Algorithmic Bias: Data as a Weapon with Ayodele Odubela
summary
Built a drone weapon‑recognition system using millimeter wave sensors that auto‑positions to suspected firearms and includes stance detection but stakeholders demand officer/suspect classification, exposing tradeoffs in false‑positive thresholds and real‑world harm. Shows facial‑recognition datasets skewed to white men, cites Northpointe recidivism and Brooklyn building biometric rollouts, and calls out emotion detection is snake oil. Recommends model cards and dataset nutrition labels, routine algorithmic impact assessments, user appeals and financial remedies, and retroactive audits of legacy models.
AI-written summary · about
chapters (11)
- 0:00DATA AS A WEAPON
- 4:07How are you privileged?
- 7:25Historical Context
- 15:41It's Proprietary
- 16:30Deploying to less powerful Communities
- 17:07Lobbying against regulations
- 17:46Overhype & Not Disclosing Limitations
- 19:02No Science, Just Vibes
- 19:50Start with Principles
- 21:29Rely on- existing research
- 22:08Standardize Documentation
related talks
47mEthics & AI: Identifying & preventing bias in predictive models
27mCrouching Harm, Hidden Bias—Addressing issues of fairness....
47mEthical Machine Learning
38mSocial Implications of Bias in Machine Learning
45mThe Myth of Neutrality: How AI is widening social divides
29mResponsible AI Challenges and Recommendations
31mDoes This Algorithm Have a Soul?
40mEthical AI: Addressing Bias & Algorithmic Fairness