Rebellion in the Solar Empire

I went to Lego Copenhagen to participate in Global Game Jam 2025. I made a game called Rebellion in the Solar Empire with Mads Sønderstrup. It's a multiplayer game played on the web browser, made in pure javascript and php, about standing up against an overzealous censorship AI that bans every word you utter.

The game

The goal of the game is given by the Rebellion: get the AI censor to ban different words. The game gets progressively harder the more you play, because every word banned means the communication gets more spotty, since you can no longer use the banned words which could be "rebellion", "bomb", or more innocuous words like "cat", "hat" or why not, "you" or "or". There's no end goal, except a continuous struggle to keep banning words and see how long you can keep talking.

You can see a gameplay video here. I tried to document it because I was pretty sure we would never set it up again, as it's a bit of a hassle and not really worth the effort.

Inspiration

The game jam theme was "bubble" - what a fantastic theme! We were inspired by the idea of social bubbles, the AI bubble, internet censorship, and a bit of sci-fi. The original idea was about organizing rebellion in an interplanetary communication scenario, where messages are delayed due to the speed of light - a physical type of information bubble. It's a cool idea, but it's a bit boring to wait for no good reason, so we didn't use this idea much.

Local AI Models

We implemented a local AI model to serve both as the censorship AI and rebellion faction, so we don't depend on OpenAI, but that also means it's hard to set up the game for any old player. We didn't provide a version of the game that can be played locally so we relied on preplanned play sessions until we could figure out a better way to distribute the game.

We used LM Studio which has a feature to run a local server where you can query the AI using an API like the one used by OpenAI. One of the players (typically me) would run a couple of AI players that behave as the censorship and rebellion users, that would contribute chats to the chatroom. They would be autonomous agents, and would decide on their own what the new mission is for the player, and what words to ban.

Censorship

The game is inspired from the internet censorship we see today by authoritarian regimes like China, which bans any form of criticism from the platforms they control, sometimes even ridiculous words, like "pepper". The only difference is, in reality, they also threaten, arrest and disappear people who don't conform to their twisted rules that change on a whim, and the most damaging part is forbidding people from talking about real issues, like human rights abuses and government oppression.

It was interesting to study different AI models' abilities and biases. There are powerful models now that can run locally, and are even less censored than OpenAI's ChatGPT - although for better or worse ChatGPT tries to not allow users to generate hateful content, so I kinda understand their angle. However, the Chinese models that we found, such as DeepSeek, that are touted nowadays (as of cca 01/2025) in the news as beating ChatGPT, are in fact rife with propaganda, genocide-denial and bias: they refuse to criticize the communist government, or to answer questions about the 1989 protests in China (or those happening today for that matter), but are happy to trash-talk Trump, Biden, or any western governments and spew other hate. You can get them to break, but the veiled attempts at censorship are plain to see.

Basically, if they had their way, history would be rewritten and the Chinese government would be absolved of their atrocities. But that's not news for anybody, I hope.

Anyway, we tried to playfully mess with this awful topic and this game is what came out.

And now it's documented! We probably won't continue working on this particular idea but maybe it will inspire others down the line.