Exploring Unstable AIs for Creative Expansion Laura Devendorf Abstract ATLAS Institute & This workshop proposal explores a concept for Information Science engaging AI in a non-deterministic manner in order to University of Colorado, Boulder collaboratively produce physical artworks with a digital laura.devendorf@colorado.edu system. It draws from a lineage of “games” played by artists, from Surrealists to Situationists, that were targeted towards automatism and creative exploration as opposed to the expression of a preconceived idea. Author Keywords Creative practice, non-determinism, AI, human- computer collaboration ACM Classification Keywords H.5.m. Information interfaces and presentation (e.g., HCI): Miscellaneous; Introduction Descriptions of creative practices offered by designers, artists, and crafts people alike often highlight the non- linear and non-deterministic elements of a creative practice. For instance, ideas often come from chance encounters in the everyday world or stem from accidents. In order to expand their space of ideas and sensitivities, many artists turn away from a pre-existing Copyright © 2017 for this paper is held by the author(s). goal or vision and instead, look for ways to work with Proceedings of MICI 2017: CHI Workshop on Mixed-Initiative Creative his or her materials (be they digital or physical) in Interfaces. open-ended, exploratory ways that may give rise to unexpected or serendipitous outcomes or “happy practices. On a more specific level, it seeks a mode of accidents.” In such practices, technology does not engagement with AI that can give rise to surprising and necessarily need to be enrolled as an assistant or beautiful results. instrument of productivity. Instead, I have argued that digital systems for creative practice can be thought of Unstable & Collective AIs as “translations” [5] —tools that allows creative At the workshop, I would like to present a concept for a practitioners to experience their idea though a new future MICI or unstable tool that is informed by symbolic and technical frame, for instance, exploring “games” historically played by artists called AI Reverb. how a 3D printing can map 3D models to sonic profiles, or how a machine learning algorithm might see an Artists Games image [6]. In the role of translator, a computational Artists games are unstable by design and typically bring system needn’t only “assist” the creative practitioner or multiple creative actors together to produce objects advance them closer to some predetermined goal state, unique to the situation of production, a form of extreme but can actively resist the maker, push back, or “break” collaboration where “players” correspond in an open their ideas in ways that may reveal new creative field of creative possibility. For instance, Surrealist potentials. As such, engaging the concept of translation artists created several games for generating artwork in design leads to tools that are unstable in the sense automatically and in a stream of consciousness fashion, that they preserve risk and unpredictability. These “Solitary and collective automatic techniques, and the unstable tools are to be used for inspiration, providing exploitation of chance are central to many surrealist the maker with a new way of seeing or understanding games…automatic techniques may be used as a their particular object of inquiry that they can take up beginning of a creative activity, to stimulate and and fold into their practice in whatever way they encourage spontaneity of utterance or image-making” choose [3,4]. [1]. The most famous of these games is “exquisite corpse,” a procedural game in which one artists draw a While I have studied AI and computer science formally, head, hides what he or she draws, and invites another I have yet to engage advanced algorithmic techniques to fill in the remaining body and legs. The result is an in the unstable prototypes that I have created to date. I outcome that neither artist could plan or anticipate. It used to see these advanced algorithms as mechanisms is an artistic product born from a collective, creative that reduced engagement in the physical world, or intelligence. eliminated the “risk” that I and others find so valuable in a creative practice. I would like to attend this Led by the writings of Guy Debord, The Situationist workshop to explore alternative engagements of AI and International extended surrealist games into the ream machine learning in what I have be calling “unstable of everyday life, developing tactics for engaging the tools,” thus, unstable AI’s. At a broad level, this project everyday that could denature the habitual and lead to joins wider calls for exploring where technology can experiences to allow someone to see beyond spectacle participate within non-linear and chance-based creative [2]. As such, these games turned away from a concrete artistic product and into a mode of sensitizing the understanding as opposed to a particular creative player. Furthermore, they fused games with life, “object” outcome. suggesting ludic engagements in everyday space. Such themes resonate