Freitag, 29.03.2024 03:00 Uhr

Risks of the use of AI in public administration

Verantwortlicher Autor: Carlo Marino Rome, 27.07.2021, 15:43 Uhr
Nachricht/Bericht: +++ Politik +++ Bericht 3362x gelesen

Rome [ENA] There are opportunities and risks of the possible use of AI in the social insurance and in public administration. The European Commission has specified its position in a legal framework for AI applications, not least with regard to applications in the public administration context. In April, it presented its proposal for an "Artificial Intelligence Act", COM (2021) 206 final, and once again invited all interested

parties to submit their comments. German Social Insurance took the opportunity to submit its Opinion on 14 July 2021. Even if AI is not currently being used in the social security sector - at least nowhere with a high risk potential - this could change in the future, e.g. through machine learning processes for supporting individual and case-specific decisions. German Social Insurance is aware of the responsibility that this entails. We, therefore, welcome clarific­tion of the ethical and liability issues that also surround the use of AI. However, there are still reservations about defining AI too broadly. Only those technologies that actually replace assessments and decisions that were previously genuinely reserved for humans should be

included in it. The opinion of the German Social Insurance is that it welcomes the draft Artificial Intelligence Act (AI Act). It has far-reaching implications for developing and using AI, not least in the public administration sector. German Social Insurance sees possible membership, contribution and benefit law implications and is aware of its responsibility in handling AI. Therefore, it welcomes clarification of issues such as the ethical and liability laws. What will also have to be clarified during further negotiations on the draft regulation and further on, as to what level of detail and depth of decision-making will European action be required.

It should be noted that the draft only uses a very broad definition of AI. It should also cover several algorithms that were developed and used in the past. A defini-tion that focuses solely on the existence of "autonomous conduct” would be far too narrow. This is cutting-edge automation that has still not yet been attained by to-day's technical systems. On the other hand, the concepts mentioned in Annex I "AI techniques and concepts referred to in Article 3, Para. 1" under sections (b) and (c) may well go too far. Therefore, AI should only be referred to when algorithms developed using one or more of the techniques listed in Annex I are used to make decisions, predictions or recommendations that are genuinely human.

Against this background, it cannot be ruled out that individual programs / applications now being used by the social insurance institutions already meet the AI characteristics. Nevertheless, no use of AI is currently seen as "high risk" in the social security institutions' own applications. However, this could change in the future if machine learning methods will be used to support both case and person-based decisions. The creation of transparency, traceability, non-discrimination and ultimate human responsibility when using machine-supported procedures will be required here. Control and monitoring mechanisms will be useful against this background.

Social security institutions can strengthen their trustworthiness by establishing their own voluntary codes of conduct or by joining the codes of conduct of representative associations when using AI applications that do not pose a high risk. Reference should also be made to the opinion from German Social Insurance about public consultation on the EC’s White Paper "Re: AI - A European approach to excellence and trust".

Für den Artikel ist der Verfasser verantwortlich, dem auch das Urheberrecht obliegt. Redaktionelle Inhalte von European-News-Agency können auf anderen Webseiten zitiert werden, wenn das Zitat maximal 5% des Gesamt-Textes ausmacht, als solches gekennzeichnet ist und die Quelle benannt (verlinkt) wird.
Zurück zur Übersicht
Photos und Events Photos und Events Photos und Events
Info.