Benchmark of algorithm approval measures

Artificial intelligence algorithms have many benefits, but introduce risks of increased inequality, unreliability, opacity and reduced privacy. The public administrations are guarantee-oriented and this leads them to prioritise the establishment of ex-ante measures, which prevent the implementation of harmful algorithms. This article presents a range of tools used by various governments around the world before putting their algorithms into operation.

Instruments: these are mechanisms that address a specific issue. The following are presented:

•  Regulation of automated administrative action in Spain
•  Global and specific awareness-raising
•  Bias detection and mitigation
•  Transparency: open source, explainability (local and global techniques) and traceability (algorithm sheets)

Working frameworks: these are more holistic approaches, which try to cover all relevant aspects of implementing an algorithmic system:

•  Recommendations for international governments (European Commission and Toronto Declaration)
•  Algorithmic Impact Assessment in Canada
•  UK Data Ethics Framework
•  New Zealand Algorithm Charter
•  Algorithmic Accountability Act in the USA