I remember when I first learned about ChatGPT. I was a first-year law student at Gonzaga University, searching for better ways to take notes and to practice “IRACing” for my legal writing class. As I browsed the web, I found recommendations for using AI, specifically ChatGPT. I investigated the program and was quickly scared off by fears of academic dishonesty, stories of hallucinated facts, and the black-box nature of the tools.

That summer, I took a course on AI ethics and privacy. My main takeaway was simple: look for technology that improves, not replaces, human agency. I became antagonistic toward artificial intelligence, especially ChatGPT, because it seemed aimed at replacement. I often debated friends who used AI to make art or write stories and songs, arguing that AI kills creativity, takes away the beauty of human creation, and strips us of agency. I still believe it does.

But I still had to learn how it works. As a teaching assistant, I created a lesson on using AI in legal research. I read the literature, watched lectures, and gathered tips for using AI tools. I taught the class and later used that lecture in my job talks after graduation. I felt I understood the mechanics: how these systems parse words and use statistics to produce prose that sounds right and may or may not be correct. I knew that prompting matters, that iterations improve answers, and that breaking a large project into smaller pieces helps the tool build better results. Yet I had not grasped the nuances of prompting–when, where, and how to iterate well.

Enter Dungeons and Dragons (DND). For those who do not know, it is a roleplaying game in which a dungeon master narrates a world of monsters and adventures, and players act through their characters. Many actions are resolved by rolling dice, such as a twenty-sided die (a d20). Try to pick a pocket, roll the die; a high roll succeeds.

I love Dungeons and Dragons. I began playing in law school and soon became a dungeon master, creating a world of kingdoms, fantasy races, geopolitical tensions, and a working economy. I realized I could use that world as a sandbox for understanding AI. Because I knew the setting, I could spot hallucinations and see whether my prompts caused them. Most importantly, I would not be feeding any sensitive information. It was a safe, controlled environment.

I started an online DND group. With all players’ permission and clear disclosure of my intent, I expanded the world using artificial intelligence. I created a project folder, set parameters, and uploaded my lore, maps, and game notes. Each week I used the system to prepare for my sessions.

Preparing a Dungeons and Dragons game is a great deal of work. I create spreadsheets and maps; draft non-player character (NPC) dialogue and story beats; design monsters and puzzles; and plan for the unexpected. It feels a lot like preparing to teach, but that is for another day.

AI improved my preparation. I went from spending roughly six days a week preparing for a session to two. I wrote less but produced more: documents, stories, monsters, lore, and visuals.

Working with version 3.5, I learned its strengths and limits. It excelled at turning a bullet list of story points into a coherent narrative, but it struggled to cross-reference across uploaded documents. It could summarize a document, then draft and pull in outside elements. I responded by tightening my prompts.

When versions 4.0 and 4.4 arrived, I refined my methods and kept testing boundaries. For example, when I uploaded PDFs of my players’ short stories and asked for a new story based on them, it mixed plots and confused characters. The lesson was clear: use it to shape text but watch for subtle blending.

Those experiments were translated to my reference work. I developed a prompt-and-iteration process that helped ChatGPT summarize a PDF with accuracy. I used it to improve the clarity of my emails and to catch small mistakes. Because of my dyslexia, I sometimes misspell words, drop words, or struggle with structure. AI became a useful safeguard and accessibility tool.

I also learned what not to ask. I do not rely on AI to summarize very large documents. I avoid pushing it for cross-references or for searching inside a document.

This blend of learning new technology through games is not new. My first dungeon master taught himself Excel by building spreadsheets to track players, gold, and levels. He inspired me to do the same with artificial intelligence and hopefully I can do the same for you with whatever technology you are trying to learn.

Hobbies provide excellent sandboxes. They let us pair what we know well with a new tool in a controlled setting. When the tool fails, we learn and adjust. We see the tradeoffs, gain context, and understand how it works.

And that is how Dungeons and Dragons made me better at my job.