ILSE
KIND 





2024





Lifetime Approval Rate 96%  







Video  
Title    Lifetime Approval Rate 96%
Year     2024
Material Behind-the-scenes documentation of the artist performing a data annotation task, with simultaneous heartbeat monitoring, LAION-5B Training Database
Specs    Full HD, stereo sound
Sound    MYNGRY 
Duration 7:25 minutes 

Sculpture 
Title    Hand 21

Year     2023
Material Silicone, Mechanics and electronics
Size     23 x 17 x 9 cm 


Sculpture
Title    Hand 41

Year     2024
Material Silicone
Size     23 x 20 x 10 cm


Sculpture
Title    Hand 71

Year     2024
Material Silicone
Size     29 x 11 x 5 cm


Sculpture
Title    Hand 14

Year     2023
Material Silicone, Mechanics and electronics
Size     21 x 18 x 8 cm





















    Perched atop the mountain’s summit, where humanity has long sought to touch the divine, the exhibition dismantles the myth of Artificial Intelligence as an autonomous force. 

Artificial intelligence is often imagined as autonomous, immaterial and self-sufficient. Yet behind every AI model lies a hidden and unacknowledged infrastructure of human labour: annotating images, evaluating outputs and training the datasets through which AIs learn to understand our world. Ilse Kind's work emerges from her own experience as an annotation worker; under unpredictable, repetitive and underpaid circumstances. This work makes those hidden bodies tangible, exposing the rhythms, repetitions and physical demands embedded within the production of machine intelligence.

At the center of the exhibition, the video Lifetime Approval Rate 96% presents a full-length Human Intelligence Task. Its title refers to the score that evaluates workers’ performance. Precisely instructed, closely monitored and bound by confidentiality, Kind captured what was allowed: the clicks of her hand on the trackpad and the rhythm of her heartbeat as she worked. These sound recordings were made into a soundscape by MYNRGY, amplifying the bodily rhythms of digital labor that usually remain unheard.

On screen, repetitive trackpad clicks annotate training data. The original data is substituted with a leaked training database, used to train AI models in generating images of hands. Medical photographs collide with artist drawings, miniature army figures, and cowboy toys. Fed with this material, AI falters most visibly in its attempts to recreate the human hand; producing distorted, anatomically flawed gestures.

The AI-generated output of the same model trained on this database take sculptural form in Click 14, Click 21, Click 45, and Click 71. Modeled with Kind’s own annotating hands and assembled through the logic of diffusion algorithms, the works collapse boundaries between human and machine, digital process and physical form. Failures that signal to AI’s lifeline of human judgment, selection, and correction. 


[Accompanying essay ’Handwerker; a human resource farm’ on request ︎]


Teaser