Fireside chat: Breaking the wall of unethical AI
summary
Advocates an evidence-based workflow: require bias assessment before launch, involve diverse beta testers, and use curated training datasets that include edge cases rather than relying on big-data proxies. Recommends operational changes like mandatory voluntary self-identification fields, product changes (pronouns, chosen name, color contrast, captions), and embedding DEI goals into performance metrics. Cites concrete harms — resume-sorting that screened out women, smart‑home abuse and facial‑recognition misuse — as reasons to act now.
AI-written summary · about
chapters (12)
- 0:00Intro
- 0:38Introducing Keith Plummer
- 3:50Diversity equity and inclusion
- 7:13Good practices and working groups
- 9:20Bias and fairness
- 12:27Status quo or average
- 14:29Longterm integrated strategy
- 16:35Proper alignment
- 18:20Product inclusion
- 19:34Use cases
- 22:57Recommendations
- 25:44Audience questions
related talks
36mKeynote: Ethics in AI: Building Fair and Just Intelligence Systems
05mToward Ethical AI
47mEthics in AI Panel
47mEthics & AI: Identifying & preventing bias in predictive models
39mThe Ethics of Artificial Intelligence - What...
45mEthical use of AI: seeking, understanding, and countering the risks
56mEthics in Machine Learning Panel
26mPanel Discussion: Ethical AI - From talk to practice