Lightening Talk about the power of diverse teams
summary
A facial‑recognition system is built by feeding many pictures into a machine‑learning pipeline that creates a neural map trained on a homogeneous dataset, and it unlocks phones reliably for white users but fails for non‑white faces, forcing people to make odd expressions for it to work. The concrete claim is that dataset composition — not the model architecture — causes the failure; the team ships a working product for some users and a biased, unusable one for others. facial recognition bias dataset-driven failure
AI-written summary · about
related talks
56mGlobal Bias in AI: When Western Data Shapes the World
47mDiversity and Inclusion for Algorithmic Bias Considerations
54mHow diversity drives excellence in our data driven tech world
39mBecoming Conscious of Our Unconscious Bias in Design
10mWhy diversity matters in Machine Learning communities
45mBias in Artificial Intelligence
10mWe want to believe, but we don't
29mUnintended consequences of technology: Enabling NextGen discrimination