While using statistics in law enforcement is nothing new,1 cutting-edge technology that uses big data is changing the face of law enforcement (Buchholtz 2020; Degeling and Berendt 2018; Egbert and Krasmann 2019; Ferguson 2017b; Sheehey 2019; Nissan 2017; Perry et al. 2013). As data and statistical tools have improved over time, a new police strategy called “predictive policing” (hereafter, PP) has come into practice (Saunders et al. 2016; Kreutzer and Sirrenberg 2020; Kulkarn and Akhilesh 2020). Based on the assumptions that certain aspects of the physical and social environment encourage predictable acts of criminal wrongdoing, and that interfering with that environment would deter the would-be crimes, PP aims to “forecast where and when the next crime or series of crimes will take place” by identifying trends and relationships that may not be readily apparent to us among the collected data (Uchida 2014, p. 3871; see also Ferguson 2017a; Moses and Chan 2018). Techniques involving large quantities of digital information have been evolving at a rapid rate. As Ferguson (2017a) notes, while the social scientific research supports the insights behind PP, police adoption of the strategy has outpaced established scientific findings. More significantly, when a police department declares its adoption of PP, it could be doing things that vary greatly in their technical sophistication, effectiveness, and ethical concerns. As such, while there is an increasingly heated debate about the effectiveness and potential impacts of the emerging techniques involving large quantities of digital information, the discussion is easily conducted without careful awareness of the differences among various methods and practices of PP (Egbert and Krasmann 2019; Ferguson 2017b). Besides, myths and pitfalls may hinder proper evaluation of PP’s development and deployment, such as assuming that AI actually knows the future, or focusing on prediction accuracy rather than tactical utility (Perry et al. 2013). As a proactive policing model, the targeted units of crime predictions of PP can range from different sizes of geographical areas to individual people. Based on its focuses, PP can be divided into three subdivisions:

中文翻译：

---

关于基于人的AI预测策略

虽然在执法机构中使用统计数据并不是什么新鲜事，但使用大数据的尖端技术正在改变执法机构的面貌（Buchholtz 2020; Degeling and Berendt 2018; Egbert and Krasmann 2019; Ferguson 2017b; Sheehey 2019; Nissan 2017; Perry等人，2013年）。随着数据和统计工具的不断完善，一种称为“预测性警务”（以下简称PP）的新警务策略已付诸实践（Saunders等人，2016； Kreutzer和Sirrenberg 2020； Kulkarn和Akhilesh 2020）。基于以下假设，即物质和社会环境的某些方面鼓励可预见的犯罪行为，而干扰该环境将阻止可能的犯罪，PP的目的是通过识别收集到的数据中我们可能不容易发现的趋势和关系，“预测下一次犯罪或一系列犯罪将在何时何地发生”（Uchida 2014，p。3871；另见Ferguson 2017a； Moses和Chan 2018）。涉及大量数字信息的技术正在迅速发展。正如Ferguson（2017a）指出的那样，尽管社会科学研究支持PP背后的见解，但警方采用该策略的速度已经超过了既定的科学发现。更重要的是，当警察部门宣布采用PP时，它所做的事情在技术上的复杂性，有效性和道德方面的差异可能很大。因此，尽管有关涉及大量数字信息的新兴技术的有效性和潜在影响的辩论日益激烈，但无需仔细了解PP的各种方法和实践之间的差异，即可轻松进行讨论（Egbert和Krasmann 2019; Ferguson 2017b ）。此外，神话和陷阱可能会妨碍对PP的发展和部署进行正确的评估，例如假设AI确实了解未来，或者侧重于预测的准确性而不是战术的实用性（Perry等人，2013）。作为一种积极的警务模型，PP犯罪预测的目标单位范围可以从不同的地理区域到个人。根据其关注点，PP可以分为三个细分：讨论很容易进行，无需仔细了解PP的各种方法和实践之间的差异（Egbert和Krasmann 2019; Ferguson 2017b）。此外，神话和陷阱可能会妨碍对PP的发展和部署进行正确的评估，例如假设AI确实了解未来，或者侧重于预测的准确性而不是战术的实用性（Perry等，2013）。作为一种积极的警务模型，PP犯罪预测的目标单位范围可以从不同的地理区域到个人。根据其关注点，PP可以分为三个细分：讨论很容易进行，无需仔细了解PP的各种方法和实践之间的差异（Egbert和Krasmann 2019; Ferguson 2017b）。此外，神话和陷阱可能会妨碍对PP的发展和部署进行正确的评估，例如假设AI确实了解未来，或者侧重于预测的准确性而不是战术的实用性（Perry等，2013）。作为一种积极的警务模型，PP犯罪预测的目标单位范围可以从不同的地理区域到个人。根据其关注点，PP可以分为三个细分：例如假设AI确实了解未来，或者专注于预测准确性而不是战术实用性（Perry等人，2013）。作为一种积极的警务模型，PP犯罪预测的目标单位范围可以从不同的地理区域到个人。根据其关注点，PP可以分为三个细分：例如假设AI确实了解未来，或者专注于预测准确性而不是战术实用性（Perry等人，2013）。作为一种积极的警务模型，PP犯罪预测的目标单位范围可以从不同的地理区域到个人。根据其关注点，PP可以分为三个细分：