Artist and Machine: An Iterative Art Performance

• Susie Fu
• susie@susiefu.com
• Independent Artist, Portland, Oregon, USA

Keywords: Artificial Intelligence, Human-Machine Collaboration, Human/Machine Labor, Authorship, Iterative Performance, Selfie Culture, Continuous Improvement .

Artist and Machine is an allegorical performance series that examines the impact of artificial intelligence and the dialogue between these machines and their creators. The piece condenses the development of AI-powered technology into the confinement of a performance space, taking normally unseen data and processes and making them physical and experienceable for the viewer. It asks the audience to consider the achievements, consequences, and implications of a society dependent on and intertwined with automated machines.

The Performances

In each performance of Artist and Machine, the audience sees an Artist and a Machine sitting side-by-side, performing the same task (Fig. 1). Over the span of several hours, they continuously draw portraits of the people they see, on two parallel strips of paper that document the passing of time and people. Using Neural Style Transfer techniques (Gatys 2015), the Machine tries to draw like the Artist.

After drawing hundreds of people by the end of each performance, the Artist has developed and practiced her technique. Then, the machine takes the updated training data from the Artist and also improves its imitation. The feedback loop continues for the next performance iteration.

Compare and Contrast

The Artist operates serially, picking one volunteer at a time (Fig. 2). Able to see and draw multiple people at once, the Machine is far more prolific but is still developing its human touch (Fig. 3-4). The Artist takes many breaks while the Machine needs only paper refills. The two performers reflect the complicated distinction between the labors of humans and machines.

As expected, participants interact very differently with the two performers (Fig. 5). They have an intimate and uncomfortable staring session with the Artist, seldom repeating the experience. In contrast, the viewers voluntarily return multiple times to the Machine. They choose to become users, treating the Machine like a mirror, a device, a service.

Unexpectedly, the participants themselves become the real performers. The polarizing behavior of the audience reveals the actual performance: how we change and respond to the inevitable shift towards an automated world.

Additional Information

Technical Description: Artist and Machine uses open-sourced libraries: Fast Neural Style (Johnson 2016), OpenCV, Torch, and custom code. The installation hardware setup includes a MacBook Pro loaded with a custom trained model (trained using an Nvidia GPU), a webcam, a thermal printer, and a Raspberry Pi.

Future Work: To date, there have been three performances in the United States. As continued iteration is a central component of the series, live performances will continue in galleries, conferences, and other spaces in the United States and abroad. A retrospective of the performances is in progress.

References

• Leon A. Gatys, Alexander S. Ecker and Matthias Bethge. 2015. A Neural Algorithm of Artistic Style. arXiv:1508.06576.
• Justin Johnson, Alexandre Alahi, and Li Fei-Fei. 2016. Perceptual losses for real-time style transfer and super-resolution. Berlin: European Conference on Computer Vision.