Space Mission: 2045
In theory, Space Mission: 2045 does something interesting. The author explains that he was very disappointed by AI Dungeon, because you are just wandering around aimlessly in a world that has no persistent reality. I agree completely! And so the author wanted to make a game that uses LLMs, but not as the core driver of the story, but more as a sort of helpful GM running the scenario prepared by the human author. The LLM can parse commands, fill in details of interaction, and so on; but underneath we have the story written by the author and a persistent world model.
There are serious questions to be asked about how useful this approach could be. One of the nice things about parser games is that it is usually quite clear whether your command was understood and whether it changed anything in the world. Putting in an AI assistant to deal with commands that have not been understood might increase the believability and depth of the play experience; but it might, instead, just make gameplay less transparent and give the player a feeling of groping in the dark. It is to Benjamin Knob’s credit that he thought about this issue. One of the things we learn from the readme file is that the AI has access to information about the game world that has not been revealed to the player, so that text generated by the AI can reveal this information and be useful to the player. This could work well. I don’t know; but it’s something that is worth trying out, and Knob is here to try it out. That is to be applauded.
Unfortunately, playing Space Mission: 2045 is not a joyous experience. First, the AI is very slow. If I understand it correctly, the AI queries are sent to a remote server, which can be more or less busy. I don’t know whether I was playing at a particularly busy time – it was Friday evening in Europe – but I do know that responses often took 10 to 20 seconds, and sometimes more. Playing a game at this speed is, if not torture, at least an exercise in frustration. Even if the gameplay had been brilliant it would still have marred the experience.
Exacerbating the feeling of frustration are serious problems of game design. On my first attempt to play the game, I quickly landed my pod on Mars, then typed ‘exit pod’. The game gleefully told me that I had forgotten to put on my space suit and helmet, and therefore I had died in the vacuum of Mars. Game over! No way to undo.
This is the kind of cheap trick that we grew out of sometime in the 80s. Either you automatically have me put on my space suit, or you give me a warning before I exit the pod, but you do not punish my failure to read the author’s mind with an insta-death… that cannot be undone! This totally breaks the expected social contract of parser IF gameplay; and the fact that the game runs at a glacial speed makes things much worse, because now I have to sit through minutes of remote server queries just to get back to where I was.
I did decide to give the game a second chance. This time I went to Earth instead of landing on Mars. There, I was quickly captured by some fascist goons. After a fairly interminable scene in a prison cell – I could not for the life of me understand what the guard wanted from me, and the AI generated responses were not helping at all – I ended up before an evil emperor who somehow is using the New World Computing logo as the symbol of his evil empire. These scenes were cringe-worthy. Everyone was a cartoon villain, shrieking, laughing evil laughs, gleefully telling me about my execution;tiring stuff with the cliché dial turned up to eleven. And it turned out that it was all part of the worst GM trick in the book: railroading me back to the pre-made adventure. The ability to go to Earth was an illusion. I was condemned to execution, but instead of being executed, I got sent back to Mars, with my onboard computer once again suggesting that I land there and nothing changed from the initial state of the game!
Since not all my readers may have experience with pen-and-paper roleplaying games, I want to explain this a little bit more. Pen-and-paper games have always had this promise that you can do anything. You’re a character in a world, and you can do whatever that character wants! And that might be true in certain systems. But it cannot be true in systems that have a Game Master and require this GM to prepare, often with a lot of labour, an adventure for your characters to have. If the GM prepares an adventure in the lair of the necromancer, and tells you that your character meets a desperate woman whose child has been kidnapped by the necromancer… then you’d better go after that necromancer, or the entire gaming evening collapses.
Now it is very possible for roleplaying groups to get into dysfunctional states where the players don’t feel like they have real agency, and where they instinctively react against the implicit wishes of the GM, perhaps by making up plausible in-character reasons for not going after the necromancer. And now the GM might feel forced to, I don’t know, have the characters kidnapped at night and then wake up in the lair of the necromancer, because that’s the only way he can get them into the adventure he has prepared! This ‘technique’ is called railroading. It is BAD.
There are many ways that roleplaying games can avoid this horrible pitfall, and I’d be happy to discuss them if someone is interested. But railroading is a phenomenon that I had never yet seen in interactive fiction. Authors don’t need it, because as an author you are not giving the player options that will take him out of your prepared story. You can’t turn back and go home in Zork and spend your afternoon gardening; and so the Zork authors don’t need to think of a way to get you back from your home to the white house above the Underground Empire. But what if you use AI to expand the realm of the possible? Now suddenly it might seem as if the player is going to have so much freedom that he can start doing things that have nothing to do with your story. He can go gardening. And so, perhaps, we need to railroad him back in…
So, there I am, back on Mars. Having learned from previous experiences, I wear my suit and helmet, and go out of the pod. I’m on a big plain. I type “go south”.
And that, dear reader, was another insta-death. Apparently I wandered around for a long time until my oxygen supplies ran out and I died. I have no idea how this intention to just keep walking forever was implicit in my command to go south, but that’s what the game did to me. Even in the 80s, only particularly evil games worked like this; nowadays, that’s just a fundamental betrayal of the social contract between parser authors and their players. Having once again been kicked in the teeth by Space Mission: 2045, I gave up on it.