VOICE Summit 2019 | VED Talks | Nicol Turner-Lee | July 24
summary
Algorithms trained on skewed data systematically steer people toward predatory lending and denied credit, and criminal‑justice models reproduce incarceration disparities because training sets mirror biased arrests and convictions. Deep‑learning image manipulation enabled 2016 election abuses; voice systems misrecognize ethnic or foreign accents. Proposed fixes are cross‑disciplinary design teams, mandatory bias impact statements, audit trails and civil‑rights enforcement plus safe‑harbor sandboxes for anti‑bias testing and auditable training datasets.
AI-written summary · about
chapters (4)
related talks
37mKeynote: Ethical Machine Learning: Creating Fair Models in an Unjust World
38mSocial Implications of Bias in Machine Learning
42mLet's Talk about AI, ML and Bias
47mEthical Machine Learning
51mDigital Discrimination: Cognitive Bias in Machine Learning
38mSocial Implications of Bias in Machine Learning
05mToward Ethical AI
36mKeynote: Ethics in AI: Building Fair and Just Intelligence Systems