This article has been published at the 2023 FDG workshop on Human-AI Interaction Through Play, and lightly adapted for online viewing. The original publication can be found here and its bibtex entry here.
Computers have been leveraged to assist human creators throughout the history of computing. Recent advances in deep learning have allowed Artificial Intelligence (AI) to produce artefacts of high quality in creative domains such as language , visual art , and music . Most of these models are now either open-sourced or available online for everyone to experiment with. This has spurred the creativity of millions of users of all types  who strive to interact with the models, use their outputs, test their limits, and, of particular importance here, engage with them playfully .
We distinguish two types of Human-AI authoring tools: creativity support tools (CST) are designed for the system to support human initiative, while mixed-initiative co-creativity (MICC) tools endow both the human user and the AI with creative responsibility [19, 41]. We argue that neither CSTs nor MICC tools are always used as intended by their end-users -- and that it is a good thing. Specifically, these tools are often designed for improving efficiency in task-achievement, e.g. through faster turnaround times, or by offloading arguably tedious tasks to the AI; however, we often observe users spending more time interacting with the tool when they engage with the AI initiative than when they do not . Recent studies on text-to-image generators such as Midjourney  highlight that even professional creatives use these tools especially for inspiration [21, 39], a goal that arguably runs against most common conceptualisations of task-achievement. These model-based generators and their interfaces seem designed for professional, efficient interactions: e.g. via a single user input and static produced outputs. However, we often observe that users tend to interact with CSTs and MICC tools in a task-agnostic, open and subversive fashion, i.e., a playful attitude -- also confirmed through empirical study .
In this paper, we envision and motivate playfulness as an important design aspect of CSTs and MICC tools to unlock their true potential. As discussed earlier, end-users already interact playfully with most of these tools. Crucially though, this is despite the task-focused design of the tool itself, rather than supported by it. This interaction is similar to the Montessori child-centred method of education , insofar as the available models, tools and software (like the toys within reach in the Montessori method) are easily available to all to interact with in a hands-on fashion as the users' will. We project that designing interaction paradigms and AI methods that not only allow for playfulness, but invite and foster it, can support the satisfaction of more diverse user goals, and have a strong, positive effect on the user experience, the emerging human-AI interaction, and consequentially the resulting artefact.
In this paper we distinguish different notions of playfulness, discuss its relation to creativity, identify its requirements, and discuss how it could be actively supported in CSTs and MICC tools.
Csikszentmihalyi remarks that playfulness is "not an expendable luxury. It is the stuff of life, it is what gives us the experience of freedom, of transcendence, of growth" . Playfulness has been conceptualised as an experience [e.g. 15, 27], a trait [e.g. 35, 36], and, most importantly in this context, as a state or attitude [e.g. 14].
Högberg et al. define playfulness as a dimension of gameful experience: the "experience of being involved in voluntary and pleasurable behaviors that are driven by imagination or exploration while being free from or being under spontaneously created rules" . Building on previous theoretical work on pleasurable game experiences, emotions and play, Lucero et al.  define 22 categories of playful user experience in their Playful Experiences framework (PLEX). A subset, e.g. experiences of captivation, control, discovery, exploration and subversion, overlaps with Högberg et al.'s findings. In contrast, Proyer et al.  conceive playfulness as a trait; an "individual differences variable that allows people to frame or reframe everyday situations in a way such that they experience them as entertaining, and/or intellectually stimulating, and/or personally interesting”. Proyer  proposes a model with four distinct facets of adult playfulness; the most relevant facets here are: lighthearted, i.e. "seeing life as a game and not worrying too much about the future consequences of one's own behavior; liking to improvise" , intellectual, i.e. "liking playing with ideas and thoughts; liking thinking about and solving problems"  and whimsical, i.e. "finding amusement in grotesque and strange situations" .
Playfulness as a state or attitude can be best understood in contrast to play. Heljakka notes that, "whereas play is a process that may develop the capacities of the player (...), playfulness remains a way of thinking that does not refer to [the] manipulation of materials" [14, p. 212]. Crucially, "playfulness (...) may not equal play" [14, p. 214]. In design, for instance, playfulness "may be regarded as an attitude or a state of mind that helps us to see things differently or achieve unexpected results" [14, p. 211]. Similarly, Pichlmair highlights playfulness as an "accompanying attitude, a mental state of openness towards a situation" . Adopting the definition of play as "the free space of movement within a more rigid structure" , Pichlmair notes that playfulness only emerges within structures that limit freedom: "Playfulness is an attitude manifesting in the experience of approaching these limits, of exploring them" .
