Author: Billie Trinder
Publisher: Seven Pillars Institute
Publication Year: 2019
Summary: The following article argues that external enforcement of ethical principles in data analytics is unfeasible due to the pace of technological advancement and bureaucratic lag. Rather, the author argues that organizations must self-regulate, and need a well-defined set of ethical guidelines in order to do so effectively. The author argues from a position of self-interest, and ethics, which he implies are not mutually exclusive. Failing to operate ethically puts companies in a position of vulnerability, as they may face discrimination lawsuits and loss of consumer trust. In order to remedy the problem of model opacity, the author argues that analysts must employ “subject-based searches.” Using this type of approach, analysts direct the machine learning model in a certain direction to explore associations. Complications arise from the largely uninterpretable nature of many machine learning models, which create concerns over a lack of transparency. For example, current “pattern-based searches” allow algorithms to seek out patterns with no direction. Pattern-based searches are difficult to understand and interpret, shielding models and their results from justified and necessary scrutiny. Shifting away from this paradigm places accountability on analysts and allows for a more conscious identification of where things go wrong in machine learning.