Issue Algorithmic bias, privacy risks and weak democratic control of AI

Summary

The report finds that AI applications increasingly rely on learning algorithms, which can reinforce prejudice. Bias in algorithms is particularly damaging because it affects entire groups in the same way, locks in historical prejudice and is hard to detect. Dutch citizens generally hold no positive expectations of algorithms, and many fear discrimination resulting from AI applications. The rapid spread of AI brings a range of privacy risks, and much experimentation takes place even where people are vulnerable, which can result in data misuse and unexpected incidents. In the report's view, the process of democratic steering and oversight of AI systems has not yet been adequately put in place. New techniques such as facial and emotion recognition pose serious risks to citizens' privacy and can lead to bias and unjust treatment.