AIStories: Exploring Human–Artificial Intelligence Collaboration Through Narratives and Games

Introduction

Today, artificial intelligence (AI) technologies have expanded rapidly across almost every domain of human activity, from health care and finance to welfare, justice, and education.

Among AI technologies, chatbots based on large language models (LLMs) such as Claude or Gemini have become increasingly accessible to the general public. There are at least three features that make these LLMs’ tools different from earlier AI tools and suited for widespread adoption: (1) the ability to interact through natural language rather than programming commands, (2) an intuitive messaging-style interface that (apparently) requires no technical expertise, and (3) a generalist architecture trained on vast amounts of textual data, enabling them to perform a broad range of linguistic and cognitive tasks, basically conversing like a human person.

These characteristics have lowered the entry barrier for both individual and organizational sectors that are implementing these tools for a wide range of tasks. According to Eurostat, AI adoption is growing transversally from cybersecurity to communication sectors. LLM chatbots are used especially for generating creative content. Apparently, among the general public, AI tools are primarily used for personal purposes, followed by professional and learning ones.

In this scenario, it is important to consider the benefits and the risks that individuals and society are facing due to the emergence of LLMs. Generative AI tools have the potential to improve message quality, foster more empathetic communication, give feedback and enhance error detection, offer writing and language support, and improve medical image analysis. ¹ Among the risks are overreliance or improper use, sycophancy (or chatbots’ obsequious behavior that can exacerbate users’ biased reasoning), and generation and spread of inaccurate information. A current situation that could exacerbate those risks is the lack of a shared ethical framework regarding implementation. Such phenomena can lead to issues and suboptimal decisions in high-stakes domains such as health care, psychological support, legal advice, and even individual life choices.

Thus, the diffusion of these systems is bringing a set of significant challenges at different levels of environmental systems of societies, from the microsystems to macrosystems of daily life.

There are several different aspects, including the transformation of workflows among different sectors. Risks for mental and social health of AI users should not be underestimated: There is increasing evidence of deskilling in work or educational tasks due to improper or unfair use of AI technologies; moreover, psychologically frail individuals are endangered by replacing consultation with health professionals with interaction with chatbots or developing affective relationships with them. ²

AI and LLM chatbot implementation in society requires new methodologies both to study human–AI interaction and to develop formative resources that could help different populations to understand what these tools actually are and what they are not, to drive appropriate adoption and use across various contexts.

Human–AI Collaboration: The Role of Narrative Technologies

Research in human–AI collaboration has highlighted how the integration of LLMs into human cognitive and social activity is not neutral. Chiriatti and colleagues ³ proposed the System 0 framework that positions AI systems as a technological extension of human cognition, capable of influencing both System 1 (fast, intuitive thinking) and System 2 (deliberative thinking, reasoning). Embedded in the everyday digital platforms that most users already interact with, this externalized cognitive layer shapes human mental processes in ways that can be both facilitative and detrimental, depending on the context of use.

A recent review ⁴ shows that, in the domain of creative generation tasks, AI systems appear to function as a form of cognitive scaffolding, supporting not only the generative process itself but also learning, provided that the collaboration is conceived through a strategic and participatory design framework. Addressing this challenge calls for a deliberate and ethically informed design of AI systems: one that strategically allocates tasks in ways that leverage the distinct strengths of both human and artificial agents, such as human creative originality and AI’s capacity for broad information retention while expanding dialectical cognitive engagement and strengthening human oversight and safety protocols.

According to our approach, a particularly promising avenue for advancing human–AI collaboration lies in the domain of narrative technologies. A growing body of research suggests that narrative structures are powerful elicitors of sophisticated cognitive processes, including perspective-taking, the construction of shared meaning, the experience of engagement, and the sense of presence that allows users to implement their own intentionality across mediated and natural contexts. ⁵ This is especially relevant in the case of story-based games, namely, environments defined not by calculation and rule optimization, but by meaning-making, character agency, and evolving narrative contexts. Such environments place distinctive cognitive demands on AI systems, for example, the ability to sustain consistent interaction over extended turns without generating misinformation; the capacity to integrate new information dynamically as it emerges during the interaction and to maintain a consistent and meaningful overall narrative frame; and last but not least, the sensitivity to subtle emotional cues and the ability to infer other agents’ intentions and perspectives, even when those are not explicitly shown or declared. In this sense, interactive narratives represent both a testing ground for evaluating the higher-order cognitive capacities of current AI systems and a promising design space for developing more capable, human-centered AI collaboration. It is possible that, when facing an interactive narrative along with an AI (e.g., as a coplayer), users would have the opportunity to gain a better understanding of both its functions and its limitations.

Narrative Games as a Framework for Human–AI Collaboration

There is a long tradition of using games to benchmark AI performance, from chess, to checkers, to go; however, these traditional games are defined by strict rules and deterministic outcomes, rewarding computational power over contextual understanding. Narrative-based games present different challenges. Role-playing games such as Dungeons & Dragons, for example, have recently been explored as platforms for testing LLMs because they are characterized by rich rule systems, branching choices, feedback cycles, and the continuous construction of shared meaning. ⁶ Engaging in such games requires AI systems to deal with the intentions of other players in ways that are contextually appropriate and socially meaningful, demands that go well beyond pattern recognition or rule optimization. For example, a type of narrative-based game is represented by gamebooks, namely, interactive narratives where the story proceeds according to the reader’s choices and actions. ⁷ In this sense, gamebooks offer a structured yet flexible environment in which the interplay between human agency and narrative constraints can be examined.

