Machine Bias

Some recent applications of Artificial Intelligence, particularly machine learning, have been strongly criticised in general media and professional literature. Applications in domains of justice, employment and banking are often mentioned in this respect. The main critic is that these applications are biased with respect to so called protected attributes, such as race, gender and age. The most notorious example is the system COMPAS which is still in use in the American justice system despite severe criticism. The aim of our paper is to analyse the trends of discussion about bias in machine learning algorithms using the COMPAS as an example. The main problem we observed is that even in the field of AI, there is no generally agreed upon definition of bias which would enable operational use in preventing bias. Our conclusions are that (1) improved general education concerning AI is needed to enable better understanding of AI methods in everyday applications, and (2) better technical methods must be developed for reliably implementing generally accepted societal values such as equality and fairness in AI systems.