Algorithmic Bias, Discover How It's Affecting You!
Guiding Questions
Are algorithms tested before public release?
Yes, however, they are not specifically tested for bias or discrimination. Programmers create algorithms to perform a specific function, many times using standard testing data to get the desired output. Because it is hard to understand how a computer decides to perform a task, algorithmic bias isn't easy to find.
How can an algorithm perpetuate biases?
If an algorithm directly or indirectly provides an advantage or disadvantage to a specific demographic, the algorithm is unfair. This happens when the output of a widely used algorithm introduces bias or reinforces stereotypes.
How are algorithms tested for bias?
It’s hard to detect what an algorithm is doing on a global scale from a single computer so instead, researchers use tools for accountability and algorithmic audits to identify and resolve ethical issues that arise on the internet.
Lesson 04
"Black Box"
Algorithms bring a trade-off between performance and understanding, the more complex a task performed by an algorithm, the harder it is to understand what the algorithm is doing. Because no one can 100% confirm what an algorithm is doing, algorithms are considered "black boxes," which makes identifying bias and proving it challenging.
Algorithmic Bias
Algorithmic bias happens when a computer system produces repeatable errors that create unfair outcomes, such as privileging one group of users over other users. This bias can happen due to several different factors, including the way the algorithms is designed, unintended use of the algorithm, and lack of diversity in the data run through these algorithms, among other things. More often than not, our own societal biases and the biases of the people creating the algorithms leak into the system, causing algorithms to reproduce and reify biases that harm certain groups of users.