Second session: Facing Technologies
30 mrt. 2023
Delfina Sol Martinez Pandiani, University of Bologna
This presentation will focus on the critical role that computer vision labels play in identifying and understanding abstract concepts, such as gender, in visual media. I will explore the challenges of naming and labeling these abstract concepts in computer vision pipelines, and the limitations of binary thinking in solving these "wicked problems." I will also examine the dangers of approaching these issues as if they were "tame issues" and the opportunities that arise from a more nuanced understanding of these abstract concepts. Furthermore, I will present some recent cognitive-neuroscientific theories about abstract concept representation, and how they can be translated into components of software architectures that can learn which multimodal features may imply certain abstract concepts, and thus associate these concepts with images. Overall, the presentation will focus on the intersection of wicked problems, complexity, and computer vision from a humanistic and ethical AI perspective, arguing for combining and questioning established binary knowledge and new approaches in naming and labeling abstract concepts in visual media.