형사정책시스템 알고리즘 구축 과정에서의 편향성

The Deviation and Responses in the Construction of Criminal Policy System Algorithm

자동화된 의사결정 형태로 발전 운용하고자 하는 형사사법 알고리즘의 효율성과 효과성을 파악하며, 알고리즘의 부정적 효과라 이해하는 편향성(bias)을 줄이기 위한 논의가 필요하다. 예측적 경찰활동이란, 능동적 범죄예측의 방식에서 현장 정보나 일선 관련 정보가 아닌 순수 사회 데이터를 활용해 통계적 예측을 통해 범죄 예방 또는 경찰의 개입 대상을 결정하는 것이다. DAS, PredPol, COMPAS 등의 시스템이 존재하나, 국내는 현재까지 진화된 방식의 기술적 논의와 테스트 단계일 뿐 법·제도적으로 갖추어야 할 요소와 알고리즘 구축에 대한 논제가 많다. 그런데도 쟁점은 기술 발달 측면에 가려져 있다. 따라서 기술적 측면을 배제하고 인문사회과학 영역에서 형사사법 알고리즘 시스템 도입에 선두 국가인 미국의 사례를 통해 그 문제점과 쟁점을 파악하는 것 또한 중요하다. 예측 치안을 어우르는 형사정책 분야에서 알고리즘의 자동화된 판단이 수반하는 편향성을 검토하고, 문제를 해결하기 위한 규범적 측면과 자료수집과 활용 측면에서 세밀한 보완과 구축의 전제적 필요성을 구상해보고자 한다.

A discussion is urgently needed to identify the efficiency and effectiveness of the criminal justice algorithm, which is intended to develop and operate in an automated decision-making form and to reduce the bias that is understood as a negative effect of the algorithm. In the method of Active Crime Prediction, which comprehensively analyzes current data and accessible public data and transmits the predicted and analyzed information to the on-site police in real-time, it utilizes pure social data rather than on-site information or front-line information. It is defined as a predictive police activity that determines the target of crime prevention or police intervention through statistical prediction and is facing a change to the establishment of criminal justice procedures through an algorithmic system as a type. Overseas, systems such as DAS, PredPol, and COMPAS are being produced and operated by private companies. However, in Korea, it is only at the stage of technical discussion and testing of the evolved method so far, and there are many issues and issues regarding the elements and algorithms that must be equipped legally and institutionally. Nevertheless, the realm of these debates and issues is obscured by technological advances. Therefore, it is important to exclude technical aspects and identify the problems and issues through the examples of the United States and the United Kingdom, which are leading countries in introducing criminal justice algorithm systems in the humanities and social sciences. In this study, the problems that appeared through previous studies and cases were classified and explored according to the system operation stage (construction, utilization, management). The various issues and problems analyzed are ultimately problems that appeared in the use of the algorithm. It also had a structure that reduced it to a problem created by the person using the technology. It is interpreted as a limitation in that it cannot provide a clear solution to the ongoing debate or reflect it in the field. Therefore, in the field of criminal policy dealing with predictive policing and recidivism prediction, we examine the bias accompanying the automated judgment of algorithms, and envision the prerequisite for detailed complementation and construction in terms of normative aspects and data collection and utilization to solve problems.