Hasbro CEO Chris Cocks Talks AI Usage in D&D [UPDATED!]

Hasbro CEO Chris Cocks is convinced that the Dungeons & Dragons franchise will support some kind of AI usage in the future. Speaking today at a Goldman Sachs event, Cocks spoke about how AI products could soon support Dungeons & Dragons and other Hasbro brands. Asked about whether AI has the potential to "bend the cost curve" in terms of entertainment development or digital gaming, and how it's being used in the toy and content industries, Cocks said the following:

"Inside of development, we've already been using AI. It's mostly machine-learning-based AI or proprietary AI as opposed to a ChatGPT approach. We will deploy it significantly and liberally internally as both a knowledge worker aid and as a development aid. I'm probably more excited though about the playful elements of AI. If you look at a typical D&D player....I play with probably 30 or 40 people regularly. There's not a single person who doesn't use AI somehow for either campaign development or character development or story ideas. That's a clear signal that we need to be embracing it. We need to do it carefully, we need to do it responsibly, we need to make sure we pay creators for their work, and we need to make sure we're clear when something is AI-generated. But the themes around using AI to enable user-generated content, using AI to streamline new player introduction, using AI for emergent storytelling, I think you're going to see that not just our hardcore brands like D&D but also multiple of our brands."

Wizards of the Coast representatives has repeatedly said that Dungeons & Dragons is a game made by people for people, as multiple AI controversies has surrounded the brand and its parent company. Wizards updated its freelance contracts to explicitly prohibit use of AI and has pulled down AI-generated artwork that was submitted for Bigby's Presents: Glory of the Giants in 2023 after they learned it was made using AI tools.

A FAQ related to AI specifically notes that "Hasbro has a vast portfolio of 1900+ brands of which Magic: The Gathering and Dungeons & Dragons are two – two very important, cherished brands. Each brand is going to approach its products differently. What is in the best interest of Trivial Pursuit is likely quite different than that of Magic: The Gathering or Dungeons & Dragons." This statement acknowledges that Hasbro may use AI for other brands, while also stating that Wizards is trying to keep AI-generated artwork away from the game. However, while Wizards seems to want to keep AI away from D&D and Magic, their parent company's CEO seems to think that AI and D&D aren't naturally opposed.

---

UPDATE -- Greg Tito, who was WotC's communications director until recently, commented on BlueSky: "I'm deeply mistrustful of AI and don't want people using it anywhere near my D&D campaigns."

 

[tetrasodium]

Because even if you have to parse their words or read between the lines, CEOs are representatives who can speak to which direction the company or product line is going.

Also in this case gives reasonable indication that one of two things could be assumed,
A: Wotc is potentially designing for a style of "play" that is wildly non-reflective of anything seen in the wild.
B: Wotc is headed by someone with such a loose grasp on d&d that they can't even blow wind that is almost adjacent to plausible.

Either of those reflect poorly on expectations for the two remaking core books yet to be released & got a not insignificant amount of discussion over the last 23 pages

 

[billd91]

As badly as we want to believe, human art isn't always an option. Prices for human art are far from universally affordable. And the time frames to commission a piece are long when you count searching for an artist and the many other preliminary steps. And that's not including the actual process of creating the artwork, which can take days or weeks.

Now, maybe I'm an outlier. But I have never had my character concept in my head long enough before a campaign to commission a piece. Nor have I ever been attached enough to that concept to be worth the hundreds of dollars a nice piece can cost. And that is waving any socioeconomic concerns one might have with such a purchase even under ideal circumstances.

I think morals have a place, but much like choosing where to shop, there is much more to this question than just those morals. Sometimes, quite often in fact, the choice is AI art or no art. And in those cases, I wonder if the conversation shouldn't change.

EDIT: This post only applies to casual, private uses in the context of players in a DnD campaign. It is not about broad commercial use.

This is why many of us have clipped art from other works over the years. I can't draw worth a damn, but if I clip a piece off the internet for my PC portrait for my home game, it's not really harming the artist even if it is violating their copyright (though, my horizon for distribution is pretty much just the eyes around my gaming table).

That's still a WORLD of difference from using an AI that's been plundering the work of thousands of artists without permission. That's far beyond the scope of me using art without permission around my table.

 

[Azzy]

Because even if you have to parse their words or read between the lines, CEOs are representatives who can speak to which direction the company or product line is going.

How about reading tea leaves?

 

[DinoInDisguise]

This is why many of us have clipped art from other works over the years. I can't draw worth a damn, but if I clip a piece off the internet for my PC portrait for my home game, it's not really harming the artist even if it is violating their copyright (though, my horizon for distribution is pretty much just the eyes around my gaming table).

That's still a WORLD of difference from using an AI that's been plundering the work of thousands of artists without permission. That's far beyond the scope of me using art without permission around my table.

My post took issue with presenting human artists, professionals in the quoted post, as an alternative for D&D players. Not on the larger issues confronting generative AI as a whole. And was not meant as a commentary on the ethical use of such tools.

I strongly believe that I have no place in lecturing people on the relative morality of such behaviors. And as such I have no intention on weighing in on that. I would, simply, encourage each person to decide for themselves what they are comfortable with.

 

[TiQuinn]

How about reading tea leaves?

Okay. Sure. Whatever.

 

[Velderan]

I spent a chunk of time this summer having the commercial version of ChatGPT (much more powerful) run a short campaign (levels 3-5) for a solo adventure that eventually grew into a party of four.

It was a pretty interesting experience. I would describe it as collaborative; I gave it general instructions as to setting and the very broad adventure ideas, but then it came up with the details, right down to a room by room description of a dungeon, and running combat.

