MACHINE VISIONS

       

(Installation View) 

Materials: generative video, 3D printed objects, levitation platform

Forthcoming exhibition: Hong Kong  2022

Description: a video installation that stems from the use of a Generative
Adversarial Network (GAN), which is a Machine Learning (ML) algorithm
that attempts to reproduce a given dataset. This GAN was trained on over
2,500 tools and technologies from Paleolithic stone implements to
contemporary gadgets, such as cellphones, creating in its output strange
renditions of familiar objects. Some of the generated models are recognisable
as human-made tools for an unspecified or imaginary use, while others are
abstractions that are remnants or ruins of the algorithm’s training process.

In the background of the installation, there will be a generative video with
a morphing landscape that is controlled by an ML text generator known as
GPT-2. This text generator was trained on science fiction novels about AI and
technology, philosophical texts, and technical manuals about tools and AI.
The text is generated and spoken out loud as an ongoing monologue.
It will be played in multichannel sound to allow the voice to travel
throughout the installation. As the text is generated and spoken, it
is analysed for its sentiment and delivery which alters the video landscape,
seasons, weather, and terrain in the video in real-time.

Video Documentation (work in progress, watch in HD):