While a user's playfulness as a trait is out of our control, we argue for designing for playfulness as both a state and an experience.
Since we aim to support playfulness in CSTs or MICC systems, we must distinguish playfulness from creativity. Bateson and Nettle  found empirically that people who consider themselves playful also believe that they come up with new ideas. They conclude that "there is a conceptual overlap, but it is probably best seeing play and playfulness as a facilitator of creativity without them being redundant”. Proyer et al.  similarly support a robust association between the concepts but note that "these relations do not indicate redundancy”, and that "there is good evidence for arguing that playfulness is a facilitator of creativity and creative responses" . While the latter consider playfulness a trait, Bateson and Nettle hold that playfulness as a state/attitude can be encouraged "and, in so doing, creativity and hence innovation can be enhanced" . These findings support that playfulness is distinct from creativity, and both concepts should thus be considered separately in the design of CSTs or MICC systems. Moreover, they highlight that designing for playfulness as a state not only has the potential for fostering the user's desired mode of interaction, but could also strengthen creativity as a potential goal. The latter overlaps with the use of AI systems for inspiration as highlighted in Section 1.
Based on the previous findings, we argue that designing for playfulness in the interaction with CSTs or MICC systems is a worthy and natural goal, with first support coming from research on casual creators . While playfulness may serendipitously occur in all types of creative process, we expect that a consequence-free scenario would allow playfulness to thrive. Therefore, we envision playful CSTs and MICC tools to be intended for novice and amateur users wishing to explore the technology and/or the problem space, as well as for professional creatives (such as artists or game developers) during conceptualisation stages, far from deadlines, or freed from a task-oriented mindset. Moreover, playful co-creative tools can elucidate the affordances and constraints of current AI algorithms by providing low-effort challenges that motivate users to engage with specific aspects of an AI, such as playfully exploring prompt engineering in text-to-image generators .
Identifying where playfulness lies in such tools forms interesting directions for future empirical and theoretical exploration. We consider means to support playfulness either in the human (Sec. 4.1) through UI design, in the AI (Sec. 4.2) through algorithms, or through interventions to their dialog (Sec. 4.3).
Figure 1: Playful interaction paradigms instantiated in Quick, Draw! The in-game screenshot shows the game-provided prompt and timer (top), the in-progress user drawing and the AI vocalized guesses (bottom).
First, we see the opportunity for human-AI authoring tools to be explicitly designed to encourage user playfulness as a state and experience. User interfaces (UI) in current MICC and CST tools are visually simple and unappealing, possessing only the absolutely necessary UI elements to complete the task. This is often due to the fact that such tools are developed by researchers with limited resources for refining the user experience design or too narrowly focused on task-oriented productivity. Perhaps the most playful interaction paradigm in AI-assisted interaction is in Quick, Draw! where the user draws a quick doodle while the AI vocalises the predicted object (see Fig. 1). The AI in this case could be considered de facto playful as it emulates voices of team members in the Pictionary game (Angel Games, 1985). Looking at more specific patterns that enhance playfulness is the constant in-progress feedback, the alternate modality (voice) that does not distract the user from their task, and the ad-hoc short time constraint for drawing (20 seconds) that enhances the challenge while signalling to the user not to take the task too seriously. It is worth noting that in Quick, Draw!, the AI acts more as a critic than a co-creator, as it does not produce its own drawings; however, its vocalised guesses may influence the player to refine their drawing and thus influences the creative process. While operating on similar sketches as Quick, Draw!, the experiment by Karimi et al.  (where both AI and human produce sketches in parallel) has a much more task-oriented and barebone interface; however, the goals of that project is to foster creativity rather than playfulness. Other AI-assisted sketching tools that focus on AI and human taking turns contributing to a common canvas [8, 16] embrace the exploratory, low-effort, and casual nature of sketching  but do not explicitly foster or reward it in terms of either the AI or the interface. While many MICC systems target improvisation, e.g. a user changing their work based on an AI's suggestion [8, 20, 24], the medium also impacts how playful a task is perceived: for instance, full-body interaction inspired by improvisational theatre  or dance  can foster playfulness merely due to the physical activity itself.
