Obscured by algorithms: how to prove "yes" if computer says "no"
summary
Companies train algorithms by paying humans to answer queries (Facebook M) and harvest that data to build models, while IBM prototypes a virtual audit trail for Watson to show decision paths. Public agencies use predictive analytics to rank fraud suspects (selecting a top 100 for inspectors) and build risk profiles from six legacy plus 10–15 new parameters; automatic enforcement proposals get rejected. Face recognition fails on underrepresented groups; three EU countries already use algorithms to assess tax liabilities.
AI-written summary · about
chapters (14)
- 0:00Start
- 7:26Deterministic Model
- 9:21Deterministic Programming
- 15:11Building Models To Fight Fraud
- 24:14What Is an Algorithm
- 26:09Security Protocols
- 27:22Adaptive Algorithms
- 28:17Scope Creep
- 30:12Power of Algorithms
- 40:06We Should Only Employ Algorithmic Decision-Making if We Ask the Right Questions
- 40:58How Much Power Does the Algorithm Have
- 45:43Contact Concept of Intelligibility
- 47:42Facebook Newsfeed Algorithms
- 50:24Problems of Algorithmic Decision-Making
related talks
01h 12mInvestigating algorithmic systems - Algorithm auditing.
54mEthical Machine Learning
36mCorporate Digital Responsibility
26m(Un)ethical Artificial Intelligence: How to Keep the AI Fair for Everyone
01h 04mData Hemorrhage, Inequality, and You...
01h 16mBEYOND ETHICS WASHING:IMPACT ASSESSMENTS, AUDITS, AND OVERSIGHT FOR AI
19mThe Case for Auditing Algorithms: Disloyalty and Bias M
38mIs My AI Racist? Or How Not To Break The Law With Machine Learning