We created a problem for ourselves that was fun -- games -- but once we'd beaten "games", the reward function kept running and now we're adversarially inventing new ways to "optimally" beat harder and harder versions of "games", ie new objectives, new limitations, new allowable orderings of elements within the games, etc.
I wonder what 50 years of this will yield!
I think the greatest form of it may be the act of saying "fuck you!" to every bell curve imaginable. Fuck bell curves, I'm going to do something else.
I really want to start a cult of bell curve hate.
> I was thinking of it like a mating display.
What is "life", if not a paperclip-optimising AI applied to our own DNA?
In other words, whether you choose paper clips, Bibles or human DNA as your optimization goal, over very long timescales (like a billion years) the optimal algorithms for all of them converge. If you want to make a solar mass of paper clips, you need a lot of infrastructure in common with a solar mass of any other goal.
I'm reminded of Schmidhuber's PowerPlay (https://arxiv.org/abs/1112.5309), too. It just trains on increasingly difficult problems with no direct objective function beyond solving the current problem and devising a harder one.
So this isn't necessarily what you'd call a randomiser — often the mods aren't randomly generated at all, but carefully curated by the event organisers. But it's à propos in terms of 'new experience every time'.
One neat aspect of this approach is how unexpected the mods can get; e.g. your player character could be swapped out for a character from a different game. Then there's the extreme variant of mystery races, where the runners don't even know which games they're going to play.
1) There's lots of different categories to compete in. As new possibilities are discovered in the game, the community congrats around goals of speedrunning any given seed. In these scenarios, its generally about building the skills to be proficient at a particular goal in record time regardless of the seed.
2) most of the modern roguelites I play include a "Daily Seed," where every player gets to play the game exactly once each day and be ranked on a leaderboard. You compete on a given seed. If you want a global ranking system, you might consider it by track-record. Something like, "of the last 50 games this player played, what is their aggregate ranking vs other players' last 50 games played", or do that same comparison, but only in the cases where those players shared a seed for the RNG.
It's true that there's no absolute "best game", but there is "best game ever on this particularly interesting seed."
This leads me to believe that this approach can both make speedrunning continue to be new and interesting to participants as well as result in a scenario where the record can mean something. Yes, there are ways to cheat the system if you really want to cheat, in all but a live in-person audited game. It shifts the focus slightly from reverse-engineering of the game and towards being a better all-round player. And it's more interesting for machine solving, too: how can I build an AI that will take the optimal route regardless of seed? How can I know that the optimal route was achieved?
An interesting different type of randomizer would be one which preserves everything like that, but adds better randomness (i.e., randomness independent of the emulator) to the game's own RNG. Currently, tool-assisted speedrunners and sometimes even live human speedrunners exploit the fact that a game's RNG is deterministic and predictable. Breaking that would lead to additional speed running research on the "original", "canonical" game.
As far as making a tool-assisted run goes, a much better but deterministic randomizer wouldn't affect the process very much, and a non-deterministic one would make it basically impossible to play at all.
You'd prevent real-time RNG manipulation, but that's pretty uncommon to start with.
And you could just say "no RNG manipulation", you don't need to edit the game code.
He ran Head-to-head bingo at ESA recently: https://www.youtube.com/watch?v=nMPSoThpXt8
As it is a marathon run, it features a little bit less risk-taking than non-marathon runs (I don't recall if this was a moderated seed or completely random).
For most popular speedrunning games, randomizers reward a completely different skillset than standard speedrunning. Typically, world record speedruns are a result of a community collaboration between theorizers, glitch finders, and routers, who devise the optimal "route" through the game -- what are the minimum objectives necessary to beat the game? What is the fastest order through the objectives? Then, running the game itself can require technical skill to move the character at speed with precision (particularly relevant in platforming games like Super Mario, etc) and execute tricky, timing-dependent glitches (notorious are 1-frame tricks, which require a certain input at a specific frame in a game running at 30 or even 60 frames per second).
Route optimization is somewhat equivalent to solving a traveling salesman problem, and is usually performed over the course of days or weeks in online chatting groups. In randomizer speedruns, however, the speedrunner must come up with a good route on the fly, since the particular randomization is essentially guaranteed to never have been seen before, as well as actually run the game. Needless to say, this demands, and shows off, a huge amount of expertise from the runner, and is a big treat to appreciate.
Personally, though, I think bingo speedrunning competitions are even better. Here, the goal is to win bingo on a randomly generated 5x5 grid of in-game objectives - but any square completed by you or your opponent becomes inaccessible for the other player! This retains the difficulty of routing on the fly while adding direct head-to-head competition between speedrunners. In some of the best cases, mindgames can ensue when speedrunners are aware of their relative skillsets which give them a comparative advantage at completing certain in-game objectives. Sniping squares becomes a fascinating strategy.
[0] True any% stands for beating the game with any percent completion. Communities for certain game sometimes rally around other categories that restrict or ban certain un-fun glitches (any% with certain limitations) since speedrunners ultimately want to have fun.