through related art movements, like AI Reverb: Fusing AI and Artists’ Games the chance inspired “event scores” of Fluxus artists Drawing from this lineage of artists’ games or event Yoko Ono [7] and La Monte Young [8] which prompt scores, I imagine a system called AI Reverb that aesthetic engagements in the more mundane prompts an artist to take action in response to happenings of the everyday. For instance, one of Ono’s directives supplied by an artificial intelligence agent. In event scores, entitled Tunafish Sandwich Piece, keeping with my interest in engaging with the everyday requires its viewers/actors to: and materials, I imagine a system composed of a camera and text output screen. The camera captures Imagine one thousand suns in the the present scene, say, the maker sketching on a piece sky at the same time. of paper. That scene (i.e. the paper) is processed as Let them shine for one hour. input to an AI that classifies what the object that the Then, let them gradually melt into maker is drawing (much like Google’s Quick Draw the sky. application 1.) Based on how the AI identifies the Make one tunafish sandwich and eat. [7] object (for instance, if it thinks the artist is drawing a cat), it can supply instructions for the maker to perform Ono’s event score fuses poetry with practice, creating a on the drawing (e.g. draw multiple tails on cat, throw prompt for a sensory engagement in sun and away drawing of cat and draw a dog instead, etc.). As sandwiches alike. The goal of the work is less oriented the artist performs the command, the drawing changes, around a “thing” produced, and more focused on how the classification of the drawing changes and a new set the execution of the instructions shaped the person of commands is born ad infinitum. The outcome of such who executed them. a system represents a reverberation between an artist and a machine, a surrealist drawing born out of a While each example is targeted towards different collective human-machine intelligence. outcomes and audiences, whether it be an automatically generating thing, a critique of spectacle, In AI Reverb accuracy and the correctness no longer or an attempt at anti-art, they share in common a function as meaningful bounds for the design space of vision of making where control extends beyond an interaction. Like a game of telephone, the pleasure individual maker or audience. They position the artist emerges from the moments in which the AI makes an as one of many numbers of forces capable of producing incorrect prediction or works in an imperfect manner. creative work, not necessarily the individual who stands And while I have illustrated the concept within the above controlling what is produced. As such, these relatively simplistic realm of drawing, one could games tend to be oriented towards sensitization and 1 https://quickdraw.withgoogle.com/ Figure 1: A concept sketch for AI embraced everyday life. For instance, an AI agent that Reverb demonstrating feedback feeds back on signage in shop windows and billboards between human doing sketching, in order to direct artists to take different actions within AI interpreting sketches in public space. progress and offering instructions for human to interpret. The tension with a project AI Reverb this is differentiating between a system that “breaks” well and a system that is just plain broken. At the present concept stage, I cannot predict exactly where those lines will be drawn. I see the sketch provided here as a benchmark for a new way of thinking about AI more than a set of plans for a system that I plan to enact. One of my goals for the workshop, then, is to refine this direction, learn more about the tools and techniques available, and hopefully gather feedback from artists in the group about how such a system might adapt into exploratory phases of their own practice. References 1. Alastair Brotchie and Mel Gooding. 1995. A Book of Surrealist Games. Shambhala, Boston. 2. Guy Debord. 2000. Society Of The Spectacle. Black & Red, Detroit. 3. Laura Devendorf. 2016. Strange and Unstable Fabrication. 4. Laura Devendorf, Abigail De Kosnik, Kate Mattingly, and Kimiko Ryokai. 2016. Probing the Potential of Post-Anthropocentric 3D Printing. In Proceedings of the 2016 ACM Conference on Designing Interactive Systems (DIS ’16), 170–181. https://doi.org/10.1145/2901790.2901879 5. Laura Devendorf and Kimiko Ryokai. 2015. Being the Machine: Reconfiguring Agency and Control in Hybrid Fabrication. In Proceedings of the 33rd Annual ACM Conference on Human Factors in imagine a mode of AI reverb that more readily Computing Systems (CHI ’15), 2477–2486. https://doi.org/10.1145/2702123.2702547 6. Lucian Leahu. 2016. Ontological Surprises: A Relational Perspective on Machine Learning. In Proceedings of the 2016 ACM Conference on Designing Interactive Systems (DIS ’16), 182–186. https://doi.org/10.1145/2901790.2901840 7. Yoko Ono and John Lennon. 2000. Grapefruit: A Book of Instructions and Drawings by Yoko Ono. Simon & Schuster, New York. 8. La Monte Young (ed.). 1963. An Anthology of Chance Operations. La Monte Young & Jackson Mac Low.