Didact: A Cross-Domain Capability Discovery System for Defence
Didact is a prototype system that addresses fragmented defence information by integrating Australian policy documents and research publications into a knowledge graph with a composite retrieval-augmented generation pipeline and an interactive Evidence Rail to facilitate efficient, auditable capability discovery through natural language conversations.
Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
Imagine you are a military strategist trying to solve a complex puzzle. The pieces of this puzzle are scattered across two very different rooms: one room is filled with thick, official government reports and policy documents (the "Defence" room), and the other is filled with thousands of academic research papers and scientist profiles (the "Research" room).
Usually, if you want to find a specific piece of information, you have to run back and forth between these two rooms, manually searching through piles of paper. It's slow, confusing, and you might miss how a new scientific discovery connects to an old government policy.
Didact is a new digital tool designed to fix this mess. Think of it as a super-smart, highly organized librarian who doesn't just fetch books but actually understands how the two rooms are connected.
Here is how Didact works, broken down into simple concepts:
1. The Two-Track Library System
Didact doesn't just dump everything into one big pile. It keeps the two types of information separate but connected, like two different tracks on a train system:
- Track A (Documents): It reads through official Defence reports and policy papers. To keep things secure (just like real government work), it has "security clearance" levels. Some documents are public, and some are "classified," and the system ensures you only see what you are allowed to see.
- Track B (The Web of Knowledge): It looks at Australian research papers and builds a giant, interactive web (a "knowledge graph"). Instead of just reading the text, it understands the relationships: "Who wrote this?" "Which university are they from?" "What topic does this cover?"
2. The "Composite" Detective
When you ask Didact a question (like, "Which Australian researchers are working on advanced sensors for dangerous environments?"), it acts like a detective with two different toolkits:
- It searches the Document Track for official reports that mention these sensors.
- It searches the Web Track to find the specific scientists and universities involved.
- It combines these answers into one smooth conversation, rather than giving you two separate lists.
3. The "Evidence Rail" (The Magic Window)
This is the most unique part of Didact. In many AI chatbots, you just get an answer, and you have to trust it. Didact is different.
Imagine you ask the librarian a question, and they hand you a summary. But next to that summary, they slide out a transparent "Evidence Rail."
- This rail shows you exactly which documents and which parts of the research web the AI used to build its answer.
- It's like a "Show Your Work" button. You can click on a name in the rail, and it zooms in to show you the original research paper or the specific government policy that supports the claim.
- This prevents the AI from "hallucinating" (making things up) because every single sentence is tied to a real source you can inspect.
4. Why This Matters for Defence
The paper explains that for defence policymakers, trust is everything. They can't just accept an AI's guess; they need to know why the AI thinks a certain technology is important.
- The Problem: Information is currently "fragmented" (scattered everywhere).
- The Solution: Didact brings these scattered pieces together into one view, but it keeps the security boundaries intact.
- The Result: A capability manager can quickly see the "big picture"—connecting a new scientific breakthrough to a national security priority—and show their boss exactly where that information came from.
How Well Does It Work?
The authors tested Didact with 50 different types of questions (like "summarize this," "find this expert," or "explain this policy").
- Speed: It answers quickly (usually in about 2 seconds).
- Accuracy: It found the correct source document 68% of the time as the very first result, and 76% of the time within the top two results.
- Completeness: In 90% of cases, the information it pulled actually covered the answer needed.
In a Nutshell
Didact is a prototype tool that helps defence experts talk to their data using plain English. It acts as a bridge between official government reports and academic research, ensuring that every answer comes with a clear, clickable trail of evidence so users can verify the facts themselves. It is currently being tested in Australia but is designed to work in any field where important information is scattered across different, disconnected sources.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.