Roll for Initiative: how to make the world of AI a more ethical place
summary
Claims concrete fixes and evidence not platitudes: a project demonstrates visibly de-biasing a GAN, chatGPT is used to generate an insult roll table, and dataset sourcing practices like Amazon Mechanical Turk get criticised. Cited numbers include 80% of AI professors are men and 72% of conference presenters, and a study finding 3.4% label error in ImageNet (482,702 images). Practical mitigations discussed are curated smaller models, hyperparameter and sample-weight tuning, and multi-pronged remediation rather than single quick fixes.
AI-written summary · about
chapters (4)
related talks
26mBuilding ethical AI: Transparency, bias mitigation, and trust
28mEthical AI?!
21mBe more responsible around AI - Less Bias, More Ethics
27mChallenges of Generative AI
38mBuilding Responsible AI in the Age of GPT
35mAI ethics for tech writers
43mAI Engineer Ethics: Shaping Responsible Generative AI
06m7 practical tips for safer AI and language models