Radical Innovation for Citizen Participation: The MultiPoD Architecture
How can we encourage citizens to engage in the democratic process while helping to eliminate biases and identifying factors that hinder participation? In recent years, experimental approaches have proven particularly helpful in tackling these questions. They include techniques to understand how people make decisions or how the specific presentation of information can influence choices.
MultiPoD combines experimental exercises with behavioural insights from the digital deliberation tools by partners webLyzard and Decidim. The goal is not just to map and understand the changes in behaviour, culture, and society required to promote political engagement, but also to encourage behaviours that increase participation.
4 pilars of the MultiPoD Architecture: A New Approach to Political Participation
The MultiPoD architecture is an innovative approach to enhancing citizen participation, leveraging the power of language models and knowledge graphs to facilitate more inclusive and informed deliberation. A language model is a type of artificial intelligence designed to process and understand human language. A knowledge graph is a graphical representation of knowledge that showcases entities and their relationships. By supplying real-world insights, knowledge graphs improve the accuracy and reliability of AI. In the context of MultiPoD, a Culture-Specific Language Model (CSLM) and an evolving Culture-Specific Knowledge Graph (CSKG) are the core components of the architecture, enabling the platform to capture, annotate, federate, transform, connect, and empower citizen participation.
1. Capture and Annotate: The First Steps in the MultiPoD Workflow
The MultiPoD workflow begins with the capture and annotation of content from existing and new Decidim platform instances, using APIs from partners such as webLyzard and their Web crawler, as well as the Incriptis.org open-source text extraction library. We use this comprehensive set of language resources to train the culture-specific language model and evolve the culture-specific knowledge graph. It provides a rich layer of metadata to classify each interaction and enrich it with factual and affective metadata.
2. Federate and Transform: Overcoming Language Barriers
The federate and transform steps in the MultiPoD workflow involve using metadata annotations to interlink related threads in the deliberation process. This two-step process includes state-of-the-art translation through eTranslation and post-processing to improve the output of the translation component using generative AI and the CSLM. This customisation aims to engage minorities by incorporating their views and supporting experiments to identify the best ranking algorithms for content recommendations that promote inclusion. These experiments will draw directly on the rich set of community resources ingested in the capture/annotate step.
3. Connect and Empower: Enhancing Citizen Participation
The connect and empower steps in the MultiPoD workflow focus on identifying unknown expertise and structuring interactions to engage linguistic minorities. The goal is to support decision-making by highlighting best practices developed in other Decidim communities. The development of the MultiPoD extensions to Decidim’s hybrid deliberative solutions will be guided and evaluated through iterative feedback loops in the Citizen Assemblies of Work Package 4, including citizens with communication impairments.
4. Reveal and Explore: Uncovering Barriers to Participation
The final step in the MultiPoD workflow is the reveal and explore phase, which involves using webLyzard’s Visual Analytics Dashboard to provide unprecedented transparency through aggregated visualisations on deliberation processes. This open architecture will make the collected information freely available to the research community and other projects. Its advanced query and on-the-fly drill-down features will reveal barriers to political participation and provide new insights into how to overcome these barriers and engage minorities.
Connection Comes First
With the culture-specific language model and knowledge graph, MultiPoD advances the field of neurosymbolic AI. In addition to these cutting-edge developments, the project focuses on connection and inclusion on both the community and technology levels. MultiPoD will use citizen feedback to connect and improve existing tools. These improvements feed back into the community to connect diverse groups of people all over Europe and beyond.