Using AI to Build Fair and Equitable Workplaces
summary
A workforce analytics platform that mines digital communications with NLP and pretrained models (BERT, word embeddings) claims to score sentiment, collaboration, inclusivity and centrality without surveys to surface inequities, but training data reflects societal bias. Detection uses correlation heatmaps, subgroup confusion matrices and disparate‑impact metrics; mitigation includes reweighting samples, label flipping, per‑group thresholds and adversarial de‑biasing; explainability uses local surrogate explanations (perturbations like LIME) to flag biased decision combinations and to audit pretrained embeddings for stereotyped associations.
AI-written summary · about
chapters (20)
- 0:00Intro
- 0:36Today's Speakers
- 1:33Where We're From
- 3:42The Current Environment
- 5:50Equality Requires Fairness
- 9:16Bias Exists in Technology Too
- 11:23Why We All Should Care
- 13:58Creating Fair & Equitable Workplaces
- 17:12Attacking the Problem
- 17:58Understand the Data
- 20:43Checking Model Bias
- 24:42Local Explainability
- 26:44Tackling Bias
- 27:34Data Set Manipulation
- 30:49Debiasing Outputs
- 32:31Bias-aware Algorithms
- 34:11Word Embeddings
- 35:21BERT et al.
- 37:15NLP Explainability
- 38:41Resources
related talks
40mResponsible AI Practices: Fairness in ML
02mThe real project of AI ethics
37mKeynote: Ethical Machine Learning: Creating Fair Models in an Unjust World
31mOperationalizing Responsible AI
36mKeynote: Ethics in AI: Building Fair and Just Intelligence Systems
47mEthics & AI: Identifying & preventing bias in predictive models
40mEthical AI: Addressing Bias & Algorithmic Fairness
23mRemoving Unfair Bias in the Machine Learning Pipeline