How does an AI system semantically compute with images or texts? Where do assumptions, biases, and prejudices arise, and how can we navigate these contingent results in both conceptual and practical ways?

In this extended workshop, we will experiment with Python programming to gain insights into how these black boxes operate. The course also serves as an introduction to how we, as artistic researchers, can utilize generative AI while maintaining a critical standpoint grounded in its technical conditions.

Most image-generating models are based on systems that translate between language and images, a novel phenomenon in which visual and linguistic meaning intersect computationally. Human vision is a deeply contextual process, drawing on personal history, cultural frameworks, and embodied experience. Machine vision, by contrast, operates through statistical analysis of vast datasets, reducing image complexity to mathematical relationships.

In practice, we will explore the mechanisms of these vision systems through hands-on programming exercises. We will investigate where biases originate and how they can be emphasized or made visible through artistic strategies, opening possibilities for developing visual-based installations or moving image practices. We will discuss ideas from multiple perspectives and references, engage with diverse voices, and situate our practice within broader conversations on critical AI in art.

Related topics include Ollama, Contrastive Language-Image Pre-Training (CLIP), Large Language and Vision Assistant (LLaVA), and image-generating models such as Stable Diffusion.

• Active participation throughout the workshop
• Engagement in practical exercises

The course will be held in collaboration with Ting-Chun Liu.

Maximum number of participants: 15

Language of instruction: English