Child in the Wild (2017)

Child in the Wild is an interactive installation that enables human participants and a child robot to co-create an immersive audiovisual artwork through the use of the robot's artificial neural networks to enable object and image recognition.

The resulting artwork dissolves the boundaries between computational and physical phenomena, displaying an aesthetic that is a real hybrid of the physical and the digital, of human and machine learning, of natural and artificial intelligence, and of real and synthetic evolution. It is an artwork and aesthetic that emerges from the interaction between robot, people and virtual environment, neither one taking precedence, rather collaborating on a genuinely post-digital, post-convergent artwork.





Child in the Wild comprises a child robot in a stroller, interacting with visitors to the gallery. Participants show the child robot objects or pictures on their mobile device and the child robot will try to guess and say what the object is. The child robot's perception of the material world is audio-visually displayed via projection on to the gallery walls and floor, and surround sound system. Over time the image and sound disintegrates into its component pixels and sonic grains, which, via an artificial intelligence network, agitate into co-evolving colour and sound fields and re-influence teh child robot's perceptions.




Child in the Wild uses deep learning artificial neural networks running on the child robot's embedded computer to enable it to recognise and converse about its surroundings. The projections, which can cover the whole space (3 walls and floor) are manifested from the child robot's inner state as it interacts with human visitors. The complex, evolving sound score is also a generative expression of the child robot's state as the sound interacts with the decomposing visual memories of the child.

Child in the Wild investigates learning, robot-machine interaction, the development of material understanding and the notion of desire this entails. For Jacques Lacan, desire is a desire for recognition from the other, and the sense of what the other desires, what it lacks. [1] For Julia Kristeva, desire is an impulse towards the other, that "leads the way, is tied to lack, and expresses the body's needs." [2] In contrast, for Deleuze and Guattari desire is a productive force that creatively incorporates into the self that which is other to it. [3]

Contemporary deep learning artificial neural networks are primarily based on fixed notions of representation and signification. [4] For Lacan, this would imply both fixity of meaning and fixation of desire, which in turn implies limitation of jouissance, rather than an unbounded, uncontrollable jouissance. [5] But it is jouissance that is at the root of creativity, art and other wild systems.

For Child in the Wild, we propose artificial desire as a motivational force for accomplishment in non-human entities, whether it be goal oriented behaviour, artistic creation, self learning or interactions with humans and others. Accordingly, the child robot interacts with the sound and vision of the projected virtual environment using principles of artificial desire that we have developed, and which evolve over the period of the exhibition, creating an emergent, recombinant relationship between robot, artwork and the public.




Child in the Wild takes a sophisticated hacker approach to the development and production of the artwork. The child robot is comprised of 3D printed parts and runs on a raspberry pi. Though the technology is maker aesthetic, the resulting capabilities of the robot are significant. Similarly, though the audiovisual aesthetic is of realtime games, its generative, evolving composition is sophisticated and nuanced.

The use of artificial neural networks to develop the child robot’s learning and desire is a technique that departs from traditional programming of robots, with less emphasis on prescribing abilities and more on learning and experience as the basis for capability. [6] Along with this learning oriented approach comes the issue of relative control. Artificial neural networks are often regarded as a black box regarding function. Yet in the cultural imagination promulgated by Asimov, robots are controlled servants that cannot harm humans. How does this correlate with non-human entities whose development is based not on hard coded rules, but on emergent models of both supervised and unsupervised learning? Further, how can deep learning be integrated with embodiment and desire in order to draw inspiration from living models?

Child in the Wild is an immersive experience as an installation. It invites participants to interact directly with the robot to co-create the art experience. The encompassing projections surrounding the participants on all walls and the floor position the participants and the child robot within a capsule of their shared desires. The disintegration of images into colour fields from which the sound is generated harks to the changeability and fragility of memory and perception, perhaps even to its unreliability. Child in the Wild seeks to challenge and motivate both human and non human participants to develop a unique, timely experience.




The work can be set up in any gallery, and can scale from a discrete single-screen display with headphones up to a full room immersive environment, using our custom mirror-based projection system, with surround sound.




  1. Jacques Lacan. 2014. Chapter 2: Anxiety, Sign of Desire. Anxiety, the Seminar of Jacques Lacan Book X. Polity Press. p. 23.
  2. Julia Kristeva. 2010. The Passion According to Motherhood. Hatred and Forgiveness. Columbia University Press. pp.80-81.
  3. Gilles Deleuze and Felix Guattari. 1983. Anti-Oedipus: Capitalism and Schizophrenia. University of Minnesota Press. p.49.
  4. G.E. Hinton. 2007. Learning Multiple Layers of Representation. Trends in Cognitive Science, 11, pp.428-434.
  5. Bruce Fink. 2007. Fundamentals of Psychoanalytic Technique: A Lacanian Approach for Practitioners. Norton and Company. p. 269.
  6. Ogata, T. et al. 2009. Prediction and Imitation of Others' Motions by Re-using Own Forward-Inverse Model in Robots. ICRA'09, IEEE International Conference on Robotics and Automation. p.4144-4149.



This project has been assisted by the Australian Government through the Australia Council, its arts funding and advisory body