Can an algorithm be fair in two different ways?
Imagine two groups of job applicants. A screening system could make the same proportion of one kind of error in each group, while its recommendations are more reliable for one group than the other. Both comparisons sound relevant to fairness. Which matters depends on the decision, the starting populations, and the harm caused by different mistakes.
Questions you could explore
- Who should decide which mistakes a hiring system must minimize?
- What changes when the groups have different starting rates of the outcome being predicted?
Start reading
- Kleinberg, Mullainathan and Raghavan / arXivInherent Trade-Offs in the Fair Determination of Risk Scores
The abstract introduces a mathematical conflict among specific fairness conditions, with exceptions that matter.
Ready to try your ideas aloud?
Your topic carries over. Record up to 2½ minutes, depending on your plan.