Machine Unlearning for Artists
Workshop, 2026
Welcome to "Machine Unlearning for Artists"
This hands-on workshop demystifies Artificial Intelligence for creators. Moving beyond the illusion of the AI "black box," you will explore the hidden mechanics, systemic biases, and critical vulnerabilities of modern machine learning models. By adopting a hacker mindset, you will learn not just to use AI, but to actively challenge, manipulate, and creatively exploit it.
Key Topics & Practical Explorations:
- Demystifying AI Basics: Understand core concepts like weights, biases, and epochs. Train and deploy your own image classifiers using tools like Teachable Machine and Google Colab (ConvNet, MobileNet).
- Surveillance & Ethical Implications: Explore the dark intersections of AI, mass surveillance, and "automation bias." Analyze real-world tools like PimEyes and discuss creative resistance strategies such as "CV Dazzle" anti-surveillance makeup and subvertising.
- Algorithmic Bias & Machine Perception: Look inside the "engine room" of AI. Expose the systemic flaws in massive, internet-scraped datasets (like LAION-5B) and see how models like OpenAI's CLIP truly perceive the world.
- Adversarial Attacks (Vision & Audio): Learn how to trick machine perception using simple text overlays (typo attacks), data poisoning, hidden backdoor triggers, image scaling exploits, and targeted audio noise against models like Whisper.
- Model Extraction & Analysis: Discover how to reverse-engineer AI. Extract, translate, and clone functional AI architectures, and use feature visualization to see exactly what triggers a neural network's decision.
- AI Red Teaming & Prompt Injection: Hack Large Language Models (LLMs) and AI agents. Master jailbreak techniques (like "DAN" or "Stan" modes) to bypass safety guardrails, hijack chatbot memories, and manipulate a simulated AI art auction bot to reveal secret reserve prices.
