Main content
Bias in machine learning
Problem
HireView is a startup that claims to speed up the process of interviewing job candidates. A candidate submits a video answering interview questions and HireView analyzes the video with a machine learning algorithm.
The algorithm scores the candidate on various aspects of their personality, such as "willingness to learn" and "personal stability". HireView engineers trained the algorithm on a data set of past videos that were scored by employers and psychologists.
In a test of the algorithm, HireView engineers discover that the algorithm always gives lower scores to people who speak more slowly.
What is the most likely explanation for the algorithm's bias?