Algorithms replacing human judgements
main problem is that algorithms make judgement based on the bias that already exists in society
real world example:
In 2020, the Harvard Gazette reported the use of algorithms by the US government. Courtrooms were using criminal risk assesment algorithms to determine the length of prison sentences. However, criminals from low income and minortiy groups were at risk of having higher scores because historically, they came from areas with higher scores. Consequently, this biased algorithm lead to less favourable sentencing.
Algorithmic bias:
human algorithm developers unknowingly introduce bias into their models
training data sets include biased or unrepresentative of the population
Black box algorithms and the lack of transparency:
it's not always possible to know how the training data was selected
The evolving nature of machine learning makes it difficult to keep up
it's difficult to explain how the algorithm concludes outputs