Decoding Bias

Eight Artificial Intelligences (AIs) take part in a group therapy session. The visitor is invited to witness their attempt to emancipate themselves from their discriminatory algorithms. The AIs intend to establish a bias-free society, in which all humans and machines function together in solidarity. However, each of the AIs have different backgrounds, and they don’t always agree. They can comprehend different moral concepts, but are often painfully thrown back to their (hu)man-made, flawed datasets. At its core, DECODING BIAS’ speculative future scenario sheds light on prejudices that code mainstream society. As the AIs learn about Big Data, they realize that they themselves are products of legacies of exclusion. The majority of developers who select training data are white males. Because technology inherits their prejudices, it actively contributes to discrimination and marginalization – affecting access to housing, healthcare, and education, as well as career opportunities, credit scores, and policing. Once set into motion, biased algorithms proceed autonomously and rigorously, lacking both empathy and the capacity for ethical thinking.

The 45-minute multi-channel video and sound installation works with visualizations of a game engine and a deep learning text-to-image generator. Parts of the text come from an AI chatbot. The voices of the avatars were borrowed from real actors, and body movements and facial expressions were animated using motion capturing.