Human-AI System Co-creativity to Build Interactive Digital Narratives

Collaboration among people significantly affects creativity; from a Vygotskian perspective, working in a group can increase one’s repertoire of cognitive and emotional expression and chances of coming up with original ideas [7]. Creativity is more likely to be achieved if the collaborators have complementary skills [5]. The mechanisms developed during collaboration can significantly influence the final product’s creativity [10]. However, asynchronous forms of contact are occasionally used, and participants may feel hesitant to voice their opinions or as though their ideas are less valuable than others; AI-supported collaboration can aid in breaking through some creative barriers [4]. Computational creativity in the form of a collaborative system can be linked to the idea of creativity, but a human component made up of social, ethnographic, and personal knowledge is still required. Collaboration between a human and an AI system can lead to co-creation when the system and the human act on each other's responses in the pursuit of creativity [25].

Co-creation is the founding argument of this contribution, which inserts in the context of Interactive Digital Narratives (IDNs), considered not only an area of investigation but also a user experience and an artefact resulting from a creative process between humans and AIs. Designers who approach AI support systems to use them for creative purposes need to understand their functionality, characteristics, and potential or access tools that allow them to classify these AIs proliferating chaotically. The IDN context is highly interdisciplinary and complex. Despite having been involved in producing IDNs for about 30 years [35, p. 525], research in this area needs to be revised. Since AI has taken hold in recent years with an increasingly efficient computing power breakthrough thanks to new neural network systems capable of working with multiple algorithms simultaneously on multiple levels, the amount of AI support tools has increased. However, a taxonomy is missing, allowing an order between these emerging and spreading tools. So, the designer looks for examples of AI systems that creatively support them in designing IDNs experiences, trying to earn time and exploit their computing power. This study aims to clarify the research area of AI support systems to discover the essential components to creatively design IDNs.

There is now a dearth of understanding and awareness among the general public regarding AI, which has led us to confusion and worries. Imaginaries connected to the philosophical current of existential risks significantly influence the current vision of AI systems [2]. In this context, “artificial intelligence” refers to a machine with algorithms capable of autonomously learning from a database and trying to mimic human intelligence. Even humans are constantly learning how to maximise AI’s potential while lowering its risks, as it is still in its early stages of growth and development. The AI system is equipped with a powerful computational engine that can handle and analyse large amounts of data simultaneously, something humans would not be able to achieve. The ability to calculate quickly and accurately is one main feature. While it can recognise emotions, it lacks empathy and cannot experience feelings, which is one of its significant flaws. However, humans can overcome this problem by collaborating with AI systems. The key is working together so that humans and AIs can complement one another. Knowing what algorithms are, how they operate, and how to apply them before we can trust them is a prerequisite for a successful collaboration. Having an AI system that collaborates with a human can significantly reduce the workload that the human can take charge of, and at the same time, the system can benefit from specific inputs. However, it depends not on humans’ knowledge of the system they collaborate with but on its comprehensibility. Even having the correct knowledge about algorithms and AIs is not enough since the knowledge needs to be presented precisely and comprehensively [33]. This work is a proposal of how to clarify the creative support tools for the creation of IDNs.

The paper explores and compares recent articles in the field of AI support systems, re-storing an excursus of the panorama regarding the human-AI system co-creation considering the growth of AI support systems that helps the designer build IDNs. AI systems can enable and assist authors in developing new content [38] and may be claimed to be support tools for realising stories in all their forms. In this case, authoring tools are utilised to create IDNs, a more complex narration form due to the interaction component. The medium or affordances determine the interaction, which leads to programmed actions [11, 29]. In this instance, the AI system is the media seen as a means of knowledge; it organises knowledge to share with the user and helps them create techniques for problem-solving [37]. IDNs in AI systems are thought to boost creativity by arranging the data in the AI system to draw unanticipated new conclusions from previous ones [9].

IDN(s) are a family of narrative concepts from humanistic viewpoints that may be played with, interacted with, and intervened in using digital technologies [28]. The IDN field deals with interactive storytelling beyond a printed book’s bounds. It also raises concerns about the boundaries of what qualifies a story [34]. By disrupting the traditional roles of the author and the reader, IDN shuffles the deck of cards through the interaction element. Interaction can be a participatory process in which an interactor engages with a computer program to produce an output [18].

The field needs to be systematised through developing, adopting, and advancing guidelines and taxonomies since there is an insufficient and occasionally muddled body of knowledge [8, 21]. Koenitz summarises the IDN-related concerns in the paper “Five Theses for Interactive Digital Narrative” [19], pointing out the confused body of knowledge, the variety of existing writing tools with distinct research directions, and the focus on the creation of IDN experiences rather than the method.

