Organisational measures for governments to use algorithms ethically

Public administrations have many challenges to address that could benefit from algorithms, but at the same time we are expected to avoid any side effects. For this reason, in 2020, the Generalitat de Catalunya generated a proposal for a framework for algorithm ethics, consisting of a set of principles and organisational measures for the responsible implementation of algorithms.

Principles: we should not do everything that is technically possible. We must use mature technologies for the most sensitive applications, while at the same time learning about the most disruptive technologies in innocuous areas. And we should not always apply all the same controls, but they should be graduated according to risk.

Ex-ante measures: during the design process, a socio-technical analysis is required, which understands the problem to be solved and the groups that could be affected by the algorithm, and the data protection analysis that we are already carrying out. In terms of procurement, it should follow technical specifications that place responsibility to provide the necessary data and analysis to assess biases and improve the performance of the algorithm on the supplier. This requires a cross-departmental interdisciplinary team to assist the departments in carrying out the socio-technical analysis and drafting the specifications, and intuitive tools to perform basic quality control of the analyses provided by the suppliers.

Ex-post measures: if the algorithm is limited to recommending actions to public employees responsible for the final decision, they will have to be trained to make up for the biases of the algorithm; however, if the algorithm acts directly, it will need internal monitoring. It is also necessary to facilitate external control, with a fact sheet for each algorithm on the transparency portal, explaining its design and indicating the body before which it can be challenged, and explainability tools that allow the algorithm's decisions to be interpreted.

Training and awareness-raising: in order for everyone to play their part, specialised training will be needed for the professionals who will have to carry out the analyses; awareness-raising and basic training for public employees involved in the operation of algorithmic systems, for those responsible for deciding on their implementation, and for those involved in their design, and also awareness-raising for citizens, both users of a particular service and citizens in general, so that they can exercise their right to understand and judge.