Human First AI: How We're Building The Narrator

The Narrator is a work of love. Three good friends, with nearly a hundred years of tabletop experience between us, wanting to take that experience somewhere new using the newest technology we have.

AI has faults. Many, many faults. But it is also the first technology that could ever give us a true tabletop roleplaying experience in a digital form. Video games are great, but they never offer the sheer breadth of choice you get playing a TTRPG. Baldur's Gate is a fantastic adaptation of tabletop for the video game space, and it still can't match the freedom of a real game run by a real person. AI can. It can adapt to the choices players make and think through the consequences. It can cope with the chaotic creativity of a human mind, even when it can't match that creativity itself.

So we decided to see what it can do, as a GM. To see what we can do with it, as gamers and creators. To see what YOU can do with it, as players.

The problem

AI is full of ethical dilemmas, and our system is no exception. People have real concerns: that AI steals intellectual property, that it shows no creativity of its own, that it invades gaming spaces that were sacred without it. We hear all of that as real and founded. Frankly, we share a lot of it.

But we also know AI makes some things possible that never were possible before. Amazing things. The trick is building those amazing things without abandoning our own moral philosophy along the way.

I personally don't have a great feeling about where AI is headed when we hand it autonomous control over things like missile strikes and nuclear codes. I have seen The Matrix. And The Terminator. But that's why WE don't use AI for warfare. We use it to play games.

And if we're going to use it at all, we owe the people it affects some honesty. We want to be mindful of the creators AI touches, and the people who lose work to it. We want to respect the artists whose work was stolen by AI scraping. We want creators credited and paid for what they make. We want to employ humans first, as a rule.

Our answer: Human First AI

That's what we call it. What it means, in plain English, is this.

Humans are the creative engine of our games, and human creativity is only ever supplemented or narrated by the AI. We value humans first. We use humans first to create things, wherever we can. And if we can't, we're honest about it.

This means human-authored stories, played by humans. Fully authored stories that still give players the freedom to go anywhere and respond however they can imagine. The AI runs the game and narrates as it goes. That's it. We put human authors, artists, and eventually voice actors first, and we only let the AI fill in the gaps.

Think of it like the DNA in Jurassic Park. Dinosaur DNA made up the bulk of the sequence, and only the gaps were filled in with DNA from amphibians, like frogs. The Narrator's DNA is human-authored, human-created content. The AI only handles the gaps.

Yes, that didn't work out great for the folks in Jurassic Park. But we'd argue hubris played a bigger part in that than the frog DNA did. Avoiding that exact mistake is why we're having this conversation in the first place.

What that means in practice

Saying "human first" is easy. Here's what it has actually cost us so far.

The stories. First, we had to decide how much we were willing to let the AI write, and how much improvising we were comfortable handing it. Early versions had the AI generate the whole story, every line of prose included. But it got hard to tell ourselves there was anything human-first about that. So we switched to fully human-authored story seeds: every beat, every NPC, every optional path written by a person. That was our first real step toward the company we want to be.

The model. The second step was looking into an ethically sourced language model, the AI itself. This is where we hit our first real setback. A few models are built on cleaner, consented training data, and we looked closely at two. Comma is an open model trained on openly licensed text. Apertus is a model from the Swiss government, built to respect creators who opt out of AI training. We were genuinely excited by both.

But our research found that the cleaner models available today can't yet handle what a live game demands: the interpretation, the improvisation, the constant coping with the unexpected. So for now, the only way to move this dream forward is to use the same kind of public model everyone else in the space uses. That means, today, we run on a commercial model, and we won't pretend its training was clean.

We're not happy about it, and we're not going to hide it. We're watching this space closely, and the moment an ethical model can handle the thinking a TTRPG needs, we'll move to it. We're genuinely looking forward to that day, because it's the last piece of the puzzle.

The data. On privacy and data, we draw a hard line for our customers. We will not use your data to train any AI model, even our own. We only use data to improve the player experience and improve our product, never to train an AI. We will never sell your data to a third party. Your data is safe with The Narrator. Privacy and data protection are among our highest concerns.

The art. Next, we threw out all the AI-generated placeholder art and replaced it with work made by real, paid human beings. We can't afford much as a startup, so we kept it in house. In this case, literally: our first (and so far only) paid artist is Scott's eight-year-old son, Arthur, who was paid in Funko Pops and in D&D sessions run by Uncle Sean. Every piece of art on our site is created by real human beings.

The voices. Finally, voice narration will use AI trained on real, paid actors. No generic AI voices. Only voices trained on actors who are credited.

The vision

Put it all together and that's our idea of a human-first AI company. One that meets these problems head on, because they're real and they're legitimate. They're also solvable, as long as we're willing to try.

So join the waitlist to get early access to the alpha, and support the Human First approach to AI TTRPGs.

Sean Burke