Content Aware Studies

Artistic inquiry into technologies of artificial neural networks (AI) and machine learning to reconstruct missing fragments in sculptures and friezes from the periods of classical Antiquity as well as to generate never-existed, yet genuine documents of synthetic histories.

Marble, polyamide, machine learning algorithms, 3D scanning, 3D printing, video installation

Technical and artistic assistance: Matthew Lenkiewicz and Artem Konevskikh 

Content Aware Studies series initiates an inquiry into the possibilities of machine learning technologies as tools for speculative restoration. A pretrained neural network model is directed to replenish lost fragments of the friezes and sculptures as well as generate never existed objects of classical antiquity by means of computer vision and machine learning performed on natural and synthetic datasets consisting of thousands of 3D scans of classical sculptures. The algorithm generates results convertible into 3D models, which are then 3D printed in polyamide and used to fill the voids of the original marble sculptures. It tends to faithfully restore original forms, while also produces bizarre errors and algorithmic interpretations of, so familiar to us, Hellenistic and Roman art.
As recent research in General Adversarial Networks has shown outstanding results in hyperrealistic image rendering, it may potentially be applied as a meta-archeological tool for restoration of past and modelling of the futures. However it seems that the role of that synthetic agency in such applications is yet to be critically examined.

The current stage of the project focuses around questions on how would such a canonised aesthetic as classical antiquity look when seen and interpreted through the lense of machine vision, synthetic cognition and sensation.
What visual and aesthetic qualities for such guises would they convey when perceived through our human-centric lens? And what of our historical knowledge, mythology and interpretation, encoded into the aesthetics of the datasets will survive this digital digestion? It questions methods of preservation and reconstruction along with new challenges in those fields posed by automation and artificial learning and production.
The project renders and explores juxtaposed classical and generative methods, their aesthetic refractions, automated forms of production and speculative archeology as a form of history production, while calling into artistic and ethical implications of such actions and readressing questions and notions of human-and-non-human-centric.

Results of the analysis and interpretation of an antique portrait produced by artificial neural network and based on a manually collected dataset consisting of about 10,000 3D scans of sculptures from collections of museums of the Metropolitan, the Hermitage, the British Museum, the National Museum of Rome, and other renowned collections of antiquity.


CAS_08 Hellenistic Ruler;

Marble, Polyamide; Machine Learning Algorithms
Dimensions: 19x26x21;

CAS_09 Colos­sal head of Her­cules

Mar­ble, Polyamide; Ma­chine Learn­ing Al­go­rithms
Di­men­sions: 24x32x20

CAS_07 Tele­phos Frieze
Bot­ti­cino Mar­ble; Ma­chine Learn­ing Al­go­ry­thms
Di­men­sions: 56x67x17

CAS_10 Telephos Drapery;

Carrera marble, Machine Learning Algorithms
Dimensions: 60x40x14

Linux based machine equipped with  GPUs
Performing GAN training during the exhibition and outputs results in real time 


Marble, polyamide, machine learning algorithms, 3D scanning, 3D printing, video installation

CAS_03 Lu­cius_Verus

Car­rera Mar­ble, Polyamide, Ma­chine Learn­ing Al­go­rithms
Di­men­sions: 42x37x32

CAS_04 Parthenon_South_XI_31
Carrera Marble, Machine Learning Algorithms
Dimensions: 120x100x10cm;


‘Open Codes’, ZKM, Karlsruhe, Germany, 2018
Curator: Peter Weibel

Marble, polyamide, machine learning algorithms, 3D scanning, 3D printing, video installation

CAS_05 Julia Mamea; 2018
Crema Marfil Marble, polyamide, machine learning algorithms
Dimensions: 20x35x21

CAS_06 Female Portrait; 2018
Crema Marfil marble, polyamide, machine learning algorithms
Dimensions: 22x26x23