Avoiding Machine Learning Becoming Pseudoscience in Biomedical Research

Abstract

The use of machine learning harbours the promise of more accurate, unbiased future predictions than human beings on their own can ever be capable of. However, because existing data sets are always utilized, these calculations are extrapolations of the past and serve to reproduce prejudices embedded in the data. In turn, machine learning prediction result raises ethical and moral dilemmas. As mirrors of society, algorithms show the status quo, reinforce errors, and are subject to targeted influences – for good and the bad. This phenomenon makes machine learning viewed as pseudoscience. Besides the limitations, injustices, and oracle-like nature of these technologies, there are also questions about the nature of the opportunities and possibilities they offer. This article aims to discuss whether machine learning in biomedical research falls into pseudoscience based on Popper and Kuhn's perspective and four theories of truth using three study cases. The discussion result explains several conditions that must be fulfilled so that machine learning in biomedical does not fall into pseudoscience

Ähnliche Arbeiten