Author: Vivek Katial
Publisher: Multitudes
Publication Year: N/A
Summary: The following article starts by introducing the profound effect that algorithms and, in particular, decisions from algorithms have on our life today. The author goes on to define algorithmic bias as “the ability of algorithms to systematically and repeatedly produce outcomes that benefit one particular group over another.” The article then goes on to state that most commonly these biases come up due to a lack of diversity within the dataset that the analyst might be using. Then, the author goes through the 5 steps of the machine learning life cycle and how biases can arrive at all 5 points. Most notably, in the first phase, when the data is collected and prepared. Biases can come up during this phase due to the data collected not being completely representative and not fully reflecting the whole world. Finally, the article states, that, to mitigate these biases, it is essential to track the quality of input data. The threat is that without fully monitoring the quality of the data, some of these biases can start to take form throughout the model building process.