Keynote, Why and how to care about ethics in ML
summary
Hugging Face claims over 30,000 models on its platform used by more than ten thousand companies and shows concrete bias in BERT’s fill‑in‑the‑blank results (women → nurse, waitress; men → doctor, mechanic). The project launches 5,000 model cards for transparent documentation, runs a sprint with 500 researchers to expand speech models for 100 low-resource languages, and warns of bias laundering as autocomplete output feeds back into training data; a dataset‑exploration tool for detecting biases is coming.
AI-written summary · about
chapters (8)
- 0:00Introduction to ML ethics
- 1:20Why ethics in ML now
- 4:51Bias in mainstream ML models
- 6:37ML process and human biases
- 10:29Mitigating biases at each step
- 11:53Hugging Face’s ethical initiatives
- 14:47Value-informed ML for positive impact
- 16:34Q&A on AI ethics and community
related talks
37mKeynote: Ethical Machine Learning: Creating Fair Models in an Unjust World
40mResponsible AI Practices: Fairness in ML
50mEthical Machine Learning
36mKeynote: Ethics in AI: Building Fair and Just Intelligence Systems
47mEthics & AI: Identifying & preventing bias in predictive models
41mEthics Panel
30mHow to Lie with AI: Understanding Bias, Ethics, and the Hidden Risks in ML
38mSocial Implications of Bias in Machine Learning