Keynote Speaker – Margaret Mitchell
summary
Creates early image‑description systems in 2011 that surpass human BLEU scores yet fail blind users because BLEU misses real needs; builds a Seeing AI prototype and a storytelling model that exhibits an everything is awesome failure mode. Implements the workflow in Python using NLTK TensorFlow scikit‑learn pandas and Gradio Streamlit, and publishes open‑source tooling on Hugging Face to measure data with Zipf checks co‑occurrence counts and normalized pointwise mutual information plus a data set card system.
AI-written summary · about
chapters (12)
- 0:00Introduction and background
- 1:32Evolution of programming
- 6:01Transition to machine learning
- 8:27Real-world impact of AI
- 12:31The importance of foresight
- 14:42The 'everything is awesome' bias
- 17:55Bias in the ML pipeline
- 22:51Value judgments in deployment
- 26:08Tensions in data development
- 29:51Operationalizing data ethics
- 33:30Measuring data quality
- 37:49Future directions and closing
related talks
45mKeynote: Django Reimagined For The Age of AI with Marlene Mhangami
49mKeynote 4: Here Come The Robots - Python and Machine Learning
58mKeynote: The past, present, and future of AI for application developers
28mKeynote, Why and how to care about ethics in ML
59mKeynote: The past, present, and future of AI for application developers
33mKeynote: Fine-tuning our way towards openness in AI
39mKeynote 3: How did you know? Explaining Black Box Model Predictions in Python
26mKeynote: AI in the Real World: Today and Tomorrow