We moreover argue that playfulness as a state can be fostered by freeing the user from a task-oriented mindset and framing the activity appropriately as a task-agnostic, exploratory endeavour. This could be accomplished through an on-boarding activity for more complex or more task-specific human-AI interactions. For instance, the intricacies of prompt engineering for text-to-image generation could be demonstrated through a playful interface that pushes a user to add or remove some of the words of their prompt and visualise the result in opposition to the image produced by the original prompt. Beyond the framing of the tools' goals, design patters from games and gamification [15, 38] can structure the interaction flow, similar to the time constrained micro-tasks in Quick, Draw!. Points, badges, or leaderboards can be used as rewards and achievements to motivate especially novice and amateur users to learn the different possibilities of CSTs or MICC systems playfully. Specifically designed Easter Eggs can also reward and surprise users who explore the systems and find them by trying different features.
Second, we envision new ways to impart playfulness as a trait of the AI by integrating algorithms for playfulness which operate separately from the system's authoring functionalities, and steer its interactions both with the UI system and with the human user. As a starting point to this very challenging endeavour, we propose to formalise central aspects of a playful state (such as curiosity, an attraction to the unexpected, and a tendency to subvert) in the agent's policy, i.e. the basis of its behaviour. To this end, we advocate to embrace intrinsic motivation  such as artificial curiosity  and models of competence allowing for autonomous goal-discovery and skill acquisition [4, 5], as well as variants of novelty , surprise , and quality-diversity search  to support goal-free or goal-directed exploration. While we explicitly advocate AI that is playful in a different way than its user, thus having the ability to complement the latter, some common ground is likely needed. In particular, we expect pure exploration by the AI to be insufficient for inspiring and maintaining playfulness in its user. To support and defy the user's expectations through more directed behaviour, e.g. defined as quality in quality-diversity search, the AI must perform efficient inference on the goals of their interaction partner. To this end, approximate Bayesian methods for goal inference [18, 42] promise to be of value. Next to engineering for playfulness, we highlight the ability of a tool to explain  its AI's (shifting) process and to verbalise its playfulness as equally important.
Third, we propose means to encourage user playfulness as a state and experience by intervening in the dialog between the user and the system. Human-computer interaction within MICC tools is often likened to dialog , and playfulness can be injected in that layer through humour and misunderstanding. This type of playfulness hinges as much on the interface as on the AI itself, but here we focus on how the AI can interpret human input and provide feedback playfully. We envision that the openness often associated with playfulness [28, 33] is to be achieved by an AI that playfully (mis)interprets the human co-creator's work or intent. This can be approached in many ways, indicatively:
After perceiving the user's input, the format of the AI response can also be playful. Here, we borrow from the whimsical facet  and "lack of self-importance" (regarding both human and AI)  within playfulness and envision an AI that role-plays (e.g., through chat) an idiosychratic personality. Examples of such "personalities" could be a taunting AI that prods users to be more original, or one that expresses a specific historical era or style in terms of its aesthetics and its vocalization via natural language filters  and text-to-speech voices or profiles .
As noted in Section 2, playfulness has received extensive attention in terms of what it is and how it emerges. However, in light of current authoring tools built around human-AI interaction, the playfulness of the human user, the AI, or their interaction is overlooked -- treating such tools instead as professional, sombre, task-focused interfaces. Future research should identify how different types of users become playful with existing tools, explore methods for imparting playful properties to the AI itself or to its interaction with the user, and develop robust methods for evaluating playfulness via e.g. interaction data and questionnaires.
This research is the culmination of work carried out during the Schloss Dagstuhl Research Meeting 23033: Game Measures and Player Experience, which occurred during 15-18 of January, 2023. All authors participated in the working group on this topic during the Research Meeting, and contributed equally. The author order is based on the contribution to the writing process for this paper.
 Patrick Bateson and Daniel Nettle. 2014. Playfulness, ideas, and creativity: A survey. Creativity Research Journal 26, 2 (2014), 219–222.
