Minutes of council meetings

Municipality of Wijchen

The AI system helps in the reporting of council meetings. The system does this by automatically converting spoken text to written text.

Last change on 27th of November 2024, at 11:15 (CET) | Publication Standard 1.0

• Publication category: Other algorithms
• Impact assessment: Field not filled in.
• Status: In use

General information

Theme

Organisation and business operations

Begin date

2024-09

Contact information

[e-mail]

Link to publication website

https://www.wijchen.nl/over-deze-website

Responsible use

Goal and impact

Writing high-quality minutes of council meetings via artificial intelligence. Security and confidentiality are guaranteed.

The Al system employs sophisticated processes designed to handle transcription and summarisation efficiently and securely.

1. Transcription phase: the audio of the meetings is fed into the AI system. The system converts the spoken language into text. This is done with models that recognise speech. The system uses advanced deep learning models trained on different datasets to recognise speech and convert it into text. The model analyses the sound, 'sees' linguistic patterns and writes text. It does this in near real-time, depending on the length and difficulty of the speech. Moreover, the model recognises different speakers by their voices;
2. Summary phase: after the text is written, the system creates a summary with a Large Language Model (LLM). LLMs are language models that can process large amounts of text. The model understands the context by recognising important themes, details and the general structure of the conversation. Then the model converts this information into a concise summary. The summary accurately reflects the content and intent of the original meeting. The summary process consists of several steps to ensure the completeness and consistency of the final outcomes.

Considerations

The use of the algorithm will make it easier to offer written minutes (also) for interested residents. This will improve our services and the transparency of municipal administration.

Human intervention

• The supplier's note-takers check the minutes and provide feedback. This feedback is incorporated into the system with the aim of training the system;
• The final minutes are shared with the municipality;
• The municipality reviews the minutes, adjusts them if necessary and sends them back with feedback. In this way, the supplier's note-takers and the system can learn from them;
• Both parties have regular contact with the aim of allowing the system to learn in a focused way. Think about how to take minutes per speaker, how to summarise, how to handle a vote, and so on.

Risk management

Council meetings are open to the public. The minutes are therefore 'public information'.

Operations

Data

Field not filled in.

Technical design

The supplier teach the AI system to take minutes at 3 levels:

1. comprehensive summary minutes;
2. verbatim summary on speaker;
3. verbatim transcription by speaker.

The municipality is currently running a pilot focusing on level 2: verbatim summary by speaker. This means noting down for each speaker what was said. The spoken language is converted into written language and well-written sentences are made out of it, staying as close as possible to what was said. The recordings received from public council meetings are converted into minutes by the AI system. Because the supplier's note-takers can make immediate manual adjustments and feed them back to the system. And the system learns what needs to be changed next time.

In summary, the procedure is as follows:

1. finding out and noting who said what;
2. arriving at well-written Dutch sentences;
3. being able to arrive at correct summaries;
4. being able to arrive at correct decision-making;
5. articulating the decision-making process correctly;
6. Arrive at the correct actions
7. articulating the actions correctly.

External provider

Emma Handson