How do I get to the municipal lost property office? When is the vehicle registration office open? How do I register a business? Employees of Hamburg's public authorities answer questions like these on a daily basis. For some time now, they have been supported by Frag-den-Michel - a text-based dialogue system that was developed as part of a research project with the DAI Laboratory at TU Berlin and is now operated and further developed by Dataport. With the support of AI, the chatbot has the right answer to many recurring enquiries, 24 hours a day, seven days a week.

Michel: "Hello! How can I help you?"
Caller: "Where can I apply for a new driver's license, please?"
Michel: "There are several locations in Hamburg for this service. Here are the addresses."
Caller: "Great, thank you very much."
Michel: "You're welcome, see you next time!"

At first glance, this is an information situation that is commonplace in many public authorities in Germany. However, in this case, Frag-den-Michel is not an employee of the city administration, but a chatbot. Since April 2019, it has been dealing with the administrative concerns of Hamburg's citizens around the clock. Now available in ten different languages, the digital assistance system "Frag-den-Michel" not only provides reliable answers, but also masters etiquette such as the typical greeting or small talk. Frag-den-Michel currently conducts around 3,000 virtual dialogues every month, with a third of enquiries coming in outside the availability of the Hamburg telephone service.

Intuitive conversation in context

Frag-den-Michel uses artificial intelligence to provide suitable answers to citizens' questions. The system analyses a question and attempts to derive the underlying intention based on a context model. To do this, it checks the input with regard to the (foreign) language used, corrects any spelling mistakes and calculates the probability of potential concerns. Frag-den-Michel provides its information in a natural dialogue and answers questions in the relevant context. Unlike common information systems, the chatbot does not play out long, complex documents, but instead provides very specific information depending on the request. The context model enables intuitive conversation and natural human-machine interaction.

The chatbot uses the content of the Hamburg Service Information System (HaSI), among other things, as its knowledge base.The usage data and feedback from users help to continuously develop the system.However, Frag-den-Michel does not operate completely autonomously, but is actively trained - for example, for questions to which it has not yet been able to provide satisfactory answers.In such cases, a suitable answer is searched for using various data sources and a decision is made as to whether the question belongs to an existing dialogue category or whether a new category needs to be added to the system.The chatbot's internal models are continuously updated with these clearly defined question-answer pairs.Automated tests are also used to regularly check the correctness of the question-answer assignment.

More networked in the future

In the future, Frag-den-Michel will expand its range of services.To this end, the City of Hamburg has derived the expectations and wishes of users from the live operation of the chatbot and will add further functionalities to the chatbot.The aim is to make the conversation even more intelligent in the future in order to get from the question to the optimal answer more quickly.There are also plans to test the integration of large language models (LLM) and the integration of forms or other online processes directly into the dialogue.Information already entered in the chat could thus be transferred without media discontinuity or multiple entries.Linked to the appointment system of the customer centres, available times could then also be booked directly and easily via Frag-den-Michel.