刑事程序中人工智慧於風險評估上的應用

AI Risk Assessment in Criminal Justice

由於人工智慧技術日趨成熟，部分國家（如美國）已經開始在刑事程序中使用風險評估系統，預測特定人可能存有的風險，決定是否拘禁其人身自由或釋放條件。不過，建置風險評估系統的每一個環節中都有著人的主觀及價值判斷。再者，因為人工智慧技術的本質，風險評估系統會有黑盒運作的疑慮。是故，我們建議，若是要在刑事程序中使用風險評估系統，應該要遵循透明性的要求。此外，系統的使用也必須要具備其當責性，以確保系統達到其所設定的目標、結果的正確性及保障人民的人身自由等權益。再者，系統的評估結果不應作為刑事程序中決定的決定性因素，且應審視系統的建置目的與程序的規範目的是否一致。最後，風險評估系統的使用必須要避免自動化依賴的情形發生。

Due to the increasing maturity of artificial intelligence (AI), several countries (e.g., the United States) have applied such technology to construct risk assessment system as a way to predict certain people’s risk of committing crimes and decide whether they should be detained or released. After thorough analysis, however, it is clear that the construction of risk assessment is not as just or objective as people might think. People’s subjective decisions and value judgment exist in every stage of the construction of system. In addition, due to the nature of AI, the decision-making processes of the risk assessment system is often unclear, meaning that the system lacks transparency. Therefore, we suggest that such risk assessment system must comply with the transparency requirement, and the use of system should be in line with the accountability requirement to ensure the system meets the set goal, the accuracy of the of its outcome, and people’s rights. Furthermore, the outcome of the risk assessment system should not be the determinative factor of the decisions in criminal procedure, and whether the function of the system is consistent with the purpose of the process should be verified. Finally, there should be effective mechanism to avoid automation complacency.