{"id":10581,"date":"2019-06-13T00:52:54","date_gmt":"2019-06-12T22:52:54","guid":{"rendered":"https:\/\/www.etalab.gouv.fr\/?p=10581"},"modified":"2019-06-13T09:07:24","modified_gmt":"2019-06-13T07:07:24","slug":"how-etalab-is-working-towards-public-sector-algorithms-accountability-a-working-paper-for-rightscon-2019","status":"publish","type":"post","link":"https:\/\/preprod.etalab.gouv.fr\/how-etalab-is-working-towards-public-sector-algorithms-accountability-a-working-paper-for-rightscon-2019","title":{"rendered":"How Etalab is working towards public sector algorithms accountability: a working paper for RightsCon 2019"},"content":{"rendered":"\n

RightsCon is the world’s leading event on human rights in the digital age. The 2019 edition<\/a>, taking place this week in Tunis, brings together 2500 participants from 130 countries (NGOs, governments, companies). Etalab has been selected to lead a session<\/a> about algorithmic accountability. <\/p>\n\n\n\n

\u00ab\u00a0With great power comes great responsibility\u00a0\u00bb<\/h4>\n\n\n\n

On this occasion, Etalab is publishing a working paper \u00ab\u00a0With great power comes great responsibility: keeping public sector algorithms accountable<\/a><\/strong><\/em>\u00a0\u00bb to present our work on algorithmic accountability and to engage the discussion with international actors (NGOs, governments and companies). Here is a short summary of our findings.<\/p>\n\n\n\n

From automation of human tasks to machine learning: a short history of public sector algorithms<\/strong><\/p>\n\n\n\n

Algorithms are nothing new for the public service. In the 1970\u2019s, the emergence of the computing industry allowed the French authorities to automate some of its administrative processes, starting with the large-scale processes, like tax and social benefits calculations. The second generation started with the use of matching algorithms, mainly for human resources management. The Ministry of Education developed a system to manage their workforce, that is, educators and teachers applying for a new position in a different school or region. Since the early 2000\u2019s, matching algorithms have been used for student allocation such as Affelnet, Admission Post-Bac, and now Parcoursup. The third generation of public sector algorithms is linked to the emergence of machine learning (ML) algorithms which represent a major shift for public policy. In the first two periods, computers were used to apply a set of rules that were sometimes complex, but always predefined. Instead, ML algorithms derive rules from observations and learning from large datasets. These systems are used for tax fraud detection, prediction of companies likely to hire (La Bonne Boite<\/a>) or that present a risk of going bankrupt (Signaux Faibles<\/a>). <\/p>\n\n\n\n

\"\"
[Extract from the working paper]<\/figcaption><\/figure>\n\n\n\n

Specific challenges for public sector algorithms<\/strong><\/p>\n\n\n\n

Private and public sectors share a series of challenges (opacity, loss of autonomy, low level of explainability, risk of bias, …) but the latter are distinguable in three ways worth noting. <\/p>\n\n\n\n

First, they must be used in the public interest<\/strong> and not a particular or private<\/em> interest. One can question the impact of YouTube\u2019s recommendation algorithm on the diffusion of harmful content. That said, it is generally accepted that, as a private company, YouTube is pursuing its own interest and not the public interest. Second, in many cases, public sector algorithms tend to implement legal rules<\/strong>. Tax calculations follow a list of rules defined by the general tax law and adopted by the Parliament. As such, some PSA are the last link from the political will to tangible effects on individuals. <\/p>\n\n\n\n

Finally, public sector algorithms tend to be unavoidable: citizens do not have the option to use a different algorithm and sometimes are not presented with the choice to opt out of using an algorithm at all. For example, a French patient in need of a heart transplant has no choice but to accept the rules and their algorithmic implementation by the Biomedicine agency. Similarly, the only way to get access to most of the higher education institutions is to go through Parcoursup, the allocation system for students, which relies on an algorithm.
In 1789, the Declaration of the Rights of Man and of the Citizen (article XV) stated that \u201cSociety has the right to require of every public agent an account of their administration<\/em>\u201d. This principle, enacted well before the invention of computers, still stands: if administrations use algorithms as tools of administration and government, then they \u2013 and their algorithms \u2013 should be held accountable. <\/p>\n\n\n\n

How Etalab is working towards sector algorithms accountability<\/strong><\/h4>\n\n\n\n

With the introduction of a new legal framework for algorithmic accountability and transparency (introduced by the Digital Republic Bill), public agencies need to be accompanied on making existing and future algorithms compliant with the new obligations. This also gives citizens access to new rights, such as an extended right to information. <\/p>\n\n\n\n

Through the National Action Plan (2018-2020) for the Open Government Partnership, France has committed to reinforcing \u201cthe transparency of public sector algorithms and source codes\u201d. Etalab, as the government\u2019s task force for open data and data policy, oversees the work on this commitment <\/strong>which lies at the crossroads between open data, open source, and open government issues.  <\/strong><\/p>\n\n\n\n

Our approach is multifaceted:<\/p>\n\n\n\n