9 - Ethics Flashcards
What was the main topic of the calendar invitation Kamala received?
Data Ethics
The invitation was titled “Urgent: Data Ethics” following a competitor’s discriminatory actuarial model.
What incident triggered the urgent meeting about data ethics?
A competitor implemented a discriminatory actuarial model leading to higher premium increases for Black patients
This incident was discovered when a hacker accessed the competitor’s data systems.
What key considerations are necessary for ensuring data integrity?
Data must be collected, stored, and used with integrity to minimize harm.
What did the CEO emphasize regarding the use of data?
No use of data that exacerbates or creates undue harm to patients will be condoned.
What does the data science code of ethics relate to?
It includes principles for data scientists to follow to prevent misuse of data.
What are the four key concepts in data ethics mentioned?
- Fairness
- Privacy and security
- Transparency and reproducibility
- Social impact of data
What is a misconception about machine learning models and bias?
That using data to make decisions eliminates human bias.
How can bias in machine learning models occur?
If the underlying data or decisions made when collecting data were biased.
What famous example illustrated the consequences of training data bias?
Google Photos algorithm mislabeling images of people with dark skin as gorillas.
What is one measure of fairness in machine learning models?
Group fairness, which measures whether subjects in each group have equal probabilities of being assigned to a certain outcome class.
What is another measure of model fairness?
Accuracy comparability across groups of interest.
What can be done if bias is found in a model’s performance?
Identify biased data sources, remove them, and retrain the model.
What is adversarial debiasing?
A method to address model bias by incorporating fairness into model development and training.
Why is there a debate about including race in prediction models?
Including race may recapitulate existing racial biases and inequities.
What are some ethical principles proposed for data scientists?
- Respect privacy of data subjects
- Acknowledge limitations of one’s knowledge
- Recognize data represents real people and situations
- Avoid causing societal harm
What does the data science oath emphasize regarding data subjects?
Respect for their privacy and security.
What should be ensured before model deployment?
Conduct fairness analysis to check for biases.
What is the potential risk of using a biased model in healthcare?
It may exacerbate existing inequities in health outcomes.
What approach can be taken to ensure equitable model performance?
Measure model performance across different demographic groups.
What is the significance of testing for fairness in data models?
To ensure that no group is unfairly disadvantaged by the model’s predictions.
What did Kamala compare the machine learning bias issue to?
The lack of diverse images in medical textbooks affecting doctor training.
What is the outcome of deploying a model with known bias?
It can lead to harmful consequences for affected populations.
What is the debate regarding the inclusion of race in prediction models?
The debate centers on whether including race perpetuates racial biases or if it is a strong predictor of outcomes.
What are some best practices for dealing with race in predictive modeling?
- Check model performance across racial groups
- Assess if race is a proxy variable
- Ensure data quality for the race variable
