AI in Government Benefits Administration Quiz

Answers and explanations appear as you take the quiz.

1. What was the core calculation flaw in Australia's Robodebt scheme?
   • It used facial recognition to identify claimants
   • It averaged annual tax-office income evenly across fortnights
   • It relied on a neural network trained on social media
   • It randomly selected recipients for debts
2. Which statement about Robodebt's technology is accurate?
   • It was an advanced deep-learning system
   • It was a chatbot that negotiated debts
   • It used satellite data to verify addresses
   • It was a flawed formula and data-matching process, not sophisticated AI
3. Which factor in the Dutch tax authority's risk model drew particular criticism?
   • Nationality
   • Preferred language of correspondence
   • Type of bank account used
   • Age of the youngest child
4. When did the Dutch cabinet resign over the childcare benefits scandal?
   • March 2015
   • June 2018
   • January 2021
   • September 2024
5. What happened to the Dutch SyRI system in 2020?
   • It was expanded nationwide
   • A court halted it on privacy and human-rights grounds
   • It was sold to another country
   • It was converted into a chatbot
6. With 1 percent fraud, 90 percent detection and a 5 percent false-flag rate, what is true of flagged people?
   • Almost all are fraudsters
   • Exactly half are fraudsters
   • None are innocent
   • Most flagged people are innocent
7. Why can labels from past investigations bias a fraud model?
   • Investigations are always random
   • Labels are encrypted
   • The model learns where investigators looked, not where fraud actually is
   • Labels are too few to train on
8. Which design is recommended for risk scores in benefits systems?
   • A score triggers human review, never an automatic debt or suspension
   • High scores automatically cancel payments
   • Scores are kept secret from caseworkers
   • Scores replace the appeals process