In this episode of Talking to AI, Paul and his friend Mark attempt a Dungeons & Dragons session led by ChatGPT as their AI Dungeon Master. Building upon a detailed and ambitious prompt, they strive to see if AI can live up to the challenge of running an engaging and rule-abiding tabletop adventure. Despite hopes for thrilling encounters after their underwhelming first attempt, the session falls short: the AI's story remains passive, adversaries are non-confrontational, and the mechanics often ignore or misinterpret essential D&D rules. Paul and Mark experience firsthand the AI's struggles with dice rolls, stat calculations, and narrative proactivity.

Throughout the journey, Paul critiques the AI's failure to set up characters properly, missed steps in gear assignment, and lack of initiative from non-player characters. Even when provided with ample rules and resources—including a book of starting scenarios and the full D&D ruleset—the AI heavily relies on player direction and fails to build dramatic tension or memorable story moments, often repeating bland phrases and showing limited strategic depth.

After reflecting on the session, Paul and Mark discuss the inherent limitations of current generative AI tools as D&D Dungeon Masters, exploring whether simpler structures or additional context might improve results in subsequent experiments. Their experience exposes both the comedic mishaps and persistent shortcomings of AI-driven roleplay, making for an entertaining but cautionary tale about the current state of AI in gaming.

🎙️ Hosted by Paul at Talking to AI — where real people, real problems, and real conversations meet artificial intelligence.