Lauuditors

Inspiration

One of the biggest challenges that lawyers and law students alike face in using AI to aid our legal research now is the “hallucination” of AI. As law students, this was a problem faced during our moot module, Legal Analysis, Research and Communication (LARC).

The current AI system that students and lawyers are utilising for legal research is inadequate and misleading, and the generation of nonexistent or inaccurate case law risks undermining the quality and honesty of materials submitted to the court. The courts have emphasised that users remain responsible for the independent verification of the materials submitted, yet manual verification of every single authority requires users to identify every authority, confirm and read the judgements fully, which adds to lawyer workload, increases costs of litigation and access to justice for the layperson. Therefore, we created Lauudit to make that verification more efficient, clearer, and more transparent, without replacing the user’s own human judgement.

What it does

Lauudit is a Chrome extension that operates as an independent audit layer over browser based legal AI tools. A user selects an AI-generated content and activates Lauudit. The extension will then: Extract the case name and citation from the selected response Check whether each citation can be verified against an available verified corpus (e-litigation/lawnet) Compares the generated case name with the canonical case title Where sufficient judgements are available, retrieve relevant passages and provide an assessment of whether the judgement appears to support the legal proposition Depending on the source evidence and audit status, flagging further uncertain findings that require human review.

How we built it

We started by discussing and drawing out what we wanted our program to do, which led to the list of steps stated above. We used these steps as a guide to create a codex prompt that could output the base design of the extension, which started as a popup that parsed through the entire AI-generated output of legal information. From there, we split the workload into frontend and backend development, slowly incorporating more ideas, such as the ability to audit specific lines of output for frontend and using eLitigation as a model for the data on the backend.

Challenges we ran into

We wanted to use eLitigation as our website for our model to access Singapore judgements from, because it is accessible by the public. However, on our hacking day, eLitigation underwent maintenance from 2pm-11pm which meant one of our main databases could not be accessed and tested.

The Corpus coverage depends on the judgements and database that are currently publicly available. Therefore our model could not access Lawnet, Lexis and WestLaw because these websites required an account to be accessed.

Accomplishments that we're proud of

We combined our knowledge and experience as law, computer science and business students to identify current gaps in the use of AI in legal research, build a model and create a pitch deck for our product.

The product has a user-friendly UI. The fact that it is a Chrome Extension makes it very convenient and quick for use, because users do not have to switch tabs when using it to verify the cases.

What we learned

We learnt that it is very difficult for AI to fact check the output of AI because of lack of access to official case viewing websites like Lawnet, Lexis and WestLaw.

In addition, the auditor does not need to be audited by another auditor, as seen from our product’s absence of AI.

What's next for Lauuditors

We want to improve our model to allow it to differentiate between obiter and ratio, to incorporate the precedential weight of cases and to better decipher the relevance of case content, which will allow us to address more complex contextually inaccurate legal propositions.

We also intend the model to be able to aid users in statutory interpretation, by being able to access Singapore Statutes Online and Hansards of parliamentary materials.

Our long term objective is for Lauudit to become a transparent and quality-assurance layer for AI-assisted legal research.

Built With

• claude
• codex
• javascript
• npm
• pytest
• python