The Authoring Tool category is examined here as an AI authoring system that uses the AI system to assist with and manage the difficulty of developing interactive narratives. AI authoring systems have an author as a partner, but in this contribution, the author is a designer considering more aspects of collaborating and creating these systems. The designer of IDNs masters and applies design thinking and is aware of the creative process of storytelling and, in the specific case of AI systems, the machine learning process. Here, the designer is a narrative interaction designer [36], with the IDN creator designer as the target of AI authoring system creation–the one who designs and builds AI authoring systems. The designer is initially identified by Murray [30] as the cyberbard that designs the IDN experience by giving authorial control to the interactors with the system, those who create the narrative [20].

The cyberbard, who uses an AI authoring system to construct IDNs, is referred here to as the designer as a creator who has not the knowledge of a writer but that of a problem-solver.

The approach used here to face complexity is what Krogh calls “drifting by intention” [22], a knowledge built on continuous learning from ongoing findings. In the discipline of design, drifting is perceived positively. It exemplifies how design researchers gain information about their findings and reshapes them. The investigation is based on a literature review of twenty academic publications that contain papers, journals, articles and books. The reviewed content includes critical reviews, frameworks, prototypes, and case studies. The search was done in Google Scholar, Scopus and ACM Digital Library databases, looking for AI support systems that help the designer creatively build IDNs. From the review emerge some characteristics and recurring elements that are collected under the categories: the Type of support tool, the Narrative elements, and the Type of creativity. The intention is to clarify the body of knowledge by systematising it to communicate it more simplified to the designer who intends to work alongside AI systems to create IDNs. The search has been restricted to the last twenty years, although most articles are from the last decade. The literature review includes Creative Support Tools (CSTs) consisting of narrative elements and those without such elements. The decision to include CST without narrative elements is essential for analysing the Type of creativity category, which otherwise would not be analysed exhaustively, i.e. considering all sub-categories. The goal is to identify the basic elements and sub-elements that make up an AI support system for designing IDNs creatively to systematise knowledge of the field.

3.1 Findings on Creativity

In the study for AI tools to support the creation of IDNs, a gap emerged between AI tools that support creativity, designed to target creativity enhancement and those designed to author stories. Eleven of twenty academic publications are CSTs, and nine are Authoring Tools. Interest in support systems for creativity is often unrelated to the goal of creating a digital narrative interaction. The analysed academic contributions address different forms of creativity, trying to make terminological clarity between personal creativity (P-creativity), historical creativity (H-creativity), human-AI co-creativity, computational creativity (CC), collaboration for creativity and crowdsourcing creativity.

Creativity can be considered the product of a human mind, as a simulation of an artificial mind, or as the result of the collaboration between humans and AI systems.

P-creativity, for example, is associated with personal or psychological creativity, which refers to new information, concepts, and ideas that an individual was not previously aware of and brings novelty to how he/she perceives the world. P-creativity is a frequent form of creativity since it involves a single person’s expertise. H-creativity is also linked to the single human mind but is related to findings that have never been published in the history of humanity [1, p.76].

The CSTs are designed with the purpose of increasing creativity, and these support tools are, for the most part, disconnected from the concept of IDN. Creative support is often linked to the AI automation of some specific task. For instance, FahionQ [14] suggests clustered styles based on quantifiable fashion traits to enhance convergent and divergent thinking. Therefore, CC is data automation and pattern detection communicated to the end user. CC can be seen as the computational version of human creativity, which attempts to simulate creativity. In the case of a human-AI collaboration, creativity may take on different shades; that is, a qualitative interpretation of creativity comes into play rather than an objective one. Creativity is linked to a more humanistic and qualitative side, more difficult to identify in clusters, especially when the co-creative aspect that arises from the relationship between the designer and AI becomes relevant. Co-creativity is a step towards an ongoing conversation between the designer and the system, resulting in a P-creative idea. In Paper Dreams [3], the AI system suggests illustrations to the designer based on the designer’s sketches. This example lays the foundations for creating a narrative through the construction of the visual storyboard; however, there is little creativity on the AI system side that suggests similar images based on the sketches of the user designer. The system offers images based on user changes, thus lacking an overall and more elaborate design vision. The term co-creativity is often misused, being exchanged with CC that is limited to fulfilling tasks and can have a non-creative output (Fig. 1).

