In a groundbreaking announcement, Taiwan’s Judicial Yuan revealed on August 27, 2023, that it has successfully developed a generative AI system designed to draft judgments for criminal cases involving driving under the influence (DUI) and assisting fraud. This AI system is set to undergo trial runs in selected courts starting September. Moreover, ongoing efforts are dedicated to crafting more judgment drafting AI systems to be used in other categories of civil and criminal cases.
This initiative primarily aims to mitigate judges’ workload and reduce the backlog especially in cases with relatively straightforward fact patterns but significant influx of filings.
In response to concerns from the public and legal communities on privacy and security, the Judicial Yuan indicated that this system was built on the TMT5 language model architecture and underwent supervised fine-tuning for judgment drafting tasks. The server is completely locally hosted, and the data training and content generation are carried out also totally locally.
The Judicial Yuan’s press release indicated that the content generated by this AI derives primarily from the indictment filed by prosecutors in each case, while the alleged crime facts and the list of evidence within the indictments are instantly analyzed to generate draft judgments for the judge’s reference. However, judges continue to wield full authority in determining facts, applying the law, and deciding sentences as mandated by the Constitution.
The press release also revealed that the Judicial Yuan will incorporate the “Guidelines for the Executive Yuan and its Agencies on the Use of Generative AI” (the EY Guidelines) as a framework for scrutinizing their development and use of generative AI. The Executive Yuan is Taiwan’s highest administrative authority, while the Judicial Yuan is the highest judicial branch of the government.
The EY Guidelines, published on July 18, 2023, currently stands as the most important government document on regulating AI. With influences drawn from the draft Artificial Intelligence Act of the EU, it encompasses a range of instructions, and those relevant to the Judicial Yuan’s AI project include:
- The use of generative AI should comply with regulations regarding cybersecurity, personal data protection, copyright, and information usage, while being mindful of the likelihood of infringement of intellectual property rights and moral rights [in the use of generative AI]. Agencies are advised to establish specific regulations or internal control measures for the use of generative AI based on the equipment and nature of their operations.
- Personnel of agencies shall not provide generative AI with data related to classified government information or information that requires consent from individuals, organizations, or any agencies for public disclosure. They should also refrain from asking generative AI questions that may involve sensitive matters or using generative AI to collect or process personal data.
- Agencies shall not solely rely on the information generated by generative AI, nor should they base administrative actions or official decisions solely on unverified output [of the generative AI].
- When using generative AI as a tool for executing affairs or providing service support, agencies should make appropriate disclosures (emphasis added).
As of the time of writing this article, there is no specific law or regulation in Taiwan that governs artificial intelligence (AI) or related legal issues. Nevertheless, on July 6, 2023, Vice Premier Mr. CHENG Wen-tsan announced that the government aims to present a draft act, possibly to be named “The Artificial Intelligence Basic Law,” to the legislature in September 2023.