Procedural generation, AI, and Storytelling
This is the second in a series of blog posts by producer Hannah Nicklin featuring in-depth interviews with the Forest of Sleep team, focussing on different aspects of the art, animation, sound music, inner-workings and influences shaping the game. Ed can be found on Twitter as @edclef.
Ed is the second of our team members I'm talking to as part of this series of long-ish reads investigating the making of Forest of Sleep. You could, perhaps, be faintly surprised he's not the first, as many people know Ed as the 'face' of Twisted Tree Games. I can already hear him cringing at that, though. After 5 months of working as a producer with Ed I've developed a keen sense for him cringing, even when hundreds of miles away.
Ed is founder of Twisted Tree Games, but it's a games label he primarily put together to release the games he makes in collaboration with others. He's really keen to emphasise the collaboration bit: Proteus was created by both Ed and David Kanaga, who in turn worked on it with many other game developers, designers, and friends. In the same way, Forest of Sleep is born out of a central collaboration between Ed and Nicolai Troshinsky, but they have built a team around them, all of whom influence the design and development of the game. Indeed, Forest of Sleep didn't really come alive in its current form until Ed began to bounce ideas back and forth with Nicolai, grappling with the question of procedural generation in stories. Initially, he and Nicolai kept talking about 'a roguelike in a forest':
We were talking about making a kind of game where you go through several encounters, it was a very glib description of it, but that was our thinking: a roguelike in a forest. Then there was some point at which Nicolai and I started talking about generating stories.
This shift happened gradually over 2015, made up of design discussions, mockups, useful disagreements and several conceptual iterations until they reached a common and agreed upon understanding of what they were setting out to make. Something that Ed says is "more complicated but hopefully also more interesting than the original idea." I ask Ed if Forest of Sleep follows on from any of the themes or questions at the heart of Proteus. He explains that initially:
[...] the expedition game idea followed on in a much more obvious way: as an aspect of walking outside and exploring in nature that is deliberately completely absent from Proteus. So it started off as the negative of Proteus. Then it became more about stories, and now it feels like the through-line is about creating a sort of reflective space to play in, and to create your own connections and meanings for things. That's a question for procedurally generated content: how does it come to mean something for someone playing? This is a philosophical rabbit hole that I'm quite obsessed with: "how does meaning arise from partial randomness inside a system?"
Ed's outlook on this is influenced by a close interest in the mechanics of divination methods like tarot cards or I Ching – systems for emergent story: systems that allow authoring to emerge between game and player. Ed explains how in tarot, or I Ching:
They're relying on random combinations, and the 'deck' itself is just an incredibly dense intricately authored object: the amount of association in the cards or in the I Ching text is deliberately huge, that's part of the mechanics of it. Even so, when you take a combination of those things it has meaning because of how you invest in it, how you frame it as a person interacting with that object. It allows meanings that aren't designed into that object to arise, meanings other than the author-imagined ones. That feels like a good guideline for me, for how to approach Forest of Sleep. I want to make something that's an interpretive experience; I want to make something that has that openness and strength of lending itself to interpretation rather than conveying a single message.
Ed is clear, however, that at the heart of the Story System in Forest of Sleep isn't an intention to replicate a human storyteller, but rather to make an instrument that the player plays with. He doesn't want us to build a game that can make up a story in the same way a human can, but rather to explore the player's storytelling instincts through a system that allows story and meaning to emerge. This is something that crosses over into the realms of computational creativity - an area of specialisation that didn't exist in AI when Ed studied it at university, but which is now a growing field. Ed has some quibbles with calling Forest of Sleep an 'AI system' though:
What I'm trying to do is definitely a computational creativity question, but I'm coming at it from a different angle [...] 'solving' AI problems in games particularly can often be a sort of Quixotic thing where people spend years and years trying to generate [human-like] stories. Whereas I often think it's better to think of the feel of the result and not just direct simulation. I think of things like Tales of Arabian Nights, and other combinatorial board games; a deck of cards where you combine them to create something interesting. [...] I'm interested in human-centred AI, rather than "can we make an AI do a thing that a human can?"
