Ikibazo gihuza ibibazo · Biragoye Urwego

Ibyemezo by'inguzanyo ya AI hamwe nibimenyesha ibikorwa bibi

Ibizamini Amabwiriza B yihariye-impamvu yihariye, CFPB izenguruka kuri moderi igoye nuburyo bwikitegererezo, hamwe nibibazo bya SHAP ishingiye kumpamvu.

Ikibazo 1 Bya 8

Dukurikije ibisobanuro B Amabwiriza B, kuki kubwira uwasabye guhakana ko yananiwe kubahiriza ibipimo byimbere bidahagije?

Reba buri kibazo8 ibibazo

Ibisubizo nibisobanuro bigaragara mugihe ufata ikibazo.

  1. Dukurikije ibisobanuro B Amabwiriza B, kuki kubwira uwasabye guhakana ko yananiwe kubahiriza ibipimo byimbere bidahagije?
    • It reveals confidential model details
    • Reasons must be specific and relate to the factors actually considered
    • It must be stated in at least two languages
    • Internal standards can only be disclosed by the FCRA notice
  2. Ni iki CFPB Circular 2022-03 yavuze ku ngero zinguzanyo zinguzanyo?
    • Complex models are banned for credit decisions
    • Creditors using complex models may skip adverse action notices
    • Creditors cannot avoid the specific-reasons requirement because their technology is too complex or opaque
    • Only banks, not fintechs, must explain complex models
  3. Icyitegererezo cy'abatanga inguzanyo gishingiye kuri banki-konte y'amafaranga-yerekana amakuru, kandi impamvu y'icyitegererezo yegeranye ifitanye isano gusa. Uruziga 2023-03 rwerekana iki?
    • The lender may use the nearest sample form reason
    • The lender must provide reasons that accurately describe the real factors, even if not on the sample form
    • The lender must stop using cash-flow data
    • The lender only needs to disclose the credit score
  4. Ibisobanuro bya Regulation B bivuga iki kubyerekeye kwerekana impamvu nyinshi?
    • At least ten reasons must always be listed
    • Disclosing more than four reasons is not likely to be helpful
    • Only one reason may ever be given
    • Reasons must be listed in alphabetical order
  5. Nigute amanota gakondo asanzwe atanga impamvu mbi yibikorwa?
    • By comparing each attribute's points with the maximum possible points
    • By asking a loan officer to write reasons by hand
    • By using the applicant's income as the only reason
    • By ranking features alphabetically
  6. Why does the choice of reference point matter when using SHAP values for adverse action reasons?
    • SHAP values are identical regardless of reference
    • Attributions are measured relative to a reference, so explaining against the average applicant can differ from explaining against one near the approval cutoff
    • The reference point must be the applicant with the highest score
    • References only matter for mortgage loans
  7. Why do lenders group correlated features into reason families before ranking them?
    • To reduce the number of features the model uses
    • So attribution for one underlying factor is not diluted across several related variables
    • Because Regulation B lists required families
    • To hide which data sources the model uses
  8. Which statement reflects the guide's view of explainability libraries?
    • Running any explainability library automatically produces compliant reasons
    • Attributions can be unstable and depend on choices, so reasons need validation for accuracy and clarity
    • Explainability tools are only needed for denied mortgage applications
    • Explainability tools also prove the model is fair