Optimising for Beauty (learning to see beauty)

Brief description

The artist is training a neural network with a dataset of beautiful people (celebrity facial dataset). By choosing to use one of the algorithm widespread algorithms used in Machine Learning and Statistical Inference — Maximum Likelihood Estimation (MLE) the artist demonstrates how the choice of this learning algorithm causes a layer of homogenisation as the machine generates new faces "a race of ‘perfect’, homogeneous specimens". The smooth and blurred faces are erased of detail as the networks learn a ideal, optimised sense of beauty. As the algorithm is not able to deal with uncertainty it is not able to see other possible views of beauty than the most common characteristic of the faces it is trained with. 

Work that the situation appears in

Title Publication Type Year Creator
Optimising for Beauty Art, Video art Memo Akten
Aesthetic characteristics
Colours
Machine P.O.V
Not machine P.O.V.

Authored by

UUID
04cbec03-8ba4-41da-8b45-d877be4dfe93