Skip to main content

European Parliament expands access to their archives with Claude in Amazon Bedrock

TL;DR

  • The European Parliament archives developed Archibot, an intuitive tool leveraging Generative AI to manage and process millions of documents dating back to 1952.
  • Archibot enables researchers and the public to quickly navigate vast document sets, find specific information, and build reports, significantly saving time and improving access to collective memory.
  • A core focus was ensuring trust and control over the AI solution, particularly through "Constitutional AI," to guarantee reliable answers and prevent data misuse.

Takeaways

  • The European Parliament archives preserve, manage, and process all documents produced and received since 1952, including legislation, internal administration, and plenary documents.
  • Archibot is an AI-powered tool designed to help researchers and the public find documents and build reports from the vast collection of parliamentary records.
  • The system was built using Anthropic-Claude technology, with an emphasis on Constitutional AI to ensure trustworthy and controlled responses from the language model.
  • A critical design principle for Archibot was maintaining strict control over the Generative AI solution, preventing data collection or extraction for other purposes, and clearly attributing information sources.
  • The real-time wisdom and summarizing capacity of Generative AI for huge document sets represents a significant shift in managing and accessing archival information.
  • Archibot has expanded multilingual access, moving beyond its previous French-only limitation, to ensure broader accessibility to the archives.
  • Accuracy, reliability, and multilingual capacities are crucial for the system, supporting the foundational principles of a democratic project by providing access to information.
  • The tool significantly eases the work for researchers, policymakers, educators, and anyone interested in the Parliament's documentation.

Vocabulary

  • Generative AI — A type of artificial intelligence that can produce new content, such as text, images, or code, rather than just analyzing existing data.
  • Archibot — The specific AI-powered tool developed by the European Parliament archives to help users find and build reports based on their extensive document collection.
  • Anthropic-Claude — Refers to the AI large language model technology from Anthropic, specifically "Claude," used as the foundation for Archibot.
  • Constitutional AI — An approach to AI safety and alignment, aiming to make AI models follow a set of principles or "constitution" to ensure trustworthiness and prevent harmful outputs.
  • Plenary documents — Official records and papers associated with the full assembly sessions (plenaries) of a legislative body, such as the European Parliament.
  • Resolutions — Formal expressions of opinion, intent, or decision adopted by a legislative body or committee.
  • Collective memory — The shared pool of knowledge and information in the memories of a social group, often preserved and supported by archives and historical records.

Transcript

Preserving collective memory is one of the fundamentals of democracy. Welcome to the archive of the European Parliament. The opportunities in charge of preserving, managing and processing all the documents that have been produced and received by the European Parliament since 1952. We have documents related to the legislation, we have documents related to the internal administration, all the plenary documents. We can find resolutions that the Parliament has been taken in positions of the Parliament's impulse or like the legal shacers with the other institutions. We started, we spent thousands of documents removed to hundreds of sizes, now we are two millions of documents. We decided to provide a way to navigate inside a huge document set. The arrival of the Generative AI is something that was a real shift. Archibot is a simple and intuitive tool that helps us and the researchers to find documents and to build reports based on the documents. It's available on our website, we have users from Gabon and from Australia, and BXS from everywhere. We built Archibot using anthropic-cloned technology. And there was one point that got the attention, this is constitutional AI. We emptied that the answer provided by the last language model is a trust-for-the-answer. The most important thing for us is trust. We wanted to make sure about one thing. If we use Generative AI, we permanently need to be under control of the solution that we've built. We didn't want to have any data that could be collected or extracted for all the purposes. We absolutely want that people know where information is coming from. Previously we had only a French as a language that could use, so we needed to move maintainable so that all countries could have an access to our archives. Accuracy, reliability, and the mutiny wisdom capacities are for us crucial. I can't imagine how much time it is saving because the archives are huge. As a researcher, it makes my work way easier. It's also way more easy for policymakers, educators, and whoever is interested. Within the European Parliament, we have shown that we can use Generative AI in a controlled way. This real-time wisdom and summarizing capacity of huge document sets was really a shift from Generative AI, you know, to me. The access to information and all the tools we provide in order to be able to make use of this information are real foundations of a democratic project.

Feedback / ReportSpotted an issue or have an improvement idea?