Help! There Are Humans in My Data!
summary
Fieldwork shows concrete costs: 350 hours of shadowing police, police systems with 685 incident codes, downstream use in predictive policing, and the Titanic dataset surviving as a third reconstructed version. One practical fix is UI change — Amsterdam replaces nested menus with a free-text box plus a simple text classifier (2019) and reduces mislabels. Data teams run iterative annotation and reclassification, but still encounter semantic shifts (e g “psychotic” vs “in a psychosis”) that break downstream decisions.
AI-written summary · about
chapters (7)
- 0:00Human factors in data
- 5:22Why humans create messy data
- 10:40Incident codes in police reports
- 17:03Ambiguous location categories
- 23:20Human limits in data entry
- 26:31Three datification problems
- 29:42Solutions for human-made data
related talks
49mKeep it Clean: Why Bad Data Ruins Projects and How to Fix it
44mKeep it Clean: Why Bad Data Ruins Projects and How to Fix it
31mEthics in technology: History, practice, and mitigation
31mInclusive Storytelling With Data
17mYour AI Deserves Better Data
11mData-centric AI: A new paradigm
47mKeynote: Using Data Science for Social Good: Examples, Opportunities, and Challenges
52mMaking Better Decisions with Data