Go to the game's main page
Review
There’s a proud IF tradition of trolling game and subversive ones that cheekily undermine their ostensible premises (let’s pick, oh, 9:05 and Nemesis Macana as respective exemplars), and Space Mission: 2045 is a fitting inheritor to that legacy. There’s perhaps no topic in contemporary IF as contentious as the use of LLMs, so what could be riper for satire? Space Mission: 2045 lures the player in with an earnest-seeming 2,500 word monograph about the weaknesses of previous attempts to use generative AI to write IF, which segues into a manifesto for the virtues of the author’s approach, followed by an extended manual for the RPG-lite system the game uses (a particularly fun gag is that in a game about traveling through space to colonize Mars, “Animal Kinship” is one of the ten skills you can pick), and wrapping up with more details on how you can interact with the AI-driven parser – again, there’s a good piss-take here where the readme implies that the response to EXAMINE [OBJECT] will always be LLM-generated rather than written by a human, but that “doesn’t necessarily mean that the details are useless.” (emphasis added).
It’s all very gung-ho, but when you launch the game, there’s a one-two sucker punch. The first is that this cutting-edge game looks extraordinarily primitive, displaying without paragraph breaks or any left-margin whatsoever – all this makes the prospect of reading reams of LLM-generated text even more terrifying. The first and a half – OK, I should go back and edit the first sentence of this paragraph now that I’ve thought of one more, but it’s late so we’re just going with it – is that we’re going on this Mars mission on behalf of thinly-veiled caricatures of Elon Musk and Donald Trump, which I really hope is meant to be a setup for jokes.
The second is that, of course, after you get done with the over-complex chargen section, nothing works. All the sophisticated features promised by the readme, like natural-language dialogue with NPCs who could adjudge whether you’d been sufficiently persuasive to convince them, or the smart parser which matches whatever convoluted sentence you care to write to the more limited list of actions that are actually possible In a particular sequence? Yeah, they sound great, and in fairness this is a game that allows you type whatever you want – it’s just that you’ll always get a response telling you there’s something wrong with the API key the game uses to interface with the LLM. It’s not particularly sophisticated, I suppose, but this is an elegant way of puncturing the pretensions of AI evangelists, demonstrating that even leaving aside the substance of what they claim, you can’t trust a chatbot game to even last all the way through a Comp before tech-company shenanigans knock it offline (meanwhile, interpreters mean we can still play forty-five year old classics with a click). It also opens up space for improvisational comedy like this:
What do you want to do? write a witty deconstruction of the folly of overreliance on generative ai
Error during AI interpretation: 401 Client Error: Unauthorized for url: https://openrouter.ai/api/v1/chat/completions
The only thing that could make the satire better is if it were actually intentional – since I glanced at some other reviews, and turns out there once was a playable game here? Sadly, Space Mission: 2045 might be more of a monument to hubris than deconstruction of same; if it ever starts working again, I’ll try to revisit it and update this review accordingly, but in the meantime, it’s one of the most effective bits of inadvertent self-mockery since the literary career of Norman Mailer.
[I believe that after the Comp closed, the game started working again; I haven't gone back to play it but I'm omitting my rating from the average]
