AML Professional Standards & Ethics 2 — Questions and Answers
Question 1: A machine learning engineer discovers their deployed fraud detection model has a significantly higher false positive rate for customers in low-income zip codes. What is the MOST ethically appropriate immediate action?
- Document the finding and schedule a review for next quarter
- Pause the model and notify stakeholders while investigating the root cause (Correct answer)
- Retrain the model with more data from those zip codes without informing leadership
- Accept the disparity as an inevitable artifact of the training data
Correct answer: Pause the model and notify stakeholders while investigating the root cause
Discovering a discriminatory disparity requires immediate escalation and pausing the harmful system while a proper investigation is conducted.
Question 2: Under the IEEE Ethically Aligned Design framework, which principle holds that AI systems must remain under human authority and not be designed to override human decisions?
- Transparency
- Accountability
- Human override / Control (Correct answer)
- Data agency
Correct answer: Human override / Control
Human override and control is a core IEEE principle ensuring humans can inspect, correct, and shut down AI systems at any time.
Question 3: A data scientist is asked to use customer purchase data collected for loyalty rewards to train a marketing churn model. Which ethical concern is MOST directly raised?
- Algorithmic bias
- Purpose limitation / contextual integrity (Correct answer)
- Model explainability
- Intellectual property infringement
Correct answer: Purpose limitation / contextual integrity
Purpose limitation holds that data should only be used for the purpose for which it was originally collected and consented to.
Question 4: When publishing an ML research paper, a researcher finds that a competing team's unpublished work led them to the same solution. What is the ethically required action?
- Publish independently since the competing work is unpublished
- Cite the competing team's work if it influenced the research (Correct answer)
- Delay publication until the competing team publishes first
- Contact the journal to claim priority based on submission date
Correct answer: Cite the competing team's work if it influenced the research
Academic integrity requires attributing any work that influenced your research, regardless of its publication status.
Question 5: Which of the following BEST describes 'dual-use' risk in the context of machine learning research?
- Using two different ML frameworks for the same task
- Research that can be applied for both beneficial and harmful purposes (Correct answer)
- Sharing a model across two separate business units
- Training a model on both labeled and unlabeled data simultaneously
Correct answer: Research that can be applied for both beneficial and harmful purposes
Dual-use risk refers to technologies or research that can be intentionally or unintentionally repurposed to cause harm.
Question 6: A company deploys an NLP hiring tool and later discovers it was trained on historical data reflecting past discriminatory hiring practices. Which legal framework in the US is MOST relevant?
- GDPR
- Title VII of the Civil Rights Act (Correct answer)
- Computer Fraud and Abuse Act
- Digital Millennium Copyright Act
Correct answer: Title VII of the Civil Rights Act
Title VII prohibits employment discrimination based on protected characteristics, which applies to AI tools used in hiring decisions.
Question 7: An ML practitioner is hired to consult on a project but lacks expertise in the specific domain. According to professional ethics standards, the MOST appropriate response is to:
- Accept the project and learn on the job to avoid revenue loss
- Decline or disclose limitations and bring in a domain expert (Correct answer)
- Deliver results without mentioning the expertise gap
- Charge a lower fee to compensate for reduced quality
Correct answer: Decline or disclose limitations and bring in a domain expert
Professional ethics require practitioners to practice only within their competence and to be transparent about limitations.
AML Professional Standards & Ethics 2
A machine learning engineer discovers their deployed fraud detection model has a significantly higher false positive rate for customers in low-income zip codes.
What is the MOST ethically appropriate immediate action?