Mitigating Bias and Fairness in AI

Introduction to the concept of Bias in AI, why it matters and how to mitigate the effects of bias in data and algorithms for AI.

Transcript

2. Big Data & AI Conference Dallas, Texas June 27 – 29, 2019 www.BigDataAIconference.com
3. Fairness & Mitigating Bias in AI Babar Bhatti Jun 28, 2019
4. Trends & Implications Ethics & Values Fairness Inequality Technology Jobs & Work Education Geo-Politics Economy Society Transparent, Reliable, Safe Deep Learning, Narrow vs AGI IoT, Blockchain, Cloud
6. Today’s Talk: Fairness, Bias, Trust ... • Power and Perils of AI - pervasive, invasive, limited • State of Fairness in AI today - Business impact of Biased Models • Black box vs Explainable AI • Ethics, Privacy --> Trust • Key Reasons why Bias occurs • Approaches to Mitigate Bias • Action Plan
7. The Trouble with Bias -- Kate Crawford, AINow
9. Explainability Problem of AI
10. Approaches to Explainability Problem
11. Explainability vs Performance
13. Deepfake Challenge
14. TOP REASONS FOR BIAS IN AI AND ML Data Bias Algorithmic Bias Human Bias (inequity captured in data)
15. Biased Medical Datasets Medical data in the US is extremely male and extremely white, and that has real- world impacts -- Quartz.com
16. ML Bias Research
17. Need Diverse, Multidisciplinary Experts Source: Algorithmic Justice League https://www.ajlunited.org/
18. Self-Regulation vs State Regulation
19. Bigger than tech - multiple levels, stakeholders and perspectives: national, organization and society / individual level Code of ethics, monitoring, governance and self-regulation - share best practices AI experts need to work with stakeholders from diverse perspectives: social sciences, policy makers, public-private partnership Ethical and fair AI methods require more research and focus - recent funding by Steve Schwarman’s to Oxford for research on AI ethics Additional Sources: • Medium Blog: medium.com/@thebabar (checklist & references) • My linkedin: Recommendations + Resources
20. Explainability Problem of AI