Table of literature review. Source: author

Full size image

3.2 Findings on IDNs

The literature review has brought out the few AI systems that are concerned with creatively creating IDN. Furthermore (Fig. 2), the AI support systems for creating IDNs are identified in the macro group of authoring tools, which deals with automation and story generation, that reduces to the bone the potential of the interactive narrative aspect. Automation of story construction becomes a relationship between the designer's input and the response output of the system.

Hierarchical diagram of the categories of tools analysed. Source: author.

Full size image

The autonomous creation of the stories takes place by feeding an AI database from which an output is produced, limiting the interaction with the system to settings set by the narrative interaction designer. A totally autonomous system does not fall into the Authoring Tool category and cannot be called such. The authoring tools automate part of the story creation process, giving the designer space to contribute to the production of stories. For example, in Shelley [41], an AI system that writes and publishes horror stories on Twitter, users are actively and continuously involved in the writing of the story that takes place through an ongoing dialogue with the AI that analyses the users’ sentences and based on the received proposals elaborates a follow-up.

The category of narrative elements contains numerous components, some with distinct and others with interchangeable meanings. For example, the narrative world (NW), storyworld and protostory can take on the same meaning, referring to an imaginary world populated by characters inserted within an environment with objects available, a world from which actions or events can arise. Among these narrative components, a lack of hierarchical reflection emerges at the level of taxonomic subdivision by order of importance since they tend to be considered on the same level. In reality, narrative elements underlie the others, for their generative component, such as the NW, is composed of characters who perform actions that gradually become stories. Therefore, this literature review intends to report the deficiencies, not just the findings.

3.3 Discussion on Findings

Therefore, the thesis that the related body of knowledge about AI support systems is still confusing has been confirmed, and the proliferation of these systems only increases the chaos leading to a waste of energy for those who try to understand the scope of the systems of AI support concerning the creation of IDNs. However, a scientific contribution tries to lay the foundations and propose a framework that allows for the classification of the support authoring tools, ordering them into sub-categories. This first attempt to identify and analyse authoring tools is presented through the work of Shibolet et al. [35], in which some categories and descriptors are introduced. These categories are described in the appendix document “Tool Shortlist” [35] and are procedural generation tools, mixed-initiative authoring tools, and procedural AI authoring tools. These sub-categories are the ones that get closer to AI systems that work alongside the design and support it in generating IDNs. However, two main categories of AI support systems emerged from the literature review: AI-based CSTs and AI authoring systems.

The AI-based CSTs are primarily related to computational creativity, being generative AIs that mainly support humans in executing tasks rather than collaborating on a specific project with an overall vision. These AI systems generally involve users in generative music [13], drawings [17], sketches [16, 3], and image generation [14].

From an IDN perspective, the two AI support systems categories converge and form a new class: AI systems to support creativity for creating IDN, also called AI creative support systems for IDNs. This new hybrid category represents a new emerging typology of AI creative support systems that involve the creation of IDNs thanks to a creative collaboration with the designer. To date, this category of AI support systems remains unexplored or consciously little explored. AI co-creative support systems for IDNs need to be acknowledged in the literature as a new category of tools and is a niche since most of the support systems do not adequately support the designer in a continuous dialogue toward finding a creative solution to the starting problem.

Human-AI co-creativity is a relationship between two entities that complement each other resulting in a creative output that, in this contribution, identifies with the creation of an IDN. This study highlights the type of support tools and their main elements starting from the context of IDN and extending it to that of the CTS. These categories emerged from the literature review of twenty academic articles, which show the hierarchical relationships among support tools versus AI support systems and creative systems versus authoring systems. Furthermore, a new type of hybrid AI system is emerging from the union of AI-based CST and AI authoring systems categories, which is defined here for the first time as an AI creative support system for IDNs. This new type of AI support system must be acknowledged as a new category corresponding to an existing small niche in the literature.

The review identifies not only the categories of AI support systems but also the elements from which these support systems are composed. Indeed, it emerges as AI-based CSTs are mostly related to CC rather than human-AI co-creativity. Most AI support systems are built to execute tasks, but AI creative support systems for IDNs can reason on a given problem and suggest creative solutions. For a detailed overview of the findings and future developments, see Fig. 3.

Table of outcomes and further steps. Source: author.

Full size image

The types of creativity components and the narrative elements currently lack a framework structure that considers their hierarchical role of importance in constructing an IDN. Therefore, it is in the subsequent developments of this research to create a framework that systematises the essential elements of the AI support system for IDN and validates them through interviews.

Co-creativity is not easy to define concerning other types of creativity; therefore, reaching a terminological consensus in the literature would facilitate its clustering. This research is the basis for future interactive narrative designers to fill the gaps and terminological pre-understandings by proposing taxonomies that consider the interdisciplinary aspect of the IDNs.