Human-Centered AI for Sustainable Development
summary
AI adoption races from Netflix 3.5 years to ChatGPT 5 days to 1M users while model training can emit huge footprints — a 200M-parameter Transformer tuning run is reported at ~280 tons CO2 per run. Real failures include Tay turning racist in 16 hours, biased hiring/hard‑coded face recognition, and a COVID chatbot giving unproven treatments. Practical fixes: audit datasets, collect representative or synthetic data, measure energy/training time and power use, reduce update frequency, add explainability, set KPIs and automated tests and include legal and environmental experts.
AI-written summary · about
chapters (9)
- 0:00Human-centered AI intro
- 3:02Human-centered AI framework
- 6:54AI failures lessons
- 8:38Measures for human-centered AI
- 11:51Sustainability and AI link
- 13:38Reducing AI footprint
- 16:03AI lowering carbon emissions
- 17:51AI adoption vs sustainability
- 19:19Conclusion
related talks
41mThe AI boom: what and why?
19mLet's Talk GenAI
27mThe Double-Edged AI Sword: Examining Applications in Different Fields
17mFocus on Data, on the Road to AI
29mThe Rise of AI: A Reality Check on Generative AI
52mSome pitfalls in AI
50mIs ChatGPT (and Other AI) the Enemy of Diversity?
31mAvoid mistakes building AI products