Mapping AI and Algorithms in Use in the Public Sector
The White Paper on Data and Artificial Intelligence 2024-2030 notes that the increasing use of artificial intelligence in both public institutions and businesses offers opportunities to create added value for society in almost every area. At the same time, it must be taken into account that the use of algorithmic solutions entails significant challenges. The increasing ability of these solutions to make complex decisions and actions without human participation leads to the need to mitigate risks, ensure control and oversight.
To increase transparency in the public sector, promote inter-agency cooperation and algorithm reuse, we have prepared an algorithmic transparency document, which contains 19 mandatory questions.
Which algorithms should be described using the algorithmic transparency document?
The algorithmic transparency document should filled in for algorithmic solutions whose use may affect the rights, freedoms or obligations of individuals. In particular, this includes the following cases:
-
Algorithms that automate the issuance of an administrative act. For example, in cases where the system automatically issues an administrative act without direct human intervention, such as the automatic issuance of a logging permit or the automatic determination of social benefits.
-
Algorithms used for automatic fact recognition before issuing an administrative act. For example, using machine vision to automatically identify a car's license plate from a speed camera image when issuing a traffic fine.
-
Decision support systems that may have an impact on the final decision that directly affects a person. For example, threat assessment systems.
-
Algorithms used to categorize or profile people. For example, a risk assessment algorithm used in supervision that identifies potential offenders or provides recommendations on who to supervise. Also, systems that perform profiling of a person and that affect his access to public services or other benefits.
-
Algorithms that automate a specific administrative action. For example, automatic transcription of a court hearing or the use of an anonymizer to remove personal data from a document.
-
Systems that are specifically based on artificial intelligence and communicate directly with a person. In particular, Bürokratt and other chatbots, through which a person receives personalized information on the information portal of an institution or access to the services of an institution.
-
Solutions that are specifically based on artificial intelligence and are systematically implemented to automate certain work processes of an official or employee. For example, the use of language models in the preparation of memos or analyses.
If your solution uses more than one algorithm, we ask you to assess whether the algorithms complement each other in achieving the same goal or act for different goals. If the algorithms serve different goals, we ask you to provide information about all the algorithms used in the form, but they can be described within the same form.transparency
What should I do with this document next?
We will compile the completed forms and publish them on https://www.kratid.ee/kasutuslood-kratid along with reviews from other institutions. In the future, we plan to create a separate website for mapping public sector algorithms, similar to the solutions already in place in the Netherlands and the UK.
Which staff member within the organization should complete this document?
The head or member of the unit responsible for developing or using the algorithm is responsible for completing the document. Where appropriate, it may be necessary to involve a developer, data scientist, or lawyer.
Where can I get more information?
If you have any additional questions or need help filling out the document, please write to algoritm@list.justdigi.ee.
Where can I get more information?