In response to the wave of AI development, the government has proposed the "Six Core Strategic Industries" plan. In line with the policy of the Executive Yuan, the Ministry of Justice intends to transform itself into a "Technological Ministry of Justice". This initiative envisages the use of AI in five key systems: procuratorates, investigation agencies, anti-corruption agencies, administrative enforcement and correctional institutions. The aim is to plan proactively for industrial development, social stability, national security and the deepening of democracy.

In connection with the prevention of drug crime, the Executive Yuan has introduced the " New Generation Anti-Drug Strategy Action Plan 2.0" under the "Comprehensive Planning Strategy" The "Recidivism Prevention Promotion Plan" will be implemented, focusing on "comprehensive protection" measures". The focus is on dealing with the high recidivism (relapse) rate among drug users and their needs for social reintegration. Based on scientific findings, the goals of "curbing drug recidivism" and "reducing new drug cases" are to be achieved. It can be seen that the early use of AI technology not only creates more opportunities for collaboration between industry and academia, but can also harness the potential of drug control research during Taiwan's crucial AI transformation.

This research, initiated in response to the Executive Yuan's "Six Core Strategic Industries" and the Ministry of Justice's "Technological Ministry of Justice" policy, began in 2021 with the pilot study of "Application of Natural Language Analysis to Indictments for Drug Use Offenses". Research has found that the use of artificial intelligence to interpret prosecution documents is highly viable, achieving a 90% accuracy rate in identifying key variables related to drug use prosecutions. Each type of crime has its own aspects, and preliminary research has highlighted the specificity of prosecution documents in the field of drug crime.

To improve the applicability and scope of AI models in the drug crime field, this project under the framework of the Ministry of Justice's "Recidivism Prevention Advancement Plan," is conducting a multi-year study, including all prosecution documents nationwide. The first phase focuses on analyzing the natural language used in the prosecution of drug use offenses and deferred prosecution. The research methodology includes annotating the text of criminal drug charges, developing an automated annotation system, and improving the machine's ability to interpret various criminal drug charges with scientific knowledge.

The research results include:
1. Establishment of the first phase of the "AI Drug Use Crime Big Data Application and Analysis System", expansion of the associated hardware and cybersecurity infrastructure.
2. Cleaning and evaluation of the "Criminal Policy and Crime Research Database", development of an interface for one-click import and visualization to improve basic research on drug prevention and recidivism.
3. Developing interfaces for annotating drug use crime prosecution and deferred prosecution documents to assist AI in learning legal language.
4. Used natural language algorithms to interpret drug use crime prosecutions and deferred prosecution documents, achieving an 80% accuracy rate in generating coded results needed for crime research.
5. Constructing machine learning models to distinguish between prosecution and deferred prosecution with a 90% accuracy rate and over 95% classification correctness.
6. Investigated key nodes in the classification of prosecution and deferred prosecution using decision tree analysis, identifying mandatory treatment records, repeat offenses, and regional medical resources as critical nodes.
7. Analysis of 168,665 drug offenders, finding that 53-65% of drug users do not commit property or violent offenses after their first offense, with most continuing drug use.
8. Analysis of the 5-year recidivism rate of deferred prosecutions with compulsory addiction treatment from the start year of the implementation of the New Generation Anti-Drug Strategy 1.0 (2017-2020), which shows a 10 percent reduction in the 5-year recidivism rate compared to 2008-2016, indicating the initial effectiveness of the long-term anti-drug policy.

For the full paper, please visit here (only in traditional Chinese version).