Panel: Building responsible, human-centric AI models
summary
Panelists argue that responsible, human‑centric AI requires concrete practices not buzzwords: apply an engineering safety factor to models, treat datasets as biased artifacts, and sit next to the user for days to observe real decisions before automating them. They cite examples: flawed recommender outputs (Amazon suggesting duplicate car seats), a support‑engineer recommendation system that improved after onsite shadowing, and CFD and weather analogies showing when models mislead. Key fixes are industry‑specific rules, practitioner certification, and legal guardrails to catch harms like video‑doorbell surveillance.
AI-written summary · about
chapters (20)
- 0:00Introduction
- 0:38Bio
- 1:20Background
- 2:57Introductions
- 5:58What is humancentric AI
- 11:10Data algorithms bias
- 14:17Who will consume the model
- 17:01Data privacy
- 17:45Human logic
- 18:59An anecdote
- 20:14How do we make them better
- 22:57Bringing the human into the conversation
- 30:29Legal challenges
- 31:36Bifurcation
- 34:19Personal story
- 37:58Industryspecific regulations
- 39:21Certifications
- 41:34Balancing needs and wants
- 42:38Diversity and inclusion
- 43:27Conclusion
related talks
47mEthics in AI Panel
40mSocially responsible AI: Practical tools for robust design
33mResponsible AI in Practice: Real-World Examples and Challenges
34mResponsible AI: The Key to Ethical, Safe and Inclusive Software Development
56mEthics in Machine Learning Panel
44mPanel: AI Feminism
33mEthics in Machine Learning Panel Q&A
26mPanel Discussion: Ethical AI - From talk to practice