Language:
Mann und Frau am Schreibtisch mit kleinem Roboter

AI Competence Centre: AI Development for the Administration

Using artificial intelligence (AI) safely and effectively is a basic requirement for modern administration. The AI Competence Centre (KI-KC) tests and promotes the use of AI in the German Federal Administration. Bundesdruckerei is working with public authorities to test AI use cases and develop user-centred AI prototypes. The resulting findings are shared within the administration in order to advance the digital transformation with the help of AI.

AI Competence Centre: AI Innovation for Public Administration

On behalf of the German Federal Ministry of Finance (BMF), Bundesdruckerei GmbH has been working through the AI Competence Centre (KI-KC) since 2023 to support the German Federal Administration in the confident handling of AI and in the development of AI applications. To this end, Bundesdruckerei is researching the use of AI for various applications in day-to-day administration in project collaborations with federal ministries. The focus lies on developing practical AI solutions that are specifically tailored to the needs of public authorities. Current projects concentrate on language models, image recognition technology and anomaly detection.

Our administration is the foundation of our social coexistence. Reliability, transparency and traceability of decisions are its most important values. As artificial intelligence (AI) becomes increasingly prevalent in many areas of life, our administration is faced with the question of how we can best utilise its potential.

Dr Maxim Schnjakin, Project Manager at the KI-KC

The KI-KC works closely with the data laboratories of the federal ministries and involves stakeholders from science, business and civil society. The AI prototypes are developed by Bundesdruckerei in collaboration with the administrative staff. 

Data & Facts at a Glance

Period:
Established in June 2023
Leadership:
German Federal Ministry of Finance (BMF)
Scope:
  • Language models
  • Image recognition technologies
  • Anomaly detection
  • Explainable AI / trustworthy AI

How the KI-KC Works

The AI Competence Centre relies on an explorative and iterative approach. In co-creation workshops, interdisciplinary project teams comprising UX experts and data scientists from Bundesdruckerei initially identify suitable use cases together with administrative staff. Once the specific use case with the greatest added value for users has been identified, suitable AI models and data science procedures are tested and compared. AI prototypes are then developed in short iterations. The aim is to quickly demonstrate the feasibility of a use case and thereby advance the utilisation of AI potential in administration. This gives the project participants an answer to the question of whether the promised added value for the group of users justifies product development.

The development process includes the following:

  1. Needs analysis
    Definition of use cases in close collaboration with the users
  2. Data preparation
    Identification and preparation of relevant data records, taking into account legal requirements
  3. Prototyping
    Development and testing of AI applications with regular involvement of a lead user group
  4. Evaluation 
    Continuous improvement of the prototypes based on user feedback

The results flow directly into further development in order to tailor the solutions more precisely to the specific needs of each public authority.

Customised AI Development for Public Authorities

What all prototypes developed by the AI Competence Centre have in common: They explore the possibilities of AI for public administration, evaluate specific technologies and test their added value by making them available to a small group of users. A selection of three current projects shows the range of possible applications.
 

ADLER: Local Language Models for Sensitive Data

ADLER (Ausführung datensicherer LLMs, evaluiert auf resilienten Geräten, “execution of data-secure LLMs, evaluated on resilient devices”) investigates how small language models can be operated locally on commercially available devices. This allows sensitive data to be processed directly on the work tool without having to upload it via networks. This strengthens data sovereignty and opens up new utilisation contexts, especially for confidential information such as VS-NfD data.

  • Goal: secure processing of sensitive data on local devices
  • Benefit: energy-efficient, application-specific models

MÖVE: Benchmarking of Language Models for Public Administration

MÖVE (Modelle für die öffentliche Verwaltung evaluieren, “evaluate models for public administration”) is a framework for evaluating language models with regard to their performance in administrative tasks, results in German and governance criteria such as social fairness, hallucinations (false or misleading results) and conformity with the AI regulation EU AI Act. The aim is to establish a sound basis for arriving at decisions on the use of language models in public administration.

