The Audio Long Read Hidden traces of humanity: what AI images reveal about our world
7 snips
Nov 4, 2024 In this intriguing discussion, Rachel Ossip, a skilled author at The Guardian, dives into the intersection of AI and human creativity. She explores how AI-generated images hold unexpected traces of human influence. Listen as she discusses the whimsical side of AI image creation, including generating playful cat visuals. Rachel also tackles the implications of biases in AI datasets and critiques the superficial solutions often proposed. This conversation reveals the fascinating challenges and ethical dilemmas of merging technology with art.
AI Snips
Chapters
Transcript
Episode notes
Early AI and Human Labor
- Early AI image generators were trained on limited datasets like flowers and birds.
- These datasets, often labeled through crowdsourcing platforms like Mechanical Turk, reveal a reliance on human labor.
AI-Generated Faces and Deepfakes
- StyleGAN, trained on Flickr images, generated realistic faces, leading to websites like "This Person Does Not Exist."
- This sparked concerns about deepfakes and threats to democracy.
ImageNet's Categorization Problem
- ImageNet aimed to categorize the world of objects but inherited flawed, biased taxonomies.
- This highlights the danger of oversimplification and bias in large datasets.