Artificial intelligence and advanced language models such as ChatGPT or Bard can be a powerful tool in the pocket of any worldbuilder if used correctly. While some designers may wish to avoid using AI to preserve a more handcrafted approach, others may find them useful for accomplishing simple tasks or generating new ideas they may not have considered at first. Using advanced language models has risks, however, and worldbuilders should first understand what AI language models are, what the drawbacks of using them as a worldbuilder may be, and if they still wish to use these tools, what types of prompts may be most effective.

Already familiar with advanced language models? Example prompts are listed at the end.

AI language models like ChatGPT from OpenAI or Bard from Google AI are programs designed to generate highly specific responses based on data they have been trained on from across the web. They operate by attempting to successfully predict the next word, phrase, or response to any user input. Unlike search engines, which present users with a list of sites pertaining to their search terms, AI language models generate the response they predict the user wishes to see based on a given prompt and their previous and on-going artificial intelligence training.

Are AI language models sentient? No. They are just mathematical algorithms that are very, very good at predicting human-like responses.

Plagiarism: AI language models extract their information from the web, as such there is no way of knowing to what degree they are effectively plagiarizing another work. While these models can be used to generate simple narrative ideas, worldbuilders will want to avoid using them to generate large segments of a world or a narrative in full.

Inaccurate Information: The information language models gather from the web is often inaccurate. This will be especially relevant to worldbuilders seeking realism within their worlds. If a worldbuilder wants real facts, it is better to find that data elsewhere. While using language models for ideas may be fine, make sure to double-check everything.

Least Common Denominator: AI language models provide likely and common results to the prompts they are given. This may mean that worldbuilders trying to design unique worlds will be steered instead towards more common ideas, concepts, and tropes.

Down the Rabbit Hole: AI language models try to provide you with the best responses to your prompts and not necessarily challenge you with new concepts or ideas. This can be a danger of any predictive algorithm that tries to present you with what it thinks you want to see instead of exposing you to new ideas. Worldbuilders over relying on these tools may lock themselves into a creative box and accidently shield themselves from sources of new inspiration.

Knowing the risks will help worldbuilders better understand how programs can be used effectively as well as what should be avoided.

Despite the risks, advanced language models can be helpful tools for worldbuilders. Yet in order to use them effectively, one must know how to create the proper prompts. Language models respond to user input, so to have them generate the response you want you must first design the correct question or request.

Generating lists may be the most helpful and straight forward use of AI but be sure to double check for accuracy.

These types of responses may generate stereotypical or commonplace responses. Worldbuilders may wish to extrapolate further on their own.

These types of responses may be risky, as they may begin to plagiarize what others on the internet have already created.