Addressing Algorithmic Bias
summary
Defines algorithmic bias with concrete failures: Google’s image tagger labeled Black people as gorillas (2015), three major facial‑recognition systems fail on a Black woman, and Amazon’s recruiting model downranks résumés mentioning “woman.” Recommends operational fixes: create an algorithm inventory, run change one input audits, score input weights, retrain or remove biased proxies, and deploy continuous, independent validation and cross‑functional governance to reduce disparate error rates even when accuracy requires trade‑offs for parity.
AI-written summary · about
chapters (17)
- 0:00Intro
- 1:10Heuristics & biases
- 3:43The cognitive bias codex
- 4:45Algorithmic bias
- 7:00Examples
- 14:57Where does bias enter the algorithms?
- 17:04Fair algorithms
- 20:26Analysing trade-offs when choosing who to protect from algorithmic bias
- 23:58Protecting groups, protecting individuals
- 25:12Accuracy is fairness
- 28:23Combating bias in algorithms
- 31:20Screening algorithms for bias
- 31:51What can we control?
- 35:41Algorithmic auditing
- 38:06Challenges with addressing algorithmic bias
- 39:33Mitigating algorithmic bias
- 43:02Outro
related talks
26m(Un)ethical Artificial Intelligence: How to Keep the AI Fair for Everyone
53mIt’s Not Fair! Detecting Algorithmic Bias with Open Source Tools
11mAddressing Privacy, Bias, and Fairness in AI Systems
54mEthical Machine Learning
21mEthics in the age of AI
47mAlgorithmic bias is everywhere (especially at Breeze) - what can we do about it?
23mDeveloping an intersectional framework to analyze biases in artificial intelligence and deep neural
30mInside AI: What Every Professional Needs to Know to Stay Relevant, Protected, and Ahead