 Alastair Brotchie and Mel Gooding. 1991. Surrealist games. Redstone Press.
 Bill Buxton. 2007. Sketching user experiences: Getting the design right and the right design. Morgan Kaufmann.
 Nuttapong Chentanez, Andrew Barto, and Satinder Singh. 2004. Intrinsically motivated reinforcement learning. Advances in neural information processing systems 17 (2004).
 Cédric Colas, Tristan Karch, Olivier Sigaud, and Pierre-Yves Oudeyer. 2022. Autotelic agents with intrinsically motivated goal-conditioned reinforcement learning: a short survey. Journal of Artificial Intelligence Research 74 (2022), 1159–1199.
 Katherine E Compton. 2019. Casual creators: Defining a genre of autotelic creativity support systems. Ph. D. Dissertation. University of California Santa Cruz.
 Mihaly Csikszentmihalyi. 1981. Some paradoxes in the definition of play. Play as context 1, 1 (1981), 14–26.
 Nicholas Davis, Chih-Pin Hsiao, Kunwar Yashraj Singh, Lisa Li, and Brian Magerko. 2016. Empirically studying participatory sense-making in abstract drawing with a co-creative cognitive agent. In Proc. Intl. Conf. on Intelligent User Interfaces.
 Edward De Bono. 2010. Lateral thinking: Creativity step by step. Harper Collins.
 Omar Delarosa, Hang Dong, Mindy Ruan, Ahmed Khalifa, and Julian Togelius. 2021. Mixed-initiative level design with RL brush. In Proc. EvoMusArt conf.
 Daniele Gravina, Ahmed Khalifa, Antonios Liapis, Julian Togelius, and Georgios N. Yannakakis. 2019. Procedural content generation through quality-diversity. In Proc. IEEE Conf. on Games.
 Daniele Gravina, Antonios Liapis, and Georgios Yannakakis. 2016. Surprise search: Beyond objectives and novelty. In Proc. GECCO. 677–684.
 Christian Guckelsberger. 2020. Intrinsic motivation in computational creativity applied to videogames. Ph. D. Dissertation. Queen Mary University of London.
 Katriina Heljakka. 2013. Principles of adult play (fulness) in contemporary toy cultures: From wow to flow to glow. Ph. D. Dissertation. Aalto University.
 Johan Högberg, Juho Hamari, and Erik Wästlund. 2019. Gameful experience questionnaire: An instrument for measuring the perceived gamefulness of system use. User Modeling and User-Adapted Interaction 29, 3 (2019), 619–660.
 Francisco Ibarrola, Oliver Bown, and Kazjon Grace. 2022. Towards co-creative drawing based on contrastive language-image models. In Proc. ICCC.
 Mikhail Jacob, Alexander Zook, and Brian Magerko. 2013. Viewpoints AI: Procedurally representing and reasoning about gestures. In Proc. DIGRA Conf.
 Antti Kangasrääsiö, Jussi PP Jokinen, Antti Oulasvirta, Andrew Howes, and Samuel Kaski. 2019. Parameter inference for computational cognitive models with Approximate Bayesian Computation. Cognitive science 43, 6 (2019), e12738.
 Anna Kantosalo and Hannu Toivonen. 2016. Modes for creative human-computer collaboration: Alternating and task-divided co-creativity. In Proc. ICCC. 77–84.
 Pegah Karimi, Jeba Rezwana, Safat Siddiqui, Mary Lou Maher, and Nasrin Dehbozorgi. 2020. Creative sketching partner: An analysis of human-AI co-creativity. In Proc. Intl. Conf. on Intelligent User Interfaces. 221–230.
 Hyung-Kwon Ko, Gwanmo Park, Hyeon Jeon, Jaemin Jo, Juho Kim, and Jinwook Seo. 2022. Large-scale text-to-image generation models for visual artists' creative works. arXiv preprint arXiv:2210.08477 (2022).
 Antonios Liapis, Héctor P Martınez, Julian Togelius, and Georgios N Yannakakis. 2013. Transforming exploratory creativity with DeLeNoX. In Proc. ICCC. 56.
