From Dungeon Masters to Data Scientists: the AI lessons in D&D

Freemavens active Dungeons and Dragons (D&D) club is a testament to our unique culture, with seasoned players regularly introducing the joys of character crafting, rolling unconventional dice, and collectively constructing epic stories to newcomers.

At first, the notion of sitting down with your colleagues to learn a new game while assuming the role of a gnome artificer wizard with a dodgy Somerset accent (a nod to my mate and AI expert Pete Trainor), out on a mission of repelling a dragon attack by rolling a 20-sided die, might seem a tad overwhelming. However, teams that bond over fantastical journeys often discover they’re better prepared to table the very real adventures that await them in their day-to-day work.

At this point you may be wondering what in holy Faerûn is going on in this article supposedly addressing and exploring AI (because we all know from LinkedIn that NO ONE IS SPAMMING ARTICLES EVERYDAY ABOUT AI). However, fret not, the threads of imagination and technology are intricately woven together, and we’re about to delve into the captivating aspirations and challenges that arise when Dungeons & Dragons meets the world of Artificial Intelligence.

Some Dungeon Masters are following a script

In D&D, the Dungeon Master (DM) assumes the role of the world’s creator and serves as the guardians of the rules, lore, and narrative possibilities that your party can explore.  Much like the DM, AI models take on the role of guiding and shaping the world of information and responses, maintaining the rules and narrative potential for your interactions.

Some DMs follow a reactive approach with limited data. They use pre-written adventure books and religiously follow the provided storyline, allowing specific player choices to unfold within defined boundaries. When the narrative is “fixed” in this context, it is colloquially known as “rail roading the players”. They feel they have agency and freedom of choice but are being channelled into fixed outcomes.

This all falls under the category of reactive AI. The system utilises a dataset, likened to an ‘adventure book’ and it responds to user prompts, much like a Dungeon Master reacts to a D&D player’s actions in our metaphor.

This doesn’t diminish the enjoyment of the game; rather, it streamlines the possible story directions. Player choices can still enrich the narrative or lead to permanent consequences, but the DM ensures the overall coherence.

But what happens when the DM (or AI) learns and evolves? Fostering creativity to mould fresh characters to encounter, adversaries to conquer, and uncharted territories to venture into? To achieve this, you must level up your Dungeon Master.

What makes an epic Dungeon Master 

To stand out as a DM, you must cultivate and foster creativity, viewing it through the perspectives of many, especially those new to the game. How do you achieve this? By ‘homebrewing’ your campaigns—stepping away from pre-written adventure books. These DM’s meticulously construct a sandbox of possibilities, weaving hooks, and threads of potential narratives, and adapting the story in real time based on the players’ agency, curiosity, and inventive thinking.

This embodies the essence of generative AI. The DM is not only incorporating player inputs but also expanding the datasets to enhance the campaign world, preparing it for the potential new directions the story might take.

This demands extensive preparation from the Dungeon Master, and it’s undeniably challenging. The hours spent in preparation far surpass the actual gameplay hours around the table. This metaphor draws parallels to the computational power and parameter expansion that the most advanced AI models are striving to achieve.

Poor Dungeon Maters allow their biases to impact players 

Individuals who’ve given D&D a try but swore off playing it again often give the criticism that the Dungeon Master was consistently aiming to thwart the players, while the party’s primary focus seemed to be on acquiring wealth and plundering every challenge they encountered.

Truth be told there are both inept Dungeon Masters’ and players, and much like in the realm of AI, they often fall prey to similar issues.

Relevant to AI, one of the most concerning parallels is how a Dungeon Masters’ biases can permeate the players’ experiences. In AI, the selective gathering of data tends to favour the most prolific sources, leading to a significant bias in favour of the capitalist industrial machine and the most outspoken voices. This bias can potentially exclude certain behaviours, cultures, and backgrounds given that the majority of AI data is sourced from the internet, which is significantly skewed in favour of Western cultures.

In the realm of D&D, this can manifest with Dungeon Masters’ spending most of their time accommodate the most vocal and dominant players. Unfortunately, this feedback loop often results in disproportionate attention on one specific person rather than the group at large.

My concern in AI is that, without the proper ethical and governance frameworks in place, we might inadvertently foster an environment that lacks regulation, enabling early adopters to influence and invest in promoting behaviours that benefit only a select few, rather than the broader community.

If D&D encounters similar issues, what insights can we extract from it to offer potential solutions and adaptable frameworks for AI? It’s likely that Wizards of the Coast, the owner of D&D in partnership with Hasbro, has not explored this line of thinking in the context of AI. Nonetheless, there may be valuable lessons to uncover.

“Rules” & “code” spoiling or structuring the fun? 

One of the most brilliantly designed aspects around D&D is the creation of a sandbox filled with potential rules. Players aren’t all-powerful from the outset and there are limitations on what can and cannot occur. Joyfully, the game encourages adaptability within this framework with emphasis on fun at the gaming table.

In practice, this approach thrives when everyone at the table aligns on their shared ‘code of fun’ for the game. Exceptional Dungeon Masters often initiate a session to delve into the game world, characters, and establish guiding principles. This not only establishes a clear understanding of the rules but, more importantly, defines the type of shared enjoyment and collective experience aspired to. It provides a vision for the Dungeon Master, prompting us to ponder: what is the vision for AI?

Without this shared code, AI’s implications remain ambiguous. When issues inevitably arise, the question becomes: how does the gaming table discuss and resolve them? It’s important to recognize that governance and regulations are not inherently at odds with innovation. This argument, lazily applied to both AI and society, overlooks the fact that rules actually serve to contain the scope and magnitude of mistakes as we collectively strive to improve the game.

A good Dungeon Master (and their players) understand that players earn their way through adventures, gradually building their skills and abilities to face more formidable challenges. They progress together, embracing certain constraints not to impede but to enhance the growth of both players and the DM, ultimately enriching the game. It’s not about slowing progress; it’s about ensuring that we’re more likely to experience a full narrative capturing the stories of many characters.

Future adventures

To maximise the benefits of AI, our focus should shift from solely increasing the power of these models to the meticulous curation and collection of experiences, just like the approach in D&D.

To bring about this transformation, the AI development community (Dungeon Masters) and the users (players) must exercise greater diligence, explicitness, and reflection on the adventures, whether substantial or minor, we should embark upon rather than merely considering what we could. This approach guarantees that the game evolves alongside the players, minimising the likelihood of stumbling into a dark cave and confronting an ancient dragon completely unprepared.

Absolutely no one, and I emphasize NO ONE, desires a TPK (Total Party Kill).

What we aspire to achieve in AI is a remarkable, ever evolving, and collaborative journey.

Roll on heroes. Roll on.