AI for Inclusive Society - Responsible AI
summary
Explains how large language models are built from public web data then fine‑tuned with RLHF, warns that models mirror societal harms with examples (Google gorilla, biased credit limits, Amazon hiring tool, Tay) and notes 180 cognitive biases and 1 billion people affected by disability as scale reasons to act. Shows a personal multilingual chat instance using GPT‑4 and points out the practical limits like a 4,000 token limit, then gives concrete steps: dataset checks, prompt design, and diverse teams.
AI-written summary · about
chapters (9)
- 0:00AI and Responsible Innovation
- 3:28AI metaphor: Blind Man and Elephant
- 9:13AI landscape and generative models
- 19:13Natural language interaction
- 24:14Societal exclusions and biases
- 35:47Responsibility in AI development
- 44:07Steps for inclusive AI systems
- 51:22AI for all and future humanity
- 53:09Q&A on AI ethics
related talks
34mResponsible AI: The Key to Ethical, Safe and Inclusive Software Development
54mAI and Ethics: Navigating the Future of Responsible Technology
40mSocially responsible AI: Practical tools for robust design
38mBuilding Responsible AI in the Age of GPT
44mAI, Corporate Responsibility & Democratic Legitimacy
19mMachine Intelligence With Humans in Mind
19mMoral Outsourcing: Humanity in the age of AI
35mDesigning Inclusive and Accessible Experiences in the Era of AI