Why Are We Like This?: The AI Architecture of a Co-Creative Storytelling Game

“By separating out character reasoning into a category of first-class actions that players can observe and perform directly, and that are added to the transcript for players to elaborate on within their stories, we hope to expose character reasoning to the players explicitly—and, thereby, to mitigate the Tale-Spin effe…”

“Many simulation-driven narrative systems [7, 17, 24] model character knowledge phenomena at a fine-grained level, tracking per-character awareness of many indi-”

“The same is true of co-creative writing processes driven by language models [3, 15, 21, 27].”

“WAWLT represents an example of AI-based game design [6] inspired by the study of existing player storytelling practices [5, 12] in simulation-driven games.”

“WAWLT is built around story sifting [22, 23] in both its implementation and its design, making central use of the story sifting and simulation engine Felt [11].”

“Several existing play experiences [8, 9, 26] make use of story sifting technology to some extent, but we believe WAWLT is the first to center story sifting as a player-facing game mechanic.”

“By allowing players to annotate events with their own descriptions, we aim to provide support for extrapolative narrativization [12]: a player storytelling behavior in which players seize on and elaborate minor details in the stories they tell about their play experiences, regardless of whether these details are expli…”

“Many simulation-driven narrative systems [7, 17, 24] model character knowledge phenomena at a fine-grained level, tracking per-character awareness of many indi-”

“Other casual creators for storytelling, such as Writing Buddy [25], and other mixed-initiative co-creative storytelling systems, such as Mimisbrunnur [28], have provided valuable design inspiration for WAWLT, but have not fully embraced the use of a fine-grained simulated storyworld as we aim to here.”

“Other casual creators for storytelling, such as Writing Buddy [25], and other mixed-initiative co-creative storytelling systems, such as Mimisbrunnur [28], have provided valuable design inspiration for WAWLT, but have not fully embraced the use of a fine-grained simulated storyworld as we aim to here.”

“WAWLT is built around story sifting [22, 23] in both its implementation and its design, making central use of the story sifting and simulation engine Felt [11].”

“This bootstrapping of social context mitigates fear of the blank canvas [4, 13] in players by providing an interesting starting scenario, with many sites of narrative potential for players to develop through play.”

“The same is true of co-creative writing processes driven by language models [3, 15, 21, 27].”

“Many of its storytelling mechanics were inspired by similar mechanics from tabletop storytelling games [2, 19, 20].”

“Advances in large language models have allowed new possibilities for their use in storytelling, such as machine-in-the-loop creative writing (Clark et al., 2018; Kreminski et al., 2020; Akoury et al., 2020) and narrative generation for games (Raley and Hua, 2020).”

“For instance, Arnold et al. [5] identified a trade-off between efficiency and ideation: while short single-word suggestions enhance efficiency without inspiring creativity, long multi-word suggestions offer inspiration but can be distracting.”

“Human-AI co-creation has recently been investigated in domains such as drawing [18, 20, 42, 43, 59], music composition [54], creative writing [26, 37], design ideation [46, 51, 69], visual metaphors [11], video game content generation [28, 47], improvisation [38], public displays [52], and dance [53].”

“For instance, Kreminski et al. [40], Stefánsson’s Mimisbrunnr [63], and the WordCraft paper [74] have all explored how generative AI can assist in organizing thoughts or actively contribute to generating text.”

“While we acknowledge that our experiments are generative in nature, we envision them as part of a larger co-creative framework akin to the framework presented in [31].”

“With the advancement of ML, many researchers have explored human-AI co-creation in creative domains including writing [4, 12, 22], music [29], drawing [7, 15, 20, 37], etc.”

“One approach involves sifting the generated storyline to identify compelling event patterns (Kreminski et al. 2020).”

“While existing studies of creativity focus on professional writing, including co-writing with LLMs (Akoury et al. 2020; Clark et al. 2018; Ippolito et al. 2022; Kreminski et al. 2020; Mirowski et al. 2022), this paper contrasts the outputs of non-professional writers with those of LLMs.”

“Other tools provide support for focused tasks, such as generating customized thesauruses [20], suggesting connections for new metaphors [21], and building characters through chat [54] or simulation [42].”

“…[40] ranging from creative visual storytelling with robots [49, 86, 87] to emergent narrative with co-creative interactive storytelling characters [42, 43, 85], this work takes a step back from the technology itself to better understand the human creative process of storytelling resulting from a…”

“Support for planning takes the form of generative ideation, usually presented as suggestions [73, 134], although some systems have a more equal and collaborative storytelling focus that weaves the AI ideas into the story text [70, 91].”

