Populating the Metaverse

Many rules of the metaverse were inspired by Second Life

Metaverses have been with us for a while. Our CEO Guy and COO Annette built islands, games and social experiences in Second Life, the 3D virtual world launched in 2003. Second Life had a lot going for it. It was a fluid 3D world where you could be who you wanted to be, buy land, and build anything - from a small house, to a multiplayer racing track. There were training islands for Accenture, art galleries, parkour experiences and virtual cinemas.

For me, it was Habbo Hotel, the online community where I spent hours customizing hotel rooms, designing mini-games and trading ‘furni’. I cared for virtual pets and earned virtual currency to spend on rare outfits.

These were not games, but more like 3D social networks before Facebook even existed.

Almost 20 years later, we have graduated from virtual worlds to The Metaverse, and, if the analysts are right, the scale will be much bigger than the first time around.

The Metaverse demands richness and depth. It needs more than mini-games, customisable avatars and gloriously surreal virtual spaces. In order to sustain the perceived virtual universe, it needs story. And the key to good story is rich, dynamic and, most importantly, interesting characters. 

However, these are different to movie characters, different to characters in books, and even (definitely) those in games. These are characters whose journeys you can join and play a role in for weeks, months, even years. 

This brings new and complex storytelling challenges, and it’s likely some of that complexity could benefit from creative support from AI systems like GPT-3 to help build  hyper-personalised experiences.  Our ventures into projects like Electric Sheep are already exploring how this will work.

To create these characters, we need to think differently about how we forge them, which is why - hot on the heels of Epic Games’ release of Metahumans - we have developed a Universal Character Model together with the Entertainment Technology Center at the University of Southern California, Digital Domain, Fable Studios and Dramatica.

The Universal Character Model defines the laws of any specific character; who they are, their motivations, their goal, and their methodology to achieve it. It is needed to create believable character-based experiences, reliable story environments and compelling stories for audiences that are cast inside the story.

The Introduction to the Universal Character Model is below, as is the link to the full ePub version for free download on the ETC website.

- + -

Introduction to a Universal Character Model

Relationships between audiences and stories are changing. After years of audiences going unnoticed by the characters they invest so much in, audiences are finally being recognized for the essential part they play in storytelling. Now, they are welcomed into dynamic story worlds that adapt to them personally from the moment they arrive.

Dynamic story worlds change depending on audience influence, and demand both strong characters and strong environments. The games industry often showcases strong environments, with systemic design and emergent narrative. The film industry gives us strong characters with personalities that we feel we know. Interactive stories give us the chance to combine both in a way where the sum is greater than its parts, because the audience is inside the story. And the audience brings complexity. They bring their own mind and their own ambitions which inherently tests the strength of our characters. A Universal Character Model (UCM) encompasses who the character is and can be utilized in a way that, no matter how the audience influences our story, the character will respond accordingly.

This paper presents the initial conceptual requirements of a UCM that are vital for creating and advancing immersive stories. We will focus on what a UCM must be capable of doing, without suggesting which particular technologies available today should be relied upon. This is because we intend the UCM to be a modular and expandable solution to interactive storytelling, one which is open to new technologies as they become available. The main requirement for the model is to be an understandable tool to improve interactive experiences for both the creator and the audience.

Broadly, a UCM is a standardized data structure that creates a character definition containing all fundamental features of a character (personality, backstory, motivations, likes/dislikes, quirks, etc). It will be populated by the author and then inserted into an interactive narrative, where it can be used as a detailed prompt for AI to generate dynamic character reactions and behavior in response to audience interactions. We argue that such a structure is not only possible, but necessary.

As interactive worlds grow, so does the workload for creatives. Creatives, writers, designers, programmers are having their workload grow exponentially with every personalized thread of character and story. Opening our story worlds for audiences to influence means the amount of content required is quickly outstripping contemporary techniques, such as hand-crafted Decision Trees. These are costly, difficult to organize, and profoundly limit the possibility space of interactions to an insufficient number of narrative inflection points. Pressures on production budgets, time scales and the capabilities of the human mind all limit the depth of immersion we can offer in interactive experiences.

In addition, today we demand that characters live across various media forms (video-games, mobile, TV, film, VR/AR, and more), and we want our immersive experiences to be long-term. It has become unfeasible for writers to individually hand-craft each instance of a character’s actions and dialogue, let alone the dynamics between large groups of characters with complex interactions.

While we believe that AI can be used as a supplementary tool to fill in gaps where our human authors cannot, we do not suggest that AI should use UCM to write our stories for us. With a UCM, the grounds for a new vision of interactive storytelling is possible. A UCM has two objectives. First, for the audience; provide them with rich, dynamic, and narratively coherent story worlds. Second, for the storytellers; provide a tool that allows them to commit to dynamic storytelling.

We will explore the conceptual requirements of a UCM through two perspectives: the data within the UCM itself, and the evolution of authorship.

We will reference the Dramatica Theory of Story (developed by Write Brothers Inc.) as a starting point for the data within the UCM because of its uniquely flexible and logical understanding of narrative. In Dramatica’s system, characters are seen not as discrete individuals, but as bundles of narrative functions. We will discuss Dramatica in some detail, however what is most important to note is that it grants us not only a method with which to distinguish the particular features of characters, but also a predictive schema through which an interactive narrative can, on the fly, be constructed in response to player decisions and inputs.

Next, we argue that a UCM fundamentally alters the evolution of authorship both in terms of the scale of the stories authors create, and how they create them. Once a character is defined within a UCM, an author could weave in AI generated character responses, leaving themselves more time to primarily focus on shaping the overall flow of a narrative, rather than the particulars of each character’s actions and dialogue. Writing becomes much more like directing, and editing like rehearsal.

Of course, hand-crafted character dialogue and action will always be necessary, much like how hand-crafted animation and procedural animation now work alongside one another. However, by shifting an author’s focus to the narrative movement of their story, and allowing a UCM approach to handle a portion of the minute-to-minute interactions, authors will be freed to focus on telling stories and immersing their audience. As an example of how authors might populate and control their character models, we will look at existing authorship tools such as Charisma.ai and its focus on conversation between audiences and characters, powered by Natural Language Processing AI (NLP).

A UCM will become the basis for creating expansive and coherent immersive experiences. We will discuss ways in which it could be utilized, such as a housing for IP to be distributed into new story worlds, a reliable prompt for AI and NLP to produce real-time hyper-personalized experiences, and a way of developing a model that aims to define our audience in the same way it does our characters. 

- + -

Read the full paper here.