In Schleswig-Holstein, ChatGPT is to be used in administration in the future. Digitization Minister Dirk Schrödter (CDU) gives the green light for the use of generative text AI in the state administration – according to the regulations of the IT Use Act (ITEG): “The use of language models such as ChatGPT will revolutionize administrative processes,” he says confidently . “The future of administration is automated, algorithmized, cloudified and data-driven.”
The Schleswig-Holstein state data protection officer Marit Hansen does not veto it, but says: “This can only work if the administration is built on a constitutional, legally compliant, fair and just foundation.” It is of course “okay” to see the chances of using ChatGPT for politics and administration, but “managing the risk is a must”. Even large IT companies are currently not adopting GPT models because the risk and compliance teams have not yet given their approval.
speeches and summaries
Dirk Schrödter explained to heise online that the possibilities would be explored and that the company wanted to play a “nationwide pioneering role”. First of all, they want to use ChatGPT “for preparatory research for speeches, presentations of facts or notes”. It is like search engines “another form of research options”. Behind this is the hope of being able to organize the administrative work more efficiently – “also in view of the forthcoming shortage of skilled workers”.
Especially where it comes to summarizing large amounts of information, a “significant acceleration in the creation of overview documents” is possible. Schrödter wants to ensure that the last content check is carried out by the experts: “It remains the case that the content generated by systems such as ChatGPT, like any other external source, must be checked by the experts of the state administration before it is included in the further work of the state administration can be incorporated.”
ChatGPT will not generate any jokes or poems for Digitization Minister Schrödter to garnish his speeches. Schrödter assured heise online that he “wouldn’t write any speeches for me either without checking them politically or in terms of content before I deliver them”. He sees the core competence of ChatGPT in the fact that there can be additional valuable information on aspects of a topic due to the “recognizable extensive information base (…). It can also make suggestions for content-related connections and transitions between topics.
Legal tech applications with ChatGPT targeted by the administration
The automated fine notices discussed in the legal tech industry and their automated contradictions are currently not an issue. Schrödter does not currently consider direct use in administrative procedures and specialist applications to be possible. In principle, however, the use is “extremely interesting”: There is still an “automation gap” “beyond clearly algorithmically decidable facts and practicable processes”.
The minister’s aim is to open up the “great treasury of information and implicit knowledge” that lies dormant in the files and information systems for administrative procedures. Schrödter is thinking specifically of information services that can automatically create and update topic-related dossiers.
Schrödter does not address the risk that company and administrative secrets or personal data of employees will flow into the ChatGPT data pool via queries or prompts. He only refers here to a release recommendation based on the IT Use Act (ITEG), in which the necessary organizational measures in the area of information security and data protection were also compiled.
Local implementation at Dataport
At the IT service provider Dataport, the necessary foundations have been created to be able to offer solutions based on generative AI and, in particular, large language models. Schrödter: “If scenarios arise in which sensitive data is to be processed, we can implement this as a project.”
A spokeswoman for Dataport told heise online that this may be the platform for AI and data usage dataportai go. This creates the technical prerequisites for the project implementation of application scenarios for the use of AI or the evaluation of large amounts of data.
Recommended Editorial Content
With your consent, an external survey (Opinary GmbH) will be loaded here.
Always load polls
State data protection officer not included
The Schleswig-Holstein State Chancellery has not yet discussed any regulatory issues. Nevertheless, Schrödter announced that the Independent State Center for Data Protection (ULD) would be approached at an early stage. The Schleswig-Holstein state data protection officer Marit Hansen said heise online: “We found out about the release of the use of ChatGPT from the press.”
Of course, the requirements of data protection law must always be met, emphasizes Hansen. The most important thing is: “Without a legal basis, the administration may not release any personal data to ChatGPT, i.e. no data about citizens.” If there were a legal basis for such processing, numerous other questions would arise, such as the implementation of information obligations towards the data subjects, exercising the rights of data subjects, information security and data protection by design & by default.
As early as April 2023, the ULD, together with other state commissioners for data protection, began a regulatory review of ChatGPT. It submitted a questionnaire to OpenAI, the manufacturer of ChatGPT, in order to obtain the information necessary for a data protection assessment. The reply period is still running. The administrative procedure is coordinated by the Federal and State Data Protection Conference’s AI task force, which has been established for years, and at the European level by the newly founded ChatGPT task force at the European Data Protection Board.
State Chancellery must ask itself questions
Hansen announced to heise online that the state chancellery would now have to answer questions about the basis for the ChatGPT release. Hansen: “This can include, for example, a data protection impact assessment, contractual documents with assurances from the provider regarding the data sources, any filters or for your own evaluation of the data entered or even service instructions.”
It would be particularly interesting for data protection officials to find out how Digitization Minister Schrödter intends to achieve the promised transparency about use if he uses ChatGPT for speeches and notes: “Are all prompts and all expenditures filed and are they also included in file inspection requests by those involved? ‘ asks Hansen. That would be “possibly appropriate for statements of facts and notes because the accuracy of the statements is usually of great relevance”. Because neither something essential should be forgotten nor should anything be added. However, such errors are typical of language models that have been trained for language but not for the truth and completeness of facts.
(mki)
