Fact-based AI sentencing information system, applicable to drunk driving, hit and run, fraud, injury and other cases.

(Photographed by reporter Yang Guowen)

[Reporter Yang Guowen/Taipei Report] The Judicial Yuan stated today that in order to respond to the expectations of all walks of life on the appropriateness of sentencing, and in response to the fact that the new system of national judges has been launched in January this year, the AI ​​sentencing information system will be fully launched today (6th). With the assistance of information power, a large amount of judgment data is sorted out and updated in real time, providing professional judges and national judges with the opportunity to inquire about the sentencing trends of similar cases, making sentencing more fair and transparent.

The Judicial Yuan pointed out that after the improvement of the sentencing information system began in September 2020, the focus of this improvement is to use digital technology to build an AI sentencing information system. In the labeling process, relevant quantitative criminal reasons are marked, and the application of the aggravated and lightened laws is searched and interpreted by automatically extracting keywords. In the labeling process, it is used as training data for machine learning instead of manual labeling, so as to achieve real-time update of data effect of the library.

Please read on...

The Judicial Yuan stated that the AI ​​sentencing information system launched today is divided into a fact-based model and an evaluation-based model.

The evaluation mode selects the sentencing factors from the content of the criminal facts identified in the judgment and summarizes them. Users can use the factual content of individual cases to inquire.

The types of crimes in this mode include drunk driving, fraud, theft, injury, hit and run and other cases.

The evaluative model starts from the reasons for sentencing consideration explained in the judgment, and then summarizes the reasons for sentencing consideration in Article 57 of the Criminal Law. In summary, users can use the sentencing conditions of individual cases to inquire.

The types of crimes in this model include cases of guns, drugs, interference with sexual autonomy, etc.

The Judicial Yuan emphasized that the AI ​​sentencing information system has four major advantages, including: 1. Real-time update of data: the judgment is newer than that of the first-generation sentencing information system, and can be continuously updated to the latest judgment information; Adjustment: Add options for sentencing data or reduce sentencing factors, list sentencing factors commonly used in judgments, and reduce the situation of too few search results or no search results due to too fine distinction of sentencing factors; 3. Provide advanced search: after querying, you can You can see the sentencing distribution range under the influence of the selected sentencing factor, and you can directly query the content of the judgment.

The last advantage is "smart learning, smart labeling": the use of semantic labeling, according to the text before and after the sentencing factor and the writing method of aggravating and mitigating the law, is conducive to the development of AI automatic labeling function, and setting can modify the labeling content in time In the background, and retrain AI automatic labeling with the results of modified labeling, which helps to continuously update data and reduce errors.

The Judicial Yuan emphasized that in order to improve the fairness, appropriateness, transparency and predictability of sentencing, and enhance the public's trust in the judiciary, it will continue to use digital technology to update and optimize the sentencing information system.

Evaluative sentencing information system, suitable for guns, drugs, and obstructive autonomy cases.

(Photographed by reporter Yang Guowen)