Investigating Bias in LLM-Based Translation

An experiment made for Bristol Digital Game Lab's 'Concept' Game Jam as part of a team of 6. The theme was algorithmic bias and the jam duration was 4 hours.

Our idea was to investigate how large language models (LLMs) may introduce bias as they translate from Chinese to English. The game allows you to translate a story, one page at a time, from Chinese to English using an LLM. You then have to answer questions about the story, considering the bias that the LLM has introduced as you do so.

An OpenAI API key is required to play, and the game only functions on Windows. Whenever translation occurs, a command prompt will briefly open so that an OpenAI API call can be made via curl.