IMDA Enhances Legal Research Process by Developing New Large Language Model for Legal Case Summarization Using Amazon SageMaker

Legal research in Singapore traditionally required manual reviewing of lengthy court judgments, some exceeding 200 pages. This process was time-consuming and inefficient, often taking 5 to 10 hours to research and look through cases.

IMDA and SAL recognized the need for an AI solution that could efficiently summarize complex legal texts while preserving the nuances of Singapore law. Existing large language models (LLMs) were not equipped to handle the specific legal terminology and mixed-language content required, making a customized implementation essential. By assessing and revising the LLM model with SG Courts and existing LawNet data, it ensured that the AI generated content was not only accurate but also aligned with the needs within the Singapore legal community.

IMDA, in collaboration with SAL, used AWS to develop GPT-Legal. By performing domain-specific finetuning and deploying this AI-driven solution on AWS, they reduced the time required to summarize long court judgments from days to minutes. As a result, legal professionals could focus on higher-value tasks, enhancing the overall quality of legal research.

To build and deploy GPT-Legal, IMDA utilized Amazon SageMaker, a fully managed service that streamlined the development, training, and deployment of its machine learning models. Amazon SageMaker provided the flexibility needed to tailor GPT-Legal to the intricacies of Singapore law. This included addressing specific legal terminology, lingo and tone, which general models struggled to capture the nuances. The availability of computational resources on AWS SageMaker is absolutely crucial for the successful training of models. 

IMDA also used Amazon SageMaker's data lineage capabilities to track both public and proprietary data during ML model training, from data ingestion through to deployment. The statutory board relied on this capability to maintain transparency and remain compliant with data usage regulations.

The statutory board also used Amazon Simple Storage Solution (Amazon S3) and Amazon Elastic Compute Cloud (Amazon EC2), a secure and resizable compute capacity service, to provide the necessary computing resources for testing and evaluating the AI solution. With this infrastructure, GPT-Legal efficiently processed and summarized over 15,000 court judgments, improving the accessibility of case information to legal professionals. 

To further enhance the reliability and accuracy of the legal summaries, IMDA integrated advanced validation tools within GPT-Legal to detect and highlight discrepancies, often referred to as "hallucinations". These tools identify and highlight sections of text or entities that may be inaccurate or unsupported by the source material. Legal researchers would then be more aware and cross-check to validate the information. The features provide legal professionals with greater confidence in the AI-generated summaries. It also helped GPT-Legal maintains the high standards of accuracy and trust required in the legal field.

Following the successful deployment of GPT-Legal, IMDA and SAL plans to introduce advanced search features, empowering legal professionals to search and compare legal summaries more effectively. With this enhancement, legal professionals can discover precedents, journals, and legal publications more effectively, improving the overall research process. 

Additionally, SAL is exploring opportunities to integrate GPT-Legal into other legal tech products, potentially expanding its application to summarize private legal documents such as prospectuses and contracts.

Looking beyond legal research, IMDA is considering the use of GPT-Legal's underlying technology in other sectors. For example, in healthcare, similar AI-driven summarization could streamline complex information processing tasks.

“Pushing the boundaries of Singapore’s legal tech scene through emerging technology, IMDA is excited to partner SAL to develop GPT-Legal in harnessing GenAI capabilities to boost legal research capabilities. GPT-Legal through LawNet AI, which 75% of Singapore's lawyers use, will speed up legal research by summarising over 15,000 Singapore court judgements. GPT-Legal, is powered by AWS, is an innovative tool designed to cater to the specific nuances of our nation’s legal system and will look to ensure that our legal professionals can confidently use AI in supporting their work, while upholding the highest standards of legal practice”, said Daniel Ng, Director, Advanced Intelligent Software Systems, BizTech Group, IMDA.