Zoom out

Zoom Out transforms a 3D-scanned home office into an attentive space that searches together with its inhabitants for moments of inspiration. Developed during the pandemic as part of the EU Horizon 2020 STARTS project MindSpaces, the work addresses a tension of remote work: the very familiarity of domestic space that enables comfort also dulls perception and suppresses creative arousal.

A Deep Reinforcement Learning agent navigates the latent space of a StyleGAN2 network trained on the Describable Textures Dataset, continuously transforming the virtual environment’s surfaces in real time. The agent’s observations are driven by the inhabitant’s galvanic skin response, measured through an Empatica E4 wristband, creating a feedback loop between physiological arousal and spatial mutation. Two prototypical systems explore different strategies for approaching inspiration: one targets a constant arousal value, the other follows a narrative arousal curve, each producing distinct experiential qualities.

The work does not aim at optimization. Its deliberately low learning rate produces an aesthetics of searching rather than finding. When training a Deep Reinforcement Learning model with physiological feedback, it takes weeks to achieve computationally meaningful results. Yet this apparent limitation becomes the work’s defining quality: a space that does not deliver inspiration on demand but searches alongside its inhabitant, uncertain and attentive. What begins as a technological experiment becomes a question about the conditions under which familiar environments can become artistically productive again.

Personal narration:

Welcome to my home office.

I have spent the last two years here. Before the pandemic, my home was my base for relaxation and time off – familiar and with low arousal.

As an artist, creativity and inspiration are essential to my work. Usually, my sources of inspiration are arousals and unplanned impressions throughout my day.

When during the pandemic I suddenly had to work from home, and I had a hard time being creative. I needed to break out, but of course, without actually leaving the room.

I 3D-scanned my home office. As you can see, the resulting virtual reality is not perfect at all. Just look at the blob of flowers on my table or my notebook without its screen. These distortions from reality already triggered something in me. So I decided to take it even further.

I started to experiment with tracking my arousal through skin conductivity. And I hooked this real-time data up with a machine learning model that then learns to transform my virtual home office.

Through a partnership of MindSpaces and Eina Idea, “Zoom out” was shown as part of SYNX at Sonar+D and STARTS AI and Music Festival at CCB Barcelona.

More information on the embedded human-A.I.-interaction, the arousal detection with Galvanic Skin Response (GSR) and the utilization of Stress-Aware Deep Reinforcement Learning in Virtual Environments can be found in this publication by myself, Maria Kyrou, Panagiotis C. Petrantonakis and Ioannis Kompatsiaris: https://dl.acm.org/doi/pdf/10.1145/3430524.3440647

Collaborators
Centre for Research & Technology Hellas (CERTH), up2metric, McNeel, Zaha Hadid Architects and Analog Native

Credits
Panagiotis Petrantonakis & Maria Kyrou – arousal tracking through Galvanic Skin Response (GSR)

Advice and support
Magdalena May – organisational psychology
Manuel Cirauqui – curation of SYNX.
Miquel Cardiel & Flor Salatino – integration in SYNX.
Mans_O – SYNX soundscape.

References excerpt

Emanuele Coccia. 2016. Sensible Life: A Micro-ontology of the Image. Fordham University Press„ New York, NY :. https://doi.org/10.1515/9780823267446

Todd M Thrash and Andrew J Elliot. 2003. Inspiration as a Psychological Construct. Journal of personality and social psychology 84, 4 (2003), 871–889. https://doi.org/10.1037/0022-3514.84.4.871

Helmut Leder. 2003. Familiar and Fluent! Style-Related Processing Hypotheses in Aesthetic Appreciation. Empirical studies of the arts 21, 2 (2003), 165–175. https://doi.org/10.2190/G6MK-6KL7-KETB-ELND

William J Dickson and F. J Roethlisberger. 2003. Management and the Worker. Routledge, Florence. https://doi.org/10.4324/9780203503010

Bernat Cuni. 2020. Deep Textures / RunwayML residency — cunicode / Digital Craftsmanship. https://www.cunicode.com/works/deep-textures