What was really cool was that it would spin adventures out of my actions. So if it described a shady-looking character in a tavern and my character spoke to them, they would typically drop hints about some sort of intrigue, lost temple, or whatever, and I could opt to follow up, or not. It felt very sand box-y.

The generative AI was very good at doing character interactions as I roleplayed the PCs and it did all the NPCs. It was also very good at coming up with vivid descriptions. It was better at math than previous iterations, but still fudged all the time, rather than really generating random numbers. It was forgetful, so in that dungeon it would sometimes give wrong directions and contradict earlier descriptions. Despite me repeatedly asking it to make the encounters more challenging, it had trouble doing so.

Also, it could add art to illustrate what the characters were seeing, which was super cool!

It was a fun experience, and with me taking an active role (often by actively rolling dice) I had a new D&D-related experience that was fun and memorable. But it was by no means independently DMing. I was leading it, so even though I didn't know where the adventure was headed, my active engagement was needed to keep it going off the rails. Nor do I think the current iteration of ChatGPT is capable of being an independent DM, due to the way it is designed.

However, I do think it would be possible to make it much better without too much work. Taking that basic model of ChatGPT and adding a better math function, maps, and adventure guidelines could allow it to be much more independent and approximate a human DM far more effectively. It does not seem like a big step, is what I'm saying. I can see why WotC and others are pursuing it.

I have played around with the latest GPT4o version as well to run a campaign for me and generally experienced the same thing. I had 3 PCs I provided instructions for and it frequently got details mixed up about the 3 of them, so interactions with NPCs were inconsistent with how the personalities it gave the NPCs would respond. Names were frequently from Tolkien, so there was a lot of repetition of names which to be fair happens with an actual DM if they just roll on a d20 table to come up with NPC names as they go. The responses to my prompts would usually be favorable, so I had to be careful if I didn’t want to find whatever mcguffin it had me looking for. For example, “I continue on to the next room of the temple in search of the idol” would likely have me find it in the next room even if that was only the second room and the priest in town had told me the idol was “deep in the temple”.

As for the images generated if 1 PC was an elf, every character it depicted would have pointed ears. Clarifying instructions to correct the image were completely hit or miss on what it understood in the corrected image. Still it was interesting to see what it could do.

That all being said, this interaction was also with the GPT4o version so it still needs some work.

 

[Oofta]

I have played around with the latest GPT4o version as well to run a campaign for me and generally experienced the same thing. I had 3 PCs I provided instructions for and it frequently got details mixed up about the 3 of them, so interactions with NPCs were inconsistent with how the personalities it gave the NPCs would respond. Names were frequently from Tolkien, so there was a lot of repetition of names which to be fair happens with an actual DM if they just roll on a d20 table to come up with NPC names as they go. The responses to my prompts would usually be favorable, so I had to be careful if I didn’t want to find whatever mcguffin it had me looking for. For example, “I continue on to the next room of the temple in search of the idol” would likely have me find it in the next room even if that was only the second room and the priest in town had told me the idol was “deep in the temple”.

As for the images generated if 1 PC was an elf, every character it depicted would have pointed ears. Clarifying instructions to correct the image were completely hit or miss on what it understood in the corrected image. Still it was interesting to see what it could do.

That all being said, this interaction was also with the GPT4o version so it still needs some work.

View attachment 379921

View attachment 379922

AI doesn't think or process information, it's just a really advanced autocomplete. On the other hand, autocomplete can be handy sometimes, you just have to accept it's limitations. Same with AI, useful in some situations as long as you accept it's limitations.

 

[LuisCarlos17f]

Can LoRAs to be protected by copyright? For example one to create gem dragoborns.

And there are some risks if everything is created automatically. For example the bard character flirts with that maid (a nPC created radomly by an AI). But later in the night when they are together alone... she is a theriantrope, and then you can guess the unpleasing surprise for the player, because this was not ready, or to kill certain monsters PCs needed some special weapon they lacked.

If there is some AI software to create quests, then this would be used in the videogame industry.

 

[Clint_L]

AI doesn't think or process information, it's just a really advanced autocomplete. On the other hand, autocomplete can be handy sometimes, you just have to accept it's limitations. Same with AI, useful in some situations as long as you accept it's limitations.

I assume you are talking about generative AI? If so, then characterizing it as simply “advanced autocomplete” is woefully oversimplified. Humans and chimps are both great apes with brains, but one can do a lot more with it than the other.

There is a huge and rapidly expanding body of research on the unexpected capacities of generative AI. To me, describing it as advanced autocomplete is like whistling in the graveyard.

 

[Oofta]

I assume you are talking about generative AI? If so, then characterizing it as simply “advanced autocomplete” is woefully oversimplified. Humans and chimps are both great apes with brains, but one can do a lot more with it than the other.

There is a huge and rapidly expanding body of research on the unexpected capacities of generative AI. To me, describing it as advanced autocomplete is like whistling in the graveyard.

My point is that it doesn't think, it doesn't analyze. It makes correlations using pattern matching. Yes, it's not "just" an autocomplete function much that a kid's peddle cart isn't a racecar. But AI is also not intelligent, at least not yet. We may or may not get there someday and advanced enough LLMs can certainly fool some of the people some of the time. But many think it's going to take a different approach or combination of approaches to have actual intelligence.

Call me a skeptic, and maybe the people that say we're on the verge of the singularity are right. But I think a different approach, like people that are starting to model the brain of a fruit fly and how it navigates are more likely on the path to true intelligence all while using a tiny, tiny fraction of the energy and complexity used to complete the same task with an AI system are the direction we need to go.

In any case, ask me again in 10 years and we'll see who's right. If the new AI overlords allow us to have that conversation.