There is growing criticism of the widespread application of machine-based recognition and data processing, especially those involving visual technologies. The inaccuracy, tolerated as a minority, is second only to the political consequences of the applied criteria. Francis Hunger & Flupke have developed a product that successfully implements this criticism at a technical level. Their “Adversarial.io”, is a webapp that alters images in order to make them machine-unreadable, while leaving them visually almost indistinguishable from the original. It accomplishes this task by introducing a noise pattern that the human eye is able to compensate for, but which induces a different description class for the ML system. The declared mission of the duo is to “fighting mass image recognition”. This attack on the new ‘machine normative’ is not an extemporary gesture, but a methodological programme.

 

Francis Hunger & Flupke – Adversarial.io

Adversarial.io from Francis Hunger on Vimeo.