  • Goal: comparison and evaluation of language models for specific administrative tasks and in German
  • Benefit: sustainable selection of suitable models via a publicly available website 

AI Language Assistant for the German Federal Administration: Efficient Information Processing

The language model project for AI-supported text analysis enables documents to be analysed quickly and comprehensively in order to extract relevant information and thus support decision-making. Automatic summaries, topic-related analyses and the identification of responsibilities increase the efficiency of administrative work.

  • Goal: optimisation of information processing in the Federal Administration
  • Benefit: fast document analysis and efficient decision-making

Achieving Sustainable Public Administration with the KI-KC

The Federal Administration’s AI Competence Centre is making a significant contribution to driving the digital transformation of public administration. By closely involving users and integrating the latest research findings, the KI-KC is developing solutions that are both innovative and practical. In addition, it supports the Federal Administration in developing and implementing solutions with regard to AI. The prototypes created are laying the foundation for sustainable product development and strengthening the Federal Administration’s expertise in handling AI. The KI-KC is thus contributing to more efficient and transparent administration and helping to successfully master the challenges of the digital future.

Bundesdruckerei GmbH brings a broad range of know-how to the table including agile development methods, an open data philosophy and the integration of state-of-the-art technologies, as in the form of language models. Moreover, the Federal Administration also benefits from Bundesdruckerei’s special expertise and many years of experience in anonymisation and pseudonymisation.

Artificial intelligence offers enormous opportunities for public administration – from the optimisation of internal processes to better citizen orientation. With the AI Competence Centre, we are creating a platform that combines innovation and user centricity to develop sustainable solutions to the challenges facing society in the future.

Camilla Dalerci, Deputy Project Manager at the KI-KC

Questions and Answers about the Project

The AI Competence Centre (KI-KC) identifies use cases for artificial intelligence for the federal administration, develops practical prototypes and expands knowledge about AI technologies. In workshops, specific challenges are analysed together with administrative staff, and solutions are developed that are closely geared to the needs of the users. The prototypes developed can analyse the feasibility of a use case and make initial results tangible. In addition, the KI-KC offers presentations, training and counselling to sustainably strengthen the competence of the administration in dealing with AI.

The KI-KC develops prototypes with the aim of making the day-to-day work of administrative staff easier by automating time-consuming routine tasks and supporting data-based decisions. Applications such as voice assistants or anomaly detection systems help to find relevant information more quickly and organise processes more efficiently. At the same time, employees learn how they can use AI and benefit from feasibility tests on customised solutions tailored to their specific needs.

A typical KI-KC project lasts for three to six months. During this time, the use case is first defined and the existing data analysed. An AI prototype is then developed and regularly tested by end users. At the end of the project, an evaluation documents the results and identifies possible next steps. This agile approach makes it possible to achieve tangible results quickly and respond flexibly to new requirements.

A prototype is a functional software solution that maps specific technical functions and can be tested by users. Its purpose is to check the feasibility of a use case and to provide initial insights into the practical benefits of an AI application. In contrast to a complete production system, non-functional requirements such as scalability or performance are only realised to a limited extent. A prototype therefore forms the basis for deciding whether comprehensive product development makes sense.

The protection of sensitive data is a top priority for the KI-KC. Strict compliance with the General Data Protection Regulation (GDPR) is ensured for all projects. Sensitive data is processed in anonymised or pseudonymised form so that it cannot be traced back to individuals. In addition, all systems run exclusively via the German Federal Government’s networks, which meet the very highest security standards.

The KI-KC focuses on the user-centred development of specific prototypes that are directly tailored to the needs of the administration. While other initiatives often emphasise strategic consulting or basic research, the KI-KC is geared towards practical solutions and their direct application. 

Do you have questions about the KI-KC?

Send us an e-mail kontakt-kikc@bdr.de.

Find the project page PLAIN here
PLAIN: a data analysis platform for the entire German federal administration

The public administration works with large amounts of data and has started to use artificial intelligence (AI). In order to process this data securely, the federal administration needs robust and efficient tools.

Events in der Bundesdruckerei-Gruppe
Events

Erfahren Sie mehr über uns, unsere Angebote und mit welchen Lösungen wir Ihr Unternehmen unterstützen können. Besuchen Sie uns auf Messen und Kongressen, treffen Sie uns bei einem unserer Vorträge oder nehmen Sie an unseren kostenlosen Online-Seminaren teil.