Reinforced Learning
by Jeroen van Loon
The Netherlands 2023
Installation with LED Screens, transport crate, and video (9:58 minutes)
Reinforced Learning is a video installation combining humour and critical thought to challenge scenarios for the future of robotisation and AI. It uses using AI-generated visual effects in a compilation of “fail videos” showcasing the irrational and inefficient actions of robotic humanoids that have replaced humans. The videos present a strange future in which robots mimic our illogical behaviour and defy our widespread belief in their non-human capacity for perfection.
The idea underlying many scenarios for the future is that by collecting enough data, reality can be interpreted and predicted, allowing “efficient” AI systems and robots to take over large parts of our reality. These short and amateurish fail videos present the exact opposite of that possible future, making it a perfect springboard for criticising and undermining these future scenarios. The artist reflects on the “Garbage in, garbage out” principle and decides it’s time to put some garbage in.
The title Reinforced Learning comes from the field of machine learning, in which algorithms learn by mimicking specific actions. The algorithm receives a reward when it successfully imitates the intended action. This work shows robots learning from the “wrong” input: human imperfection.