When the legal artificial intelligence (AI) "AiLex" was asked to analyze the school violence case in the Netflix drama "The Glory," this answer came out. It was the result of inputting the entire set of virtual case records, running to thousands of pages, based on the drama's content. The AI quickly grasped the whole story of the case and began cross-checking the characters' statements with the evidence.
It also caught the process by which Park Yeon-jin changed statements. At first, Park said, "I have never assaulted anyone," but when accomplices confessed, Park said, "It was only a scuffle." When Moon Dong-eun's (actor Song Hye-kyo) burn treatment records appeared, Park changed the statement again, to the effect that "a hair straightener did touch the body, but there was no intent to injure." Based on this, AiLex assessed the credibility of Park Yeon-jin's statements as low. In contrast, it analyzed that Moon Dong-eun's statements consistently matched the medical records and physical evidence.
Jang Il-jun, CEO of A2D2, the legal tech startup that developed AiLex, recently told ChosunBiz, "Unlike legal AIs that find precedents and statutes, AiLex reads the case itself." From case records that run to thousands or tens of thousands of pages, it extracts key evidence and identifies who changed their statements and when, helping lawyers review the records.
It also proposes defense strategies. When asked, "Who might testify favorably for Park Yeon-jin?" it reanalyzes the entire record to find testimony that would support the defense's argument. Jang said, "We incorporated the practical experience of senior lawyers who have handled numerous civil and criminal cases into the development process."
Might legal AI threaten lawyers' jobs? Jang thinks the opposite. Jang said, "As AI reduces the time spent on record review and paperwork, lawyers can take on more cases," adding, "Naturally, the barriers to legal services will come down." The following is a Q&A with Jang.
—What kind of legal AI is AiLex?
"While other legal AIs have focused on searching precedents and statutes, AiLex specializes in analyzing the evidentiary record and facts of individual cases. We launched paid services for criminal cases in Jan. and for civil cases in Apr. As of the end of Jul., more than 400 law firms and over 800 lawyers have subscribed, and the number of cases analyzed so far stands at 638. It is still in the early stages, but users are growing quickly."
—What is the biggest difference from other legal AIs?
"That it is specialized for case analysis. In criminal cases, when you print the indictment, statements, financial transaction histories, call records, and KakaoTalk chats, it runs to thousands or tens of thousands of pages. If you upload the materials to AiLex, the AI digitizes them, organizes the case summary, and analyzes the people, statements, and evidence.
In the past, lawyers had to read all case-related records and find important content themselves. AiLex reduces this workload and even suggests which points to rebut or strengthen depending on the issues in dispute. Of course, the final judgment and defense strategy decisions rest with the lawyer."
—How are lawyers responding?
"Many say the time spent reviewing records has been greatly reduced. In large criminal cases, the same person may be questioned multiple times, and the records can swell to tens of thousands of pages. But a substantial portion is not directly related to the key issues. By using AiLex, you can quickly identify important materials and issues without reading the entire record from start to finish."
—How does the AI pick out only the important content?
"Just as people understand sentences in the context of an entire text, the AI analyzes multiple documents together to determine what each means in the case. The indictment, evidence, third-party statements, and suspect statements differ in nature and importance.
AiLex first distinguishes the document types, then classifies content by importance, and analyzes the case around the core materials. We also incorporated the practical know-how of senior lawyers with extensive civil and criminal experience during development."
—What enables you to analyze records that run to tens of thousands of pages?
"It is thanks to our proprietary technology that drastically reduces memory usage. The key is not to have the AI remember the massive documents as they are. We digitize the documents page by page, identify their interrelationships, and then split the facts into small units and store them by topic.
At this point, we use "knowledge graph (Knowledge Graph)" technology to structure relationships among data. By structuring each fact and relationship, we reduce the AI's memory usage."
—You recently adopted agent AI. How is it used?
"It can be used to craft defense strategies. For example, if you ask, "Who might be a witness who would testify favorably for me?" it analyzes the entire record to find witnesses who will support the defense's claims. It can locate statements about coercion or threats in face-to-face interrogation records, or present, with grounds, statements that another defendant played a leading role."
—Is there a possibility that legal AI will replace lawyers?
"Rather than replacing lawyers, the goal is to expand the range of cases lawyers can take. Until now, there have been many cases that were hard for lawyers to accept because the fees were low relative to the time required. But if AI reduces the time for record review and paperwork, lawyers can take on more cases at lower cost. Consumers can access legal services more cheaply, and lawyers can approach new markets."
—Are there no security issues with sensitive case records?
"AiLex operates in an environment isolated from the outside. We do not use user-uploaded case records for AI model training, and the system is designed so that the development team cannot arbitrarily view client records. For AI training, we use separate materials such as statutes and publicly available precedents."
—What are your future plans?
"While we are currently focused on supporting lawyers' casework, we plan to expand in the long term to corporate legal affairs and compliance. Corporations also have many records that need to be analyzed, such as contracts, emails, and investigation materials, and some materials cannot be taken off-site. We plan to develop models for other fields and institutions that handle massive records."