Cuestionario vinculado a una guía · Medio Nivel

Cuestionario sobre auditorías de sesgo de reconocimiento facial y matices de género

Prueba el método de auditoría interseccional Gender Shades, la diferencia entre clasificación e identificación de género y los hallazgos demográficos del NIST.

Rutas de guía relacionadasAuditoría de sesgo de reconocimiento facial de tonos de género
Pregunta 1 de 8

Which companies' face analysis services did the original Gender Shades study audit?

Vista previa de cada pregunta8 preguntas

Las respuestas y explicaciones aparecen a medida que realiza el cuestionario.

  1. Which companies' face analysis services did the original Gender Shades study audit?
    • Google, Apple and Meta
    • Amazon, Kairos and Clearview
    • IBM, Microsoft and Face++
    • NEC, Idemia and Cognitec
  2. What was the Pilot Parliaments Benchmark made of?
    • Images of members of parliament from three African and three European countries
    • Celebrity photos scraped from the web
    • Mugshots from US police departments
    • Synthetic faces generated by a neural network
  3. Which scale did the researchers use to label skin type?
    • The Monk Skin Tone scale
    • The Fitzpatrick scale
    • The Munsell color system
    • A self-reported race category
  4. What task did Gender Shades actually measure?
    • One-to-many face identification against police databases
    • Emotion recognition
    • Face detection in video
    • Gender classification
  5. Which subgroup had the highest error rates in Gender Shades?
    • Darker-skinned women
    • Lighter-skinned women
    • Darker-skinned men
    • Lighter-skinned men
  6. What did the 2019 "Actionable Auditing" follow-up find?
    • No vendor had changed its system
    • All disparities had been eliminated industry-wide
    • Audited companies had narrowed their gaps, while unaudited vendors showed similar disparities
    • The original study's data had been fabricated
  7. Why can a single similarity threshold produce unequal outcomes across groups?
    • Thresholds only apply to men
    • The same threshold can yield very different false match rates for different groups
    • Thresholds are chosen at random for each query
    • Thresholds affect speed but not errors
  8. Which vendor response happened in June 2020?
    • Microsoft began selling facial recognition to all US police
    • Face++ published its training data
    • NIST banned facial recognition
    • IBM said it would leave the general-purpose facial recognition business