The Future Through Artificial Eyes, 20 Years of VPRO Tegenlicht

Independently thinking machines have captivated the human imagination since at least the nineteenth century, mostly in two opposing imaginaries: will they unlock a new, ideal world where labour, inequality and war are mere memories? Or will they bring about a doomsday scenario in which robots eventually take control? In reality, algorithms and machine learning can be found everywhere in the 21st century, but most people do not know how they work.

The Future Through Artificial Eyes marks the 20th anniversary of the Dutch television programme VPRO Tegenlicht. VPRO Tegenlicht and Het Nieuwe Instituut have collaborated with artist and designer Richard Vijgen in analysing visions of the future via the social, economic and technological developments of the 21st century by focussing specifically on the capacity of Artificial Intelligence (AI) to recognise patterns in images.

At the heart of the exhibition is a multimedia installation designed by Vijgen in which an AI trained to find patterns in images scans 20 years of VPRO Tegenlicht broadcasts and points out what it recognises. Visitors can influence the outcome by playing with the way the ‘artificial eye’ views the programme’s archive.

First fully automated loom invented by Jacques de Vaucanson (1709-82) c.1745
The automaton Digesting Duck by French inventor Jacques de Vaucanson (1709-1782), created in 1739.
Plate IV, 'Uber den Schachspieler des Herrn von Kempelen und dessen Nachbildung', Leipzig und Dresden: Joh. Gottl. Breitkopf, 1789
Plate III, 'Uber den Schachspieler des Herrn von Kempelen und dessen Nachbildung', Leipzig und Dresden: Joh. Gottl. Breitkopf, 1789
Tweet by Jo Liss, 8.12.2015 “When computers design things, they look very different. Tensile structure before/after topological optimization”
The object on the left is human-designed. The one on the right has been designed by a computer, which calculated the tensile and material strength needed to improve upon the human-made object. The alternative proposed by the computer has been generated through the use of a genetic algorithm that uses its neural network to present the most optimal version of the original object.
The Twitter account ‘neural net guesses memes’ runs popular memes through an image recognition neural network and publishes the resultant image prediction and level of confidence. It is clear that the computer cannot grasp the context and nuance of memes; it becomes an unwitting generator of a second level of humorous memes
However, when a neural network assesses real-life human concerns, such as whether a person deserves a loan or if they fit a job description, the confidence threshold becomes a sensitive issue.