Ed is describing an aspect of computational creativity which is interested in trying to figure out how humans work through simulating them. But, as he clarifies, that's not what Forest of Sleep is trying to do - we're not trying to simulate human storytelling. The game will, instead, be a framework for associations that make a space for story to emerge between game system and player.
Stories are already engines for human imagination, folk tales doubly so (Forest of Sleep focuses on the Eastern European tradition, which has particularly influenced Nicolai). Folk tales are made up of patterns and constituent building blocks, character archetypes, and familiar settings that in their simplicity leave room for inferred nuance.
We're trying to make something that's interesting to play, and which the player can push back against. Both in the sense of leaving gaps and letting the player fill those gaps with their imagination (which also relies on us framing things in a way that feels important enough that you might want to fill those gaps) and letting the player show what they're interested in by how they interact with the game.
We're doing this thing of reacting to the player, taking things from them, transforming and giving them back, rather than generating a story and the player just walking through it.
Ed explains how this is actually responding to older ideas about AI - such as the ELIZA chatbot from the 1960s, based on Rogerian psychological practices. The ELIZA bot was encoded with certain tactics of questioning that were effective because they were open enough to allow the user room to infer, and respond; to self-reflect. In this same sense Forest of Sleep is a framework for inferred storytelling. One that, Ed hopes, works on the basis of transparency: it doesn't pretend to be human, and it doesn't pretend there aren't humans who designed the framework: but in not-pretending, the game leaves room for the player-as-author's imagination to go to work.
In games you could suggest that there are two overarching trends to storytelling: largely railroaded, carefully authored experiences – and that's not a disparagement, a novel or a film also relies on a level of immersion and authorship to tell a particular kind of story. And then freer, less immersive*, more interactive, responsive frameworks for play; divination systems, folk tale traditions, tabletop and paper RPGs; these frameworks take the opposite approach. They expose the system, don't try to vanish the edges between 'authored' and 'experience' and instead make clear how that system is authored, and that the player's place in that system is one of equal authorship – leading (one hopes) to emergent practices – things that come up from the interplay between player and game, that the designer never directly envisaged.
(*immersive in the sense of 'washing away' reality and immersing you in a new totally authored universe, more interactive: handing you the tools for play and a place to play with them.)
We're creating something where hopefully the game reflects back on the player, allowing them to tell a story to themselves: it's a framework. Also, there's always in any system an author - all systems have some authorial content and conceits built into them - we're not trying to make a completely blank slate which doesn't mean anything until a person does something with it. We're creating a structure that will hopefully lend the gameplay a dramatic shape, and building formula for storytelling, made out of authored parts.
Humans are storytelling animals. There are evolutionary and psychological theories which talk about the importance of 'the storied self' (how we construct our identities through storytelling) – the act of tool building which allowed our species to grow in the first place required storied, imaginative leaps. And as many people talk about the problems of 'post-truth politics' it seems to me that at the heart of the struggles of our contemporary global community are ones driven by powerful, sometime problematic stories. One could suggest that one vital tool in these struggles; political, social, psychosocial; is story literacy.
Personally (let's not attempt to vanish my authorship at the heart of this post) this is one of the things that drew me to work with Ed on Forest of Sleep in the first place. In an era of mass media, folk and oral storytelling culture has somewhat fallen by the wayside, and it concerns me that the power of storytelling is increasingly centralised, and perpetuated by those with vested interests. The way to re-cultivate our ability to take the role of author and storyteller – and to unpick the stories of others - isn't necessarily to return to what came before, rather to find new ways of investigating authorship and the act of story. Games can be one of those places.
Of course, that's just a backdrop, a context. Mostly, Ed hopes that he's making a pleasing, explorative framework for playful storytelling. A fun thing.
I certainly hope for it to be surprising and replayable and... beautiful in a moment-to-moment sense: the visuals and the music, and having individual surprising funny situations. I also hope it works as story, that it creates stories you feel invested in, that have a tension to them, and a sense of evolution. I hope it's not just intellectually difficult as an exercise, but that it stands up in it's own right as a fun thing. And something that appeals to anyone: accessible to people without games literacy, but also fun for gamers in the sense that it's novel and interesting and plays with and subverts some tropes that you find in videogames, with lots of fresh aesthetics in it. Also... in a lot of ways I don't know what we're making! It's important to make things that you're uncertain about.