AI Bias — Bias AI

Kecenderungan sistem AI untuk memperkuat prasangka dari data training, menghasilkan diskriminasi pada gender, ras, usia, dll. Contoh: COMPAS, Amazon Rekrutmen.

AI bias: training data bias, algorithmic bias, deployment bias. Solusi: diverse data, fairness metrics (demographic parity, equalized odds), auditing, diverse teams.

Also known as: bias AI
Print

AI Bias

Definisi

AI bias adalah ketika sistem AI menghasilkan output yang tidak adil atau mendiskriminasi kelompok tertentu — biasanya karena bias dalam data training atau desain algoritma.

Contoh

  • COMPAS (ProPublica 2016) — software prediksi recidivism, bias terhadap orang kulit hitam
  • Amazon Rekrutmen (2018) — gender bias, sistem preferensi laki-laki
  • Google Photos (2015) — salah label orang kulit hitam sebagai gorila
  • GPT-3 (2020) — bias gender, ras, dan agama

Mitigasi

  • Diverse training data
  • Fairness metrics — demographic parity, equalized odds, counterfactual fairness
  • Algorithmic auditing
  • Diverse teams saat desain
  • Regulation — EU AI Act, EEOC guidance

Connected to

Not yet written

The following pages are referenced but don't exist yet — they'd make good future additions.

  • /concepts/ai-safety

References

  1. Wikipedia

Type at least 2 characters to search.

Press to navigate, to open, esc to close.