AI in TTRPGs: Innovation or Invasion?

Since releasing the Quest Portal Assistant, lots of Game Masters, Players, and Publishers have asked how it all works under the hood and if we plan on replacing humans with an AI Game Master. It’s a great question, so we decided to take a step back and explore the topic of an AI Game Master, how we are thinking about the role of AI, and how our technology works.

Game Masters bear a substantial burden on their shoulders. They have to know the rules, prepare for gaming sessions, manage logistics of getting the group together, keep the story going and roleplay a lot of characters. These are only some of the reasons why there are way more players than GMs in the world.

This GM scarcity means there are more people that want to play than actually can enjoy playing every week. Therefore, it is very understandable why some people find an AI GM an exciting idea. We have already seen some effort made to automate the role of the Game Master, to name a few:

• Mark Zuckerberg's Snoop Dogg Dungeon Master
• ChatGPT and the prompt "I want you to act like a Dungeon Master."
• AI Dungeon from Latitude

As attractive as an always-available-game-master sounds, we need to pause and think about the consequences of removing one of the players around the table. All the players around the table, including the Game Master,  find the experience incredibly rewarding.

We can. But, we have decided to draw a clear line: AI should never be used to reduce the number of people around the table, whether a player or a Game Master.

That leaves us with the question of should AI be used at all?

We believe that many of the barriers in front of people wanting to become Game Masters can be lowered with the help of AI technology. More concretely, AI should be the mech suit available to all GMs, beefing up their confidence to tackle any barrier in front of them. If weaving AI into the world of TTRPGs means we can get more people to play these magical games, then the answer should be a clear YES.

Large language models (LLMs) are the technology that power a lot of generative AI solutions out there. Since they gained popularity with the GPT models being released many have been quick to point out that those models have a tendency to make things up and present them as facts. These errors are sometimes referred to as hallucinations.

Many Game Masters have been experimenting with using ChatGPT to answer questions about the rules and lore of certain gaming systems. This is an area where hallucinations are a real problem as the answer can simply be wrong.

The Quest Portal Assistant uses a method known as retrieval augmented generation for knowledge-intensive NLP tasks, or RAG for short. This method works by understanding user input in natural language, creating embeddings, which are high-dimensional vector representations of text, and measuring semantic similarity between various sources. This allows you to go beyond keyword based search and get results for things that have similar meanings.

The sources of our Assistant are all licensed from publishers that own the content or are publicly available through an open license (CC, OGL, ORC, etc.). In fact, our publishing partners sell access to their content on our platform.

So once you ask the Assistant in your Call of Cthulhu campaign for monsters that are related to water, roughly the following process takes place:

• We send the question to an AI model that classifies the query (e.g., is it chitchat, a question about rules or lore, monster stat blocks, characters, dice rolling, etc.).
• Since this is asking about a monster we create an embedding of the query and search through our vector database for matches to that query.
• Usually matches are found, but they are not equally good. We filter out the snippets of text that don’t have a high matching score. In the case of Call of Cthulhu we are likely to find high quality matches in the Malleus Monstrorum books and the Keeper’s Rulebook.
• We send the original question from the player and the most relevant snippets from the original textbooks to another AI model that answers the question based on the available context.
• We usually answer the player's question within 1-2 seconds, but complex questions can take longer.
• Players can read the answer or click through to open the Library on the right page.
• If players are unhappy with the answer, they can downvote the response, which brings it to our attention. We investigate what went wrong. We've noticed that some errors stem from the original books being vague, so we manually write great answers and store them in a hidden "FAQ" database. For this purpose, we store queries and responses in our internal database.

By being able to scope the knowledge of the AI model we significantly reduce the chance of errors or hallucinations. The Assistant also cites its sources so you can verify that the information presented is accurate or if you’d like to know more about the topic.

Not all hallucinations are bad though, right? When the Assistant is dealing with a request that is not for factual answers then we let them loose. This is where it shines in assisting your creativity. When asking for NPC concepts or place descriptions you want there to be variability, creativity, and flexibility.

If we think about AI in the context of why it is being used, then it adds a bit of nuance to the overall discussion. We think that when it comes to your home game, you should feel free to do whatever you want and whatever your table is comfortable with. It’s not our or anyone else's responsibility to dictate what technology you use or how you use it.

Many players and game masters are already using generative AI technology for creating character art or scenes. They are bringing those creations into their game and sharing it with their table.

The demand for this is very reasonable. It feels very empowering to be able to bring your ideas to life. Generative AI technology has brought the power of creating to everyone and most people are just using it to feel better, to feel creative, to create memories with their friends and family by sharing those creations.

We decided to support your desire for creativity by making the Avatar Generator. Since we released that feature it has been used to illustrate well over 100,000 unique player characters and NPCs. We know that these creations are being used to make the shared experiences in Quest Portal better.

We also understand that some have hesitations about this technology. So, we think it is important for us to be transparent about what powers our AI. Currently, our Avatar Generator runs on top of Stable Diffusion, an open-source text-to-image model from Stability AI. We have created our infrastructure to not be dependent on one solution. We are always exploring newer models to see if they can better meet the needs of our players.

The same story goes for the Assistant and other AI tools we are creating. We use a mix of open-source models and closed-source models to solve problems. Currently, we are using many of OpenAI’s GPT models as well as experimenting with other LLMs.

Looking forward we see generative AI models go to be more ethically sourced. Stable Diffusion version 3 has an opt-out of training option for content creators. We expect new models to come to the market that are purely made from opt-in or public domain sources.

We are keeping a keen eye on these developments and are always looking to improve our user's experience as well as meet their ethical standards (which are high in our space).

Get Quest Portal Pro

Are you ready to take your tabletop role-playing games to the next level? Look no further than Quest Portal Pro—the ultimate companion for unforgettable adventures.

Get started for free

Loading...

We founded Quest Portal to exponentially improve the human experience of playing tabletop roleplaying games. We want to achieve this by building powerful technology that is perfectly cloaked. Technology that lives in the background and works hard to help make your game experience as smooth as it can be. Intuitive, mobile, and intelligent.

AI has been a part of our vision for making TTRPGs more accessible for years and will remain so as we move forward. It's a subject that we, as Quest Portal founders, have known very well since our first forays in the field together in 2008, and we have been baking AI technology into everything we do at Quest Portal.

AI is the very definition of powerful technology, but the acronym is loaded with contention. We know the strengths of AI and fundamentally believe in its ability to bring TTRPGs to millions more players. To us, there is no hesitation - we intend to continue playing a leading role in integrating AI into the roleplaying experience alongside all of the other technologies that we have built over the years.

We proudly embrace AI as a means to grow confidence, lower barriers, and support the imagination of GMs and players. AI-assisted storytelling and AI-generated art are making TTRPGs more accessible and welcoming to more people.

With the launch of Quest Portal Pro we took our first steps in a long journey towards getting more people into this hobby of ours. We invite you to come along. Those that have joined are already giving us feedback we could only hope for. Tom, an early user of the Quest Portal Assistant, puts it best:

You can try out the Quest Portal Assistant with the free trial to our Pro subscription and make a judgment for yourself if this is right for you and your home game.

We sincerely hope you join our adventure.

Gunnar, Gummi, and Steinn

Quest Portal Founders