 Antonios Liapis, Georgios N. Yannakakis, and Julian Togelius. 2013. Designer modeling for personalized game content creation tools. In Proc. AIIDE Workshop on Artificial Intelligence & Game Aesthetics.
 Antonios Liapis, Georgios N. Yannakakis, and Julian Togelius. 2013. Sentient sketchbook: Computer-aided game level authoring. In Proc. FDG Conf. 213–220.
 Grace I. Lin and Marilyn A. Walker. 2011. All the world's a stage: Learning character models from film. In Proc. AIIDE Conf.
 Rui Liu, Berrak Sisman, Guanglai Gao, and Haizhou Li. 2022. Controllable accented text-to-speech synthesis. arXiv:2209.10804 [cs.SD]
 Andrés Lucero, Jussi Holopainen, Elina Ollila, Riku Suomela, and Evangelos Karapanos. 2013. The playful experiences framework as a guide for expert evaluation. In Proc. Conf. Designing Pleasurable Products and Interfaces. 221–230.
 María Lugones. 1987. Playfulness, "world"-travelling, and loving perception. Hypatia 2 (1987), 3–19. Issue 2.
 Chloë Marshall. 2017. Montessori education: A review of the evidence base. npj Science of Learning 2 (2017).
 TJ McCue. 2022. Midjourney AI based art generator creates dazzling images from words. https://www.forbes.com/sites/tjmccue/2023/01/23/midjourney-ai-basedart-generator-creates-dazzling-images-from-words/.
 David G. Novick and Stephen Sutton. 1997. What is Mixed-Initiative Interaction?. In Proc. AAAI Spring Symposium on Computational Models for Mixed Initiative Interaction.
 Open AI. 2022. ChatGPT: Optimizing language models for dialogue. https: //openai.com/blog/chatgpt/.
 Martin Pichlmair. 2008. Venturing into the borderlands of playfulness. Technoetic Arts 6, 2 (2008), 207–212.
 Andreya Piplica, Christopher DeLeon, and Brian Magerko. 2022. Full-body gesture interaction with improvisational narrative agents. In Proc. Intl. Conf. on Intelligent Virtual Agents.
 René T Proyer. 2017. A new structural model for the study of adult playfulness: Assessment and exploration of an understudied individual differences variable. Personality and Individual Differences 108 (2017), 113–122.
 René T Proyer, Nancy Tandler, and Kay Brauer. 2019. Playfulness and creativity: A selective review. In Creativity and humor, Sarah R. Luria, John Baer, and James C. Kaufman (Eds.). Elsevier, 43–60.
 Jürgen Schmidhuber. 1991. Curious model-building control systems. In Proc. IEEE Intl. Joint Conf. on Neural Networks.
 Jakub Swacha and Karolina Muszyńska. 2016. Design patterns for gamification of work. In Proc. Intl. Conf. on Technological Ecosystems for Enhancing Multiculturality. 763–769.
 Veera Vimpari, Annakaisa Kultima, Perttu Hämäläinen, and Christian Guckelsberger. 2023. "An adapt-or-die type of situation": Perception, adoption, and use of text-to-image-generation AI by game industry professionals. arXiv preprint arXiv:2302.12601 (2023).
 Kyle Wiggers and Amanda Silberling. 2022. AI music generators could be a boon for artists -- but also problematic. https://techcrunch.com/2022/10/07/ai-musicgenerator-dance-diffusion/.
 Georgios N. Yannakakis, Antonios Liapis, and Constantine Alexopoulos. 2014. Mixed-initiative co-creativity. In Proc. FDG Conf.
 Tan Zhi-Xuan, Jordyn Mann, Tom Silver, Josh Tenenbaum, and Vikash Mansinghka. 2020. Online bayesian goal inference for boundedly rational planning agents. Advances in neural information processing systems 33 (2020), 19238–19250.
 Jichen Zhu, Antonios Liapis, Sebastian Risi, Rafael Bidarra, and G. Michael Youngblood. 2018. Explainable AI for designers: A human-centered perspective on mixed-initiative co-creation. In Proc. IEEE Conf. on CIG.
 Eric Zimmerman. 2004. Narrative, interactivity, play, and games: Four naughty concepts in need of discipline. In First person: New media as story, performance, and game. Vol. 154. MIT Press Cambridge, MA.