to test, and a goal to aim for (we express this as a numeric goal, like maximising the player's score). Then it tries lots of ideas to see what works!
New Paper: Pixie: Code-Level Mechanic Generation for Game Designers. How do you connect automatic game design to real games? Pixie is a mechanic generation system for Unity games, as an example framework for how to build AGD systems into real game platforms.
Read here: ojs.aaai.org/index.php/AI...
Read here: ojs.aaai.org/index.php/AI...
November 14, 2025 at 6:31 PM
New Paper: Pixie: Code-Level Mechanic Generation for Game Designers. How do you connect automatic game design to real games? Pixie is a mechanic generation system for Unity games, as an example framework for how to build AGD systems into real game platforms.
Read here: ojs.aaai.org/index.php/AI...
Read here: ojs.aaai.org/index.php/AI...
This paper includes, among other things, an attempt to define a subset of procedural generation work: "procedural gameplay systems" (not my best term-coining, I'll admit). I report on a survey of players and developers on what we mean when we talk about "games with procedural generation in".
November 14, 2025 at 6:18 PM
This paper includes, among other things, an attempt to define a subset of procedural generation work: "procedural gameplay systems" (not my best term-coining, I'll admit). I report on a survey of players and developers on what we mean when we talk about "games with procedural generation in".
New Paper: Game Design is Generative Design. Designing a procedural generator and designing a game share a lot of skills, problems and philosophies, even if we think of them as distinct. I try to connect them here, for good!
Read here: ojs.aaai.org/index.php/AI...
Read here: ojs.aaai.org/index.php/AI...
November 14, 2025 at 6:17 PM
New Paper: Game Design is Generative Design. Designing a procedural generator and designing a game share a lot of skills, problems and philosophies, even if we think of them as distinct. I try to connect them here, for good!
Read here: ojs.aaai.org/index.php/AI...
Read here: ojs.aaai.org/index.php/AI...
Guy turning over calendar page meme
November 14, 2025 at 5:09 PM
Guy turning over calendar page meme
Absolutely delighted by this last #aiide keynote by Jonathan Schaefer. It’s about the history of computer chess but you may know him as the guy who “solved” Checkers: en.wikipedia.org/wiki/Jonatha...
November 14, 2025 at 4:40 PM
Absolutely delighted by this last #aiide keynote by Jonathan Schaefer. It’s about the history of computer chess but you may know him as the guy who “solved” Checkers: en.wikipedia.org/wiki/Jonatha...
Cosmic Express on the stage at AIIDE 2025! @draknek.bsky.social
November 13, 2025 at 10:56 PM
Cosmic Express on the stage at AIIDE 2025! @draknek.bsky.social
me: boy i really don't have a lot of time to write this talk!
also me:
also me:
November 13, 2025 at 5:52 PM
me: boy i really don't have a lot of time to write this talk!
also me:
also me:
New Paper: "We Call This Controller Skip: AI For Speedrunning" by me, @mastermilkx.bsky.social, M Awiszus, F Carnovalini, A Dockhorn for #exag25
ceur-ws.org/Vol-4090/pap...
What is speedrunning? Why is it interesting for AI researchers? What fun, silly, playful things could we make together?
ceur-ws.org/Vol-4090/pap...
What is speedrunning? Why is it interesting for AI researchers? What fun, silly, playful things could we make together?
November 12, 2025 at 3:58 PM
New Paper: "We Call This Controller Skip: AI For Speedrunning" by me, @mastermilkx.bsky.social, M Awiszus, F Carnovalini, A Dockhorn for #exag25
ceur-ws.org/Vol-4090/pap...
What is speedrunning? Why is it interesting for AI researchers? What fun, silly, playful things could we make together?
ceur-ws.org/Vol-4090/pap...
What is speedrunning? Why is it interesting for AI researchers? What fun, silly, playful things could we make together?
AIIDE has a “test of time” paper award now - for papers from the past that has a long impact. I really love this, it’s something a few other conferences do and I think it’s an excellent way to reflect on the past.
November 12, 2025 at 3:53 PM
AIIDE has a “test of time” paper award now - for papers from the past that has a long impact. I really love this, it’s something a few other conferences do and I think it’s an excellent way to reflect on the past.