[SECURITY] Biased data – biased algorithms
Machine learning algorithms are just well-trained machines that base their conclusions on data.
However, those conclusions are only as good as the data used to train the machines. If the data is
biased, then the algorithms will be biased. This can be particularly problematic in predictive policing
algorithms, which rely on archival data to predict where future crimes will occur.
In 2020, researchers at the University of Cambridge found that a predictive policing algorithm used by
London police was biased against members of certain ethnic and racial minorities. It has been shown
that the algorithm is twice as likely to wrongly label individuals from certain minority groups as potential
criminals than it is the case with members of the majority population. This study highlights the necessity of testing algorithms for bias, as well as the need for transparency and accountability when using big