“…many advancements in AI image generation [4, 44, 74, 129], as well as cinematic [27, 72, 124, 125] and literary systems more broadly [41, 46, 48, 63, 118], these kinds of visual narrative outputs are distinct from those that specifcally derive stories from an image or image sequence, which is…”

“, wherein four of five playtesters re-ported a sense of directionlessness at least once during play (Kreminski et al. 2020a).”

“…et al., 2019b; Kreminski and Wardrip-Fruin, 2019) and to provide computational support for playful creative writing at the plot-structure level (Kreminski et al., 2020a) have revealed a preliminary inventory of several distinct but interrelated creativity support needs among creative writers,…”

“Some allowed the design of artificial characters and world elements that can autonomously interact with the user [52, 78].”

“These include systems for cocreating platform game levels [48], strategy maps [34], and certain aspects of narrative [28, 29].”

“(WAWLT) where two players can develop a story transcript while supported by an AI system with tools to inspect the story world and with suggestions to direct the plot [25].”

“And several recent efforts have been made to construct digital IEN games in which the intended player experience is one of coauthorship, rather than one of participation [30, 10–12].”

“(Kreminski et al. 2020a,b).”

“And several recent IEN play experiences [29, 11–13] have been explicitly designed with the goal of supporting player storytelling practices.”

“With these efforts made in developing human-AI collaborative writing systems and collecting human-AI writing datasets (Coenen et al., 2021; Kreminski et al., 2020), researchers have started investigating the human writing behaviors in this particular setting with the goal of effectively supporting…”

“Why Are We Like This? is a digital storytelling game that uses AI as a mixed-initiative component [12].”

“Previously, the tendency to use rule-based AI as the core for story generation stemmed from the inability of language models to generate coherent writing suggestions and narratives, which could potentially distract players [20].”

“Since an affordance of AI in education is providing individualized learning experiences for students (Kuleto et al., 2021), and given free access to the language models, SDKs and website hosting, we built four text generators for the study.”

“Storytelling, is a widely accepted pedagogical approach (Landrum et al., 2019) and a remarkable element in game design (Kreminski et al., 2020).”

“Beyond general frameworks, many practical systems have been developed to support AI-assisted story ideation and generation [6, 9, 17, 26, 37], speculative fiction writing [45], application drafting [11], argumen-tative writing [50], among others.”

“Some examples include level design [1]–[8], the creation of visual game assets [9], the design of game mechanics [10], [11], narrative development [12], [13] and even the creation of of fully working games [14], [15].”

“AI can support several activities of storytelling, e.g. monitor data sources, collect data, define characters, develop the plot over time, visualize data, measure success, etc. (Al-Doulat et al., 2020; Dur, 2012; Kreminski et al., 2020; Thorne, 2020; Yang et al., 2019 ).”

“Another line focuses on generating narrative text games or using AI tools to generate or act in games (Ammanabrolu et al., 2020; Li et al., 2024a; Latitude Inc., 2023; Wang et al., 2023a; Zhu et al., 2023; Kreminski et al., 2020; Cui et al., 2023; Zhou et al., 2023; Dambekodi et al., 2020).”

“…(Liu et al., 2026a), tool manipulation (Prasongpongchai et al., 2025), scholarly recommendation (Siangliulue et al., 2026), storytelling games (Kreminski et al., 2020), web search (Huang and Narechania, 2026), problem-solving (Luo et al., 2026), and group collaboration (Mukhopadhyay et al.,…”

“GenAI has entered the game industry, promising to support creativity through content generation guided by human input [82, 83].”

“Branching editors such as Twine, storylet-based structures, and progression visualizations offer different ways to organize and inspect narrative possibilities [11, 17, 25, 27, 50].”

“Further work in HCI and creative HCI on computational narratives[36, 37] and the use of AI for reflection tools is also particularly salient[2, 4].”

“Researchers have since developed tools for a range of creative writing contexts, including playwriting [21], screenwriting [46], poetry [13, 18], and games [22, 33].”

“The bundled example worlds and authoring tools relate Shadow-Loom to the broader procedural-content-generation tradition (Smelik et al., 2014; Sum-merville et al., 2018; Kreminski et al., 2020).”

“Prior research shows that AI writing assistance can help with idea generation, narrative development, collaborative storytelling, and iterative refinement, thereby improving both writing productivity and the quality of written output (Kreminski et al., 2020; Lee et al., 2022; Singh et al., 2023).”

“Other work introduced gamified designs such as temporary rules and goals [51, 22, 23, 41], as well as scores [7] into mixed-initiative co-writing to make the system more enjoyable to novice writers.”

“2020a) for more information about the design of WAWLT, and (Kreminski et al. 2020b) for more information about its AI architecture.”