Beyond their role as evaluative benchmarks for AI, narrative and creative tasks have also emerged as particularly effective contexts for human–AI collaboration. Creative story generation and coauthorship with AI systems, whether producing an interview with a historical figure, drafting a report, or collaboratively building a fictional world, have been identified by international bodies including the Organisation for Economic Co-operation and Development and The United Nations Educational, Scientific and Cultural Organization as high-value activities for developing AI literacy, precisely because they engage users in active, hands-on interaction with AI tools rather than passive consumption. In addition, a growing body of applied research in psychology and education points to games and video games as meaningful tools for studying and supporting cognitive development and learning in human–AI collaboration. Taken together, these converging lines of evidence point toward the need for a multidisciplinary approach to human–AI collaboration, one capable of integrating cognitive, educational, and design perspectives to build AI systems that are meaningfully human-centered.

AIStories Project

The LAHTI (Laboratory for Advanced Human-Technology Interaction) at the Milan headquarters of Pegaso University, Italy, is engaged in a research project focused on exploring narratives and games as environments for studying human–AI collaboration and enhancing AI literacy. Building on the theoretical premises outlined above, the AIStories project aims to investigate the psychological dimensions of human–AI collaboration through compelling narrative and game-based contexts. Adopting a multidimensional perspective, the project pursues two core objectives: (1) identifying the individual psychological factors that shape how users engage with and represent AI systems and (2) designing and conducting experimental studies in which human users collaborate with AI in the creation and/or fruition of interactive narratives in order to assess how AI literacy, user engagement, and usage behavior are affected by the experience.

To address the first objective, a cross-sectional study was conducted to examine the relationships among motivation to use generative AI technologies and self-perceived AI literacy along the dimensions of AI awareness and AI engagement. ⁸ The study revealed that the engagement dimension of AI knowledge, defined as the ability to actively interact with generative AI tools and linked to practical, technical skills, fosters more autonomous forms of motivation for AI use compared to the awareness dimension, which reflects theoretical knowledge alone. These findings suggest that formative and educational efforts should prioritize hands-on, experiential engagement with AI tools over purely informational or theoretical instruction. To explore the role of narrative contexts in human–AI interaction, a case study was conducted in which an LLM chatbot was asked to assume the role of reader and player with a commercial gamebook. ⁷ This experience yielded some insights into the design of collaborative AI systems. First, the chatbot demonstrated the ability to accept the fictional categories of interactive narrative, but it progressively lost track of previously established information, with measurable consequences for its decision-making; second, the system was quick in abandoning the player role when it stopped to distinguish reality from fiction. More importantly, the case study highlighted how involving AI in interactive narratives makes its limitations particularly evident in terms of reasoning, problem-solving, and decision-making, especially when those are situated in complex contexts with characters, aims, and the necessity for flexible situation awareness.

Building on this, the project now turns to an experimental phase to examine how the interactive dynamics between human users and LLMs shift across the distinct phases of a creative process. Specifically, participants will collaborate with an AI system to construct a short narrative from a set of visual stimuli. The task is structured to distinguish between two creative phases to explore how perceived agency over the creative process varies depending on the type of subtask performed in collaboration with AI. In addition, the study will examine cognitive load, participants’ perception of the AI as a collaborator, and whether prior knowledge of AI systems moderates the quality and character of human–AI collaboration in a creative task. The findings are expected to add more information about how the human–AI collaboration process works in different parts of creative generation tasks.

Luca Botturi, PhD, full professor of media in education at University of Applied Sciences and Arts of Southern Switzerland (SUPSI), has shared about the project: “AI systems do not think or create as we humans do, and the adjective intelligent in ‘artificial intelligence’ is used as analogy, nor as a descriptor. Nonetheless, in practical terms AI can be regarded as a new type of (autonomous) agents—and it is paramount to gain insights in how our cognitive, creative and meaning-making processes adapt and integrate such agents. Interactive narratives provide an ideal playground for this research challenge: at the same time, they allow LLMs to exploit their key features (text generation) in a deeply human and creative domain, blending art, playing and experiential meaning. I expect this kind of studies to deliver highly relevant results to hopefully inform both AI education and AI systems design.”

Conclusions

The AIStories project aims to deepen our understanding of the psychological processes that underpin human–AI collaboration, with particular attention to how the strategic division of creative processes, explored through narrative frameworks and game-based environments, can strengthen the collaborative synergy between human users and AI systems.

In doing so, it moves beyond descriptive accounts of user attitudes toward AI, examining instead the cognitive, motivational, and interactional factors that shape how people engage with LLMs in open-ended, meaning-rich contexts. The findings are expected to carry implications across two interconnected dimensions. First, they speak to the design of AI systems as collaborative agents. The evidence emerging from this project suggests that effective human–AI collaboration should be designed taking into account the complexity of situations where users may employ AI tools to take decisions, sometimes impacting aims and objectives in real-life scenarios.

Second, the project offers important insights into the design of formative resources that would support individuals in forming more accurate and realistic representations of AI tools, hopefully leading to improved awareness of risks and appropriate use. Engaging with AI within the testing ground of interactive narratives (both cocreating and coplaying them) allows users to develop understanding of LLM limitations in dealing with contextual information, perspective-taking, meaning attribution, and intention recognition. Formative resources for AI implementation should privilege hands-on experiences that test chatbot performance with decision-making within undetermined contexts.