Algorithmic bias in marketing

Abstract

The advent of Machine Learning and Artificial Intelligence has transformed industries to help make decisions. Algorithms can now decide on loan ratings, job fitness, healthcare diagnosis, criminal record tracking, or whom to marry or date. Such widespread adoption of the technology has grabbed attention on fairness and correctness of these algorithms as they are no longer the novelty from the laboratories but are actually being deployed in places where it affects our day-to-day life. Like any human being, however, these algorithms also carry inherent biases. Sometimes these biases can be overshadowed because the computer is assumed to be fair and lacks emotional connection. However, this has proven to be wrong. Most of these algorithms try to create a way of personalisation so that they can better serve the individual or group. However, such a process of discrimination can introduce some unwanted or, in some special cases, unlawful biases. Examples of biases in algorithms can be a disadvantage or misrepresentation for a certain gender, race, sexual orientation, religion or age. In this report, we will try to investigate different types of algorithmic biases that exist critically analyse them. Later we would move on to the study are general stages of mitigation of these biases. Along with the traditional method, we also propose a novel method that can be used going forward in bias reduction. In the end, we have taken up a case study and tried to analyse how we can work on reducing the algorithmic bias.