Political statutes are foundational documents. They define how power is distributed, how decisions are made, and how conflicts are resolved. In theory, they are written for everyone. In practice, they are read by very few.
This is not because people lack interest. It is mostly a matter of friction. Length, legal language, internal references, and historical layering make statutes difficult to approach unless you already know where to look. Over time, this creates a knowledge gap between those who regularly operate within governance structures and those who do not.
While working within the PvdA, I encountered this gap repeatedly. Even relatively simple questions often required navigating dozens of pages, interpreting procedural language, or relying on institutional memory rather than the document itself. The statutes were available, but not truly accessible.
This raised a broader question:
If rules shape participation, what are the implications when only those with expertise can truly understand them?
That question became the starting point for PvdAI.
What is PvdAI?
PvdAI is an experiment in reducing the barrier to accessing complex political documents. It allows users to explore the statutes of the Dutch Labour Party and ask questions in natural language, while keeping the original text as the single source of truth.
The goal is not to rewrite, summarize, or simplify the rules themselves. Simplification can easily turn into interpretation, and interpretation introduces authority. Instead, PvdAI focuses on orientation. It helps readers find relevant sections, understand how articles relate to one another, and surface answers that remain grounded in the original text.
AI is not treated as an expert here. It functions as a guide.
PvdAI is built as an open-source project, with the full codebase available on GitHub: https://github.com/julianaijal/PvdAI
A live demo of the project is available at: pvdai.tech.
From static documents to navigable knowledge
Statutes are usually published as static documents. PDFs work well for preservation and version control, but poorly for interaction. They assume linear reading, while most real questions are not linear at all.
PvdAI treats the statutes as a structured knowledge source instead. Articles, chapters, and sub-sections are explicitly modeled, which makes relationships between rules visible rather than implicit. This already changes how the document can be explored.
On top of this structure, semantic embeddings are used to retrieve relevant passages based on meaning rather than exact phrasing. When a user asks a question, the system identifies which parts of the statutes are most relevant and generates an answer constrained to those sections.
This constraint is essential. The system does not invent new rules or fill in gaps with general knowledge. Every answer can be traced back to specific articles. In that sense, the output remains verifiable.
Rather than replacing interpretation, the system makes it visible.
Interpretation, power, and interfaces
Every interface introduces bias. Search boxes, navigation menus, summaries, and even tables of contents influence how information is accessed and understood. Legal documents are no exception.
Traditionally, interpretive power sits with those who know where to look and how to read procedural language. By adding an AI-driven interface, some of that power shifts. Questions that once required guidance from a board member or committee can now be explored independently.
This does not remove hierarchy, but it does make it more transparent. When an answer points directly to a specific article, disagreements can move away from authority and toward the text itself. That shift matters.
At the same time, this also introduces responsibility. If AI becomes a primary access point to rules, its limitations must remain explicit. Ambiguity, edge cases, and unresolved interpretations should stay visible rather than being smoothed over. PvdAI avoids definitive language when the statutes themselves are open-ended.
Clarity should not come at the cost of legitimacy.
Why this matters beyond this project
Political parties are only one example. Many institutions rely on documents that are formally public but practically opaque. Regulations, bylaws, internal policies, and codes of conduct often function as gatekeepers rather than enablers.
AI makes it possible to rethink how people interact with these texts. Not by replacing them, but by turning them into systems that can be explored iteratively. Asking follow-up questions, validating assumptions, and checking answers against source material become part of the reading process.
This aligns with a broader shift already visible in search, structured data, and knowledge graphs. Information is no longer consumed linearly, but contextually. The challenge is ensuring that this shift improves understanding rather than hiding complexity behind fluent output.
Why the technical setup is intentionally simple
From a technical perspective, PvdAI is intentionally kept simple. There is no database of user behavior, no personalization layer, and no attempt to optimize for engagement. Questions are processed and discarded, and the statutes themselves remain static and inspectable.
This is a conscious design choice. When dealing with governance and rules, predictability and transparency matter more than technical sophistication. A system that can explain why it produces a certain answer is more valuable than one that produces fluent output without clear provenance.
In this context, AI is not a product feature. It functions as an infrastructural layer that supports access, not interpretation.
Looking ahead
Tools like PvdAI raise a broader question about how we engage with complex information. As conversational interfaces become more common, reading may increasingly mean navigating rather than consuming content from top to bottom.
The risk is not that people stop reading documents, but that they stop questioning the interfaces through which those documents are accessed.
PvdAI is a small experiment, but it points to a larger design challenge: how to use AI to reduce friction without reducing agency.
That question remains open, and it is worth exploring.