What the goblins can teach us about AI - lessons learned from two years of playing D&D with GenAI

Playing D&D with AI is a let down. You’d think it would be excellent! Hallucinating is the basis of the entire game! But, after a the first magical week, things started to slide. I’ll share my experience and what I learned about using AI after two years of tinkering with ChatGPT, Gemini, and Claude.

I’ll give a little bit of background on starting - a fun story of being holed up an Amsterdam hotel room for a week - and then get to my expectations and then what I struggled with, adapted to, and then learned to play D&D with gen AI. What’s interesting here is that you end up inventing the pattern of agentic AI. Thus, if you go through the process of being able to play D&D with AI, you learn a general pattern for using it for any application. Also, there are interesting copyright considerations: so much of the material you’re using is copyrighted, and often by individuals not Big Evil Hasbro. There’s some emerging practices and debates in the role playing with AI community. As you’d expect, people who write the content hate it - it’s offensive and threatening. On the other hand, there’s a lot of new material (random encounters) generated by AI that are VERY good. And, as with my experience, the ability to use AI brings joy that would previously not exist (or have the possibility of doing so). You’ve got all the issues of AI wrapped up in a smokey tavern!

Thanks for considering my talk!

I haven’t published on doing D&D with AI in a whole, but here are some older things: https://www.youtube.com/playlist?list=PLk_5VqpWEtiWbS-AHbk6WxgMfnpYaIx3g

And, here’s some examples of my past public talks: https://cote.io/speaking/