IMPLEMENTATION OF ALGORITHMIC SYSTEMS IN ORDER TO IMPROVE THE SUPPLY OF SERVICES TO CITIZENS: THE LABORATORY DISCLOSES ITS INVENTORY OF THE MOST COMMON BEST PRACTICES
By Emilien Miron
A growing number of organizations offering services to citizens, including public authorities, international organizations as well as multinational corporations, decide to implement algorithmic systems in order to improve the way they satisfy the needs of population.
In the ambit of the project with ULB of ACT, which purpose is to carry out an inventory of the policies and best practices presently applied when it comes to automation through artificial intelligence, Laetitia Dimanche and Jehanne Dussert have consequently analyzed 34 key projects in order to list the best practices that are most commonly observed in this field.
The projects under study have been developed in several distinct geographical regions (including in particular North America, Europe and China) and are aimed at achieving goals as varied as, for example:
- strengthening transportation security for Air Canada or Tesla;
- resolving humanitarian or political crises for Google;
- paying through facial recognition for Tencent;
- in the legal field, assessing the risk of recidivism of an offender who has applied for release on parole in the United States or predicting the decision to be rendered by a tax judge in Canada.
The outcome of the examination, which is intended to be used primarily by public administrations and has been conceived as a starting point for deeper scientific research, highlights 13 unanimously-approved best practices:
- carrying out audits;
- applying for certification proceedings or obtaining quality labels;
- setting up ethics committees;
- informing the public on a regular basis and through multiple channels of communication;
- elaborating internal policies;
- realizing polls;
- creating research centres;
- implementing testing phases;
- developing partnerships;
- organizing collaborative events;
- drafting guides;
- for public decision-makers, creating a control authority;
- for public decision-makers, establishing a legal framework regulating the use and the development of algorithmic systems.
In addition to this, the report emphasizes the fact that all those various practices are not intended to be applied at the same time during the implementation of a given algorithmic system and that, even more importantly, each of them must imperatively be put into application at the right moment so as not to lose its interest.
Moreover, to be fully effective, some best practices must be implemented on a continuous basis or at multiple stages throughout the process, as it is the case with audits or information of the public.
The report can be downloaded in PDF format by clicking on the following link (in french):
Have a good reading!
This content has been updated on 10/18/2019 at 11 h 27 min.