Ikibazo gihuza ibibazo · Hagati Urwego
Gutandukanya Ingaruka vs Gutandukanya Ubuvuzi muri Algorithms Ikibazo
Gutandukanya ubuvuzi bushingiye kubushake butandukanye kubikorwa bitagira aho bibogamiye n'ingaruka mbi, kandi usuzume ibimenyetso bitandukanye buri nyigisho isaba.
Ikibazo 1 Bya 8
Ni ibihe bintu byerekana neza uburyo butandukanye bwo kuvurwa?
Reba buri kibazo8 ibibazo
Ibisubizo nibisobanuro bigaragara mugihe ufata ikibazo.
- Ni ibihe bintu byerekana neza uburyo butandukanye bwo kuvurwa?
- A facially neutral threshold has different selection rates
- A model is retrained
- A decision-maker explicitly lowers an applicant’s score because of age
- A dataset has duplicate rows
- Niki gisobanura ingaruka zitandukanye?
- A neutral-looking practice causing adverse effects on a protected group without adequate justification
- Any model using sensitive data
- An explicit discriminatory instruction
- A model with low accuracy
- Ese kureka ubwoko nkibintu byinjira byerekana ingaruka zitandukanye?
- Yego, burigihe
- Only for large employers
- Yes, if the model is proprietary
- No; correlated features and process steps can still create adverse effects
- Ese ZIP-code ifitanye isano nubwoko yonyine irerekana uburyo butandukanye?
- Yes, automatically
- No; correlation alone does not establish intent, though deliberate proxy choice may be evidence
- Only if the model is open source
- No legal theory applies to ZIP code
- Ni iki kigomba kumenyekana mu isesengura-ngaruka zitandukanye?
- The specific practice causing the observed impact
- Only the model vendor
- The most accurate benchmark
- Every feature used by a competitor
- Which defense is relevant after a disparate-impact showing under Title VII?
- The employer says the tool was neutral
- The vendor has a patent
- The practice is job-related and consistent with business necessity
- The model uses enough parameters
- What does evidence of a less discriminatory alternative do?
- It converts impact into intent
- It may remain relevant after an employer shows business necessity
- It proves the model is accurate
- It removes the need to identify a practice
- Which conclusion is supported when an audit finds a statistical gap?
- It proves unlawful discrimination in every jurisdiction
- It proves discriminatory intent
- It requires immediate deletion of all records
- It informs review but does not itself resolve legal liability