Editing Gamemaster (section)

=== Generative artificial intelligence ===
[[Generative artificial intelligence]] (AI), built off of [[large language model]]s (LLMs), can also be used to simulate the actions of a gamemaster.<ref name=":2">{{Cite web |last=Zeoli |first=Rowan |date=2025-01-17 |title=AI Dungeon Master experiment exposes the vulnerability of Critical Role’s fandom |url=https://www.polygon.com/critical-role/510326/critical-role-transcripts-ai-dnd-dungeon-master |access-date=2025-05-02 |website=Polygon |language=en-US}}</ref><ref>{{Cite news |last=Johnson |first=Steven |last2=Iziev |first2=Nikita |date=2022-04-15 |title=A.I. Is Mastering Language. Should We Trust What It Says? |url=https://www.nytimes.com/2022/04/15/magazine/ai-language.html |access-date=2025-05-02 |work=The New York Times |language=en-US |issn=0362-4331}}</ref><ref>{{Cite web |last=Fahey |first=Rob |date=2023-04-21 |title=Could AI spell the end of the immersion-breaking NPC? {{!}} Opinion |url=https://www.gamesindustry.biz/could-llms-spell-the-end-of-the-immersion-breaking-npc-opinion |access-date=2025-05-02 |website=GamesIndustry.biz |language=en}}</ref><ref>{{Cite web |last=Garcia |first=Derek |date=2021-05-27 |title=Why AI Dungeon Is A Work In Progress, And Still Not Safe |url=https://screenrant.com/ai-dungeon-not-safe-children-kids-safeguards-problems/ |access-date=2025-05-02 |website=ScreenRant |language=en}}</ref> In 2023, both ''[[Wired (magazine)|Wired]]'' and ''Bell of Lost Souls'' highlighted the limitations of using [[ChatGPT]] as a dungeon master for ''Dungeons & Dragons''.<ref name=":0">{{Cite news |last=Gilsdorf |first=Ethan |title=ChatGPT’s Storytelling Chops Are No Match for Dungeons & Dragons |url=https://www.wired.com/story/chatgpt-learn-from-dungeons-and-dragons/ |access-date=2025-05-02 |work=Wired |language=en-US |issn=1059-1028}}</ref><ref name=":1">{{Cite web |last=Zambrano |first=J. R. |date=2023-05-26 |title=We Played D&D With a Chat-GPT Dungeon Master – Can AI Really Run Your Campaign? |url=https://www.belloflostsouls.net/2023/05/we-played-dd-with-a-chat-gpt-dungeon-master-can-ai-really-run-your-campaign.html |access-date=2025-05-02 |website=Bell of Lost Souls |language=en}}</ref> ''Wired'' commented that "ChatDM's taste for fantasy was often a bland amalgam of fantasy scenarios harvested from decades of D&D lore and Tolkienesque tropes" and it struggled "to maintain a consistent story".<ref name=":0" /> They noted this experience reminded them "that a good D&D adventure isn't like being told a story by a novelist or storyteller" as instead "the narrative unfolds communally around a table"; however, ChatD&D "ironically" might be "truer to the game's improv-oriented roots" as the "more free-form" nature means neither the players nor the dungeon master have "a clue as to where the adventure will go".<ref name=":0" /> ''Bell of Lost Souls'' noted "asking Chat-GPT to accomplish anything creative really highlights the limits of a Large Language Model" and that while it can produce "great idea-seeds", ChatGPT does not understand "the pacing of a scene in a game, or a story or adventure".<ref name=":1" /> They commented it is "fantastic at helping you iterate" and it can take "a lot of the grunt and guesswork out of the work of ideation" when working on plot development.<ref name=":1" />

''[[Polygon (website)|Polygon]]'', ''[[Boing Boing]]'' and ''[[Wargamer (website)|Wargamer]]'' reported on a July 2024 research paper, by graduate student Pavlos Sakellaridis, which examined the feasibility of a ChatGPT dungeon master built off of ''[[The Sunless Citadel]]'' (2000) [[adventure module]] and transcripts from the [[actual play]] web series ''[[Critical Role]]''.<ref name=":2" /><ref name=":3">{{Cite web |last=Toohey |first=Ellsworth |date=2025-01-14 |title=AI might be your next Dungeon Master |url=https://boingboing.net/2025/01/14/ai-might-be-your-next-dungeon-master.html |access-date=2025-05-02 |website=Boing Boing |language=en-US}}</ref><ref name=":4">{{Cite web |date=2025-01-20 |title=AI dungeon master for DnD railroads players |url=https://www.wargamer.com/dnd/ai-railroads-players |access-date=2025-05-02 |website=Wargamer |language=en-US}}</ref><ref>{{Cite thesis |last=Sakellaridis |first=Pavlos |title=Exploring the Potential of LLM-based Agents as Dungeon Masters in Tabletop Role-playing Games |date=July 2024 |access-date=May 1, 2025 |publisher=[[Utrecht University]] |url=https://studenttheses.uu.nl/bitstream/handle/20.500.12932/47209/Thesis_Final.pdf}}</ref> ''Boing Boing'' noted that "Sakellaridis compared player experiences with both AI and human Dungeon Masters" – the results showed "surprising strengths" for the AI Dungeon Master and "while human DMs maintained a slight edge in most categories, the AI excelled at creating immersive environments, scoring 4.13 out of 5 compared to humans' 3.35".<ref name=":3" /> ''Wargamer'' similarly highlighted that the "results of his tests are interestingly mixed" with player reports rating the human dungeon master as "more competent, created better flow and narrative progression, and elicited more positive feelings overall", however, they "thought the AI DM was better at creating immersive environments".<ref name=":4" /> ''Wargamer'' commented that "Artificial Intelligence is a bit of a misnomer for Chat GPT" since it is a LLM "so the robo DM's ability to create an immersive environment, in a text-based exchange, doesn't mean that it's imaginative. LLMs have been specifically designed to [[Artificial intelligence and copyright#Training AI with copyrighted data|digest large volumes of copyrighted material]] down to patterns, and then generate new text that fits into those patterns when prompted".<ref name=":4" /> ''Polygon'' stated that "the use of generative AI has been a point of repeated contention in the tabletop industry and beyond, with the technology's critics citing its [[Environmental impact of artificial intelligence|environmental impact]] and its foundations on exploitative labor from both workers based in the global south and artists whose work is nonconsensually used to train the tech".<ref name=":2" /> ''Polygon'' highlighted that this academic study used "a complex slurry of variously licensed information, with some sourced from private companies, some sourced from a group of performers, and other materials sourced from volunteers" and it "has raised questions about the hazy nature of [[fan works]]" in relation "to consent in training" of LLMs.<ref name=":2" />