To build an algorithm you need big data (of what happened in the past) and a definition of success (what to look for). Who and how does one define the definition of success — usually the people in power?
Models are opinions embedded in code. Who’s opinions are biases are injected into the algorithms. Their is a false notion that algorithms are objective and true.
The algorithms will be biased, and understanding that.
Algorithms are a set of secret rules.
Algorithms don’t make things fair. They repeat our past practices, our patterns. They automate the status quo.
“data laundering” occurs where technologists hide ugly truths inside black box algorithms and call them objective. When they are secret, destructive, and important they are “Weapons of Math Destruction”.
There is a lot of money being made in unfairness.
- Data integrity check — For example, whites and blacks smoking pot at the same rate yet blacks are more likely to be arrested. How does that pattern and bias play into other areas of data.
- Audit the definition of success
- Consider the algorithm errors — no algorithm is perfect.
- Consider the long term effects of algorithms — Consider the longer term feedback loops in place.
- Ask for the the model — Ask for the source code and understand the model. Don’t blindly trust the algorithm. “Do not stop questioning just because some of the math in front of you seems inscrutable.”
