Algorithmic Biases

Concept of Algorithmic Biases

How do algorithms work?

Algorithms is a part of the concept used in machine learning, where the computer scientist exposes the computer to a bunch of data, any kind, enough for the computer to make judgments or predictions about the information based on the patterns it notices. Algorithms are the information associated with the raw data. For example, if you wanted a computer to be able to identify when an object is a book, you would go through a set of metrics with the computer to identify whether the objects presented are books and when they are not. After testing and refining, the systems should know what indicates a book and be able to predict in the future when an object is a book. This technique, though represented on a computer, is very human-centric - people write the algorithms, people choose the data used by algorithms, and people decide how to apply the results of the algorithms.

How do algorithms work? diagram

How does algorithmic bias occur?

Since algorithms are built by humans, and most tests are done on things more subjective than identifying a book as an object, algorithmic bias occurs by bias in data sets, people designing AI models, as well as interpretation of those results. Machine learning left unchecked can perpetuate historical inequities. Without diverse teams and rigorous testing, it can be too easy for people to let subtle, unconscious biases enter, which AI then automates and perpetuates. Scientists must be aware of the historical inequities in their dataset in order to compensate for it. In order for scientists to be aware, a diverse group of people must be working on these systems. Unfortunately, machine learning tends to be dominated by men and overrepresent white people. Though representation may occur, that does not mean that biases are solved. It is nearly impossible to build a “fair” system on social questions that do not have an objective answer. Nonetheless, If these biases are not identified, they can lead to poor business decisions, which is outlined in the next page.