Artificial intelligence & digital democracy
The developments in the fields of artificial intelligence (AI) and machine learning hold great dangers, but also a lot of potential for democratic opinion-forming on the internet. Since the technological possibilities for training AI models are developing very quickly and these models are becoming more and more widespread at the same time, we as an association are also in a constant learning process. One important prerequisite is already clear to us as Liquid Democracy e.V.: Artificial intelligence can only meaningfully support our democratic discourse if its development is oriented towards the common good and the training and use of AI models is made as transparent and comprehensible as possible. It is also important to be aware of the possible discrimination risks of AI systems and to actively counteract them.
With this in mind, since 2020 we have been critically exploring and testing the opportunities, but also the risks, surrounding AI on the basis of our vision, mission and values. Drawing on long-standing and new partnerships with universities, research institutions and other civil society organisations, we are testing the integration of AI models into our participation software Adhocracy. In addition, in collaborations with universities and other partner institutions, we are researching the impact of AI-powered moderation on discussions online and developing ideas and concepts on how AI can be used meaningfully for digital democracy.
Artificial intelligence research projects
In three of our current projects, we are looking at AI language models as well as automatic detection of fake news.
The aim of this project is to create a concept for the detection and correction of discrimination in AI language models. Together with Beyond AI Collective e.V., Heinrich Heine University Düsseldorf and TU Ilmenau, we are developing ideas and processes for this on the basis of our AI language model developed in the KOSMO project.
Duration: 06/2023 - 12/2023
More information: KI-MoDi project page
The aim of KOSMO is to find out whether and how the moderation of digital discourse can be supported with AI, so that the quality of discussion is increased. In the project, we integrated a prototype for an AI-based moderation assistant into our digital participation platform adhocracy.plus. The KOSMO language model was developed by Heinrich Heine University Düsseldorf.
More information https://liqd.net/de/projects/kosmo/
Testing how to counter disinformation campaigns in digital online discussions with AI is the goal of our project DeFaktS. Together with the Forschungszentrum Informatik and the Philipps-Universität Marburg, an AI algorithm for the automated detection of disinformation will be integrated into our software adhocracy.plus and tested to see what effect this has on online discussions.
Duration: 01/2022 - 12/2024
More information: https://liqd.net/de/projects/defakts/
Our research questions
- How can we integrate AI into digital participation platforms in a sustainable and meaningful way?
- How can AI be used transparently and meaningfully to enhance the quality of online discussions?
- How can AI be used meaningfully to increase the accessibility of democratic processes?
- How can discrimination risks of AI language models (Large Language Models and Foundation Models) be identified and counteracted?
- How can Explainable AI work in practice - especially in the context of digital democracy?
- What does it take to participate in and advance the development of public good AI as an NGO with few resources?
- How can we ensure that AI models are developed, trained and operated under economically fair and ecologically sustainable conditions?
- How can we ensure that generative AI is not used to influence democratic discourse through the generation and misinformation?
Unser Lernprozess und Meilensteine
Talk bei der re:publica: High Standards and Low Budget – Künstliche Intelligenz für demokratische Öffentlichkeiten (Link)
Unsere Idee „KI-MoDi“ wird beim Ideenwettbewerb „Gemeinsam wird es KI“ des Bundesministerium für Arbeit und Soziales (BMAS) prämiert. (Link)
Liquid Academy: Wie Künstliche Intelligenz die Bürger:innenbeteiligung revolutionieren kann (Link)
Launch des Prototypen von KOSMO (Link)
Das Projekt defakts startet in die Umsetzung (Link)
Liquid Tank #5 mit Michael Puntschuh zu Thema „Sexist AI – Wie Diskriminierung in den Algorithmus kommt“ (Link)
Die Projektumsetzung unseres Projekts „KOSMO“ beginnt (Link)