Human-centered technology and the role of Responsible AI
summary
Describes concrete fixes: Search uses BERT to reinterpret ambiguous queries so explicit or stereotyped results get reduced and demoted problematic content, improving relevance for queries like “Latina yoga instructor” and general queries too. Camera teams build with image experts and Real Tone techniques to change auto exposure and auto white balance so portraits of darker skin are accurate, producing natural, more beautiful portraits. Emphasizes auditing each pipeline step, recruiting diverse reviewers, and finding internal champions to operationalize fixes.
AI-written summary · about
chapters (4)
- 0:00Introduction
- 0:21Representation and Inclusion
- 3:58How do we drive this work
- 9:17Dont underestimate your own voice
related talks
44mPanel: Building responsible, human-centric AI models
47mEthics & AI: Identifying & preventing bias in predictive models
03mShowreel
40mSocially responsible AI: Practical tools for robust design
28mEthical AI
34mResponsible AI: The Key to Ethical, Safe and Inclusive Software Development
41mWhen AI Failures Reflect Our Own Social Biases
11mData-centric AI: A new paradigm