Breaking the Chains of Oppressive Software
summary
Shows how specific systems skew results: Google Word2Vec embeds 3 million Google News words into 300‑dim vectors that reproduce gendered analogies; Google Photos fixed a gorilla mislabeling by deleting primate data; Beauty AI scored finalists mostly light‑skinned despite 40,000 Indian and 9,000 African‑American submissions; the secretive COMPAS risk score uses a 137‑question input and affects sentencing; Amazon scraps a biased recruiting prototype. Offers remedies: dataset transparency, audits, Mozilla’s anonymized code review extension, Accenture’s fairness tool, and legal/ethical checks.
AI-written summary · about
chapters (13)
- 0:00Introduction and speaker intro
- 2:35Algorithms and societal biases
- 7:26Bias in word vectors and image recognition
- 10:59Biases in LinkedIn and court systems
- 13:46Recruiting algorithms and their failures
- 15:13Legal and structural barriers to accountability
- 16:36Binary system and historical roots of bias
- 19:10Impact and growth of AI in society
- 21:36Consequences of unchecked algorithmic power
- 23:02Practical steps for ethical AI development
- 25:07Demanding algorithmic accountability
- 26:11Ethics in code education and developer power
- 27:15Call to action and closing remarks
related talks
37mKeynote: Ethical Machine Learning: Creating Fair Models in an Unjust World
38mSocial Implications of Bias in Machine Learning
27mCrouching Harm, Hidden Bias—Addressing issues of fairness....
41mMeasures and Mismeasures of Algorithmic Fairness
47mEthics & AI: Identifying & preventing bias in predictive models
47mEthical Machine Learning
37mThe AI 'Black Box'
11mMachines of Love & Grace – Building Truly Inclusive AI