AI Dungeon

by Nick Walton

2019
Multiple
Python

Go to the game's main page

Review

3 of 3 people found the following review helpful:
A long, strange history, October 13, 2024
by Lance Cirone (Backwater, Vermont)

AI Dungeon isn't a game many people would want to play today. The ethics of generative AI have been discussed countless times, the environmental impact is horrible, and it's been shoved down everyone's throats for a while now -- it's become the next tech buzzword that every company thinks they need to use.

But back in early 2021, when generative AI was in its infancy, not as readily available to the public, and restricted to just text with images being a pipe dream, AI Dungeon was something to see. I loved playing this back then. My stories rarely ever made sense and I'd try to follow them the best I could, but if I got bored, I'd prompt it with some crazy plot twist and see where it tried to take everything. The writing could also range from funny, to profound, to just bizarre. Sometimes people wouldn't even just use it for stories: they'd prompt it to write song lyrics, jokes, or lists of fun facts, which came out as incredibly mangled and hilarious in a surreal way.

Granted, even then it had its problems. The game had an insistence on making you go to school, wake up in the forest, or have your mother appear in the story, no matter how much sense it made. It was also really bad at following physical character descriptions or permanent bits of worldbuilding you wanted to add.

If you want to see what a traditional text AI output looked like around this time, try B.J. Best's You Will Thank Me As Fast As You Thank a Werewolf, it was made around a similar time and had a very similar energy when I read through it.

The game's decline was rapid. If you asked me to review it in March 2021, I'd probably say four or five stars. But around April, the game was hit with an infamous censorship system. This wouldn't be a problem if the censor wasn't so trigger-happy. Having a character or animal that was under 18, in the cleanest contexts, would upset the censor. Even saying that a person had short hair would trigger it. It would flag your story, halt it, and Latitude would read your private stories for content violations.

Fans complained, and the response of Latitude was horrible as well. If the censor got triggered, you'd get the patronizing message of "this took a weird turn, help us figure it out?" And if you clicked that, they'd double down with the text "It's possible that our systems messed up; that can happen when you're on the bleeding edge of technology" and imply it's uncommon that the player would "actually like reading." This did nothing to appease the already angry community, who were backing up their stories into Google Docs, canceling their subscriptions, and deleting their accounts. Latitude also changed their official Discord server into an unofficial one and removed their social media links from the game's homepage.

By May that month, the true source of the training data was found, with it being heavily illegal and NSFW material. Latitude's next update was releasing an in-app purchase that let you have the AI speak text out loud in a Russian accent. They also started to let random people sign up to read the flagged private stories for 7 cents each.

I dipped from the game and checked back in 2022, where it turned out to have an energy system: one ad gave you 10 turns. A Steam release that didn't even have the premium AI was made with a price point of 30 dollars. The AI itself also got even dumber -- while it had a bit of inconsistency, it was flat-out unable to write a story by this point.

MathBrush's recent review inspired me to check in on the current state of the game, and it seems like it hasn't improved too much since then. There are also competitors like NovelAI, but in a world where AI is now starting to become more of a genuine threat than just a tool to use for fun, I don't think the appeal is there to most IF authors.

Was this review helpful to you?   Yes   No   Remove vote  
More Options

 | Add a comment