雅各布·麦克惠尼/《圣迭戈之声》 今年早些时候，圣迭戈警察局因一项禁止警员使用人工智能撰写警务报告的政策而登上新闻头条。如今，该局却开始为使用此类技术敞开大门。
JAKOB MCWHINNEY/Voice of San Diego Earlier this year, the San Diego Police Department made headlines for a policy that prohibited officers from using artificial intelligence to produce police reports. Now, the department is opening the door to do just that.
7月31日，圣迭戈警察局发布了一份信息征集请求，称正在寻找一款“能借助人工智能技术协助警员及调查员撰写警务报告”的程序。该请求中写道：“此举旨在提升报告撰写效率，从而让警员能更迅速地响应各类服务请求。”
A request for information issued by the department on July 31 describes the department looking for a program “that utilizes AI technology to assist officers and investigators in writing police reports.”“The goal of this initiative is to enhance efficiency in report writing, thereby allowing officers to respond more quickly to calls for service,” the request for information reads.
这份信息征集请求还指出，相关程序必须确保使用者能够自主掌控“报告的内容与结构”。此外，该请求并非招标公告，发布目的“仅在于收集信息并用于规划工作”。文件预计该合同的价值将在50万美元至300万美元之间。
The request for information notes that the program must allow users to maintain control over the “content and structure of their reports.”It also notes that it is not a request for proposals and is being issued “solely for information gathering and planning purposes.”The document estimates the contract value to be somewhere between $500,000 and $3 million.
圣迭戈警察局发言人阿什利·尼科尔斯在邮件中强调，目前该局正处于“探索阶段”，旨在了解市场上有哪些相关技术以及此类技术的最佳应用方式。她表示，若最终决定引入人工智能辅助警务报告系统，警方会主动与民选官员及社区成员进行沟通。
In an email, SDPD spokesperson Ashley Nicholes emphasized that the department is just in an “exploration phase” during which they are working to learn what’s out there and what best practices for such technology is. Should they choose to pursue AI-assisted police report programs, they would conduct outreach to elected officials and community members, she said.
尼科尔斯写道：“鉴于当前一线警力短缺的现状，圣迭戈警察局始终在寻求提升整体工作效率的方法。随着人工智能技术的不断发展，这一技术或许能减少警员在部分事务上耗费的时间，从而让他们更快重返岗位，处理更多服务请求。”
“SDPD is constantly looking at ways to maximize efficiency throughout the department, particularly given current staffing shortages of front line officers. With the emergence of AI technology, this may be an option to reduce time spent on certain tasks to put officers back in the field responding to additional calls for service faster,” Nicholes wrote.
此次信息征集距离该局出台一项看似完全相悖的政策还不到一年。去年12月，圣迭戈警察局曾明令禁止警员使用人工智能平台生成警务报告。
The request comes less than a year after the department instituted a policy that, at first, seemed wholly contradictory. In December, the department forbade officers from using AI platforms to generate reports.
“该部门不授权使用任何生成式人工智能程序来撰写警方报告”，该备忘录写道，该备忘录最早由CBS 8报道。
“The Department does not authorize the use of any generative Artificial Intelligence programs to author a police report,” read the memo, first reported by CBS 8 .
政策变更发生在参议院法案524生效前不久。该法律要求全州警察部门开始制定规范人工智能使用的政策。它还要求部门披露是否使用人工智能撰写报告，并保留任何身份摄像头录像和音频，以便进行未来审计。
The policy change came shortly before Senate Bill 524 went into effect . The law required police departments statewide to begin to develop policies regulating AI use. It also required that departments disclose if AI was used to write reports and hold on to any body-cam footage and audio to allow for future audits.
但圣地亚哥警察部门的政策附带了一个巨大的星号：它仅禁止警员使用未获部门批准的人工智能平台。因此，看似全面禁止的政策实际上只是暂停，前提是该部门正式批准了一个人工智能平台。
But the San Diego Police Department’s policy came with a huge asterisk: it only prohibited officers’ use of non-department-approved AI platforms. So, what may have seemed like a blanket ban was really just a time-out, assuming the department officially approved an AI platform.
哈立德·亚历山大是圣地亚哥城市学院的教授，也是“社区支柱”组织的创始人兼总裁，该组织致力于支持受执法影响的圣地亚哥居民。在他看来，前景似乎全是坏的。
Khalid Alexander is a professor at San Diego City College and the founder and president of Pillars of the Community, an organization dedicated to supporting San Diegans impacted by law enforcement. To him, the prospect seems all bad.
“这只是他们试图从警务中夺走任何人性的又一个例子，”亚历山大说。
“It’s just one more example of them trying to take any kind of humanity away from policing,” Alexander said.
但对他来说，问题更深层次。他不仅担心装载着人们生活中最糟糕日子信息的系统是否真正安全，还担心对技术的过度依赖。
But it goes deeper than that for him. He not only worries about whether systems stuffed full of information about people on the worst days of their lives are actually secure, he also worries about over-reliance on the technology.
作为一名教师，他经常看到学生依赖人工智能完成课程作业。其产出充斥着错误、幻觉和废话。他设想，在执法环境中使用的通常存在缺陷的系统将带来更严重的后果。
As an instructor, he constantly sees students relying on AI to complete coursework. What it produces is work rife with errors, hallucinations and nonsense. He said he imagines a typically flawed system used in a law enforcement context will have even worse repercussions.
“问题不仅在于他们会使用它，而在于他们假设这项技术比人类更少偏见、更正确，”亚历山大说。
“The problem isn’t just that they’ll use it, it’s the assumption that the technology is less biased and more correct than human beings,” Alexander said.
目前已有少数地方执法机构开始引入人工智能技术。丘拉维斯塔警察局与圣地亚哥县警长办公室均采用了人工智能程序，这些程序能够转录民众拨打的非紧急求助电话并将其转接至相应处理部门，从而让接线员腾出精力处理更多紧急呼叫。
A handful of local law enforcement agencies have already leaned into AI. Both the Chula Vista Police Department and the San Diego County Sheriff’s Office utilize AI programs that transcribe residents’ non-emergency calls and route them to the appropriate place in a bid to free up dispatchers to take more emergency calls.
全面接纳此类技术的丘拉维斯塔警察局还利用人工智能来撰写警务报告。
Chula Vista, which has wholly embraced such technologies, also utilizes AI to write police reports.
“据估算，每位警员每班需要花费三到五个小时来撰写报告，”丘拉维斯塔警察局局长罗克萨娜·肯尼迪于11月对CBS8电视台说道，“如果这样能让警员更快重返执勤岗位，那这项技术就极具价值。”不过，执法部门使用人工智能一事长期以来饱受争议。诸多争议焦点在于人工智能平台中潜藏的种种偏见。随着人工智能日益融入日常生活，新的担忧也随之出现：比如人工智能倾向于迎合并强化用户固有观念的特性，是否会对警务调查产生负面影响。
“We’re estimating usually three to five hours per shift spent on report writing,” Chula Vista Police Department Chief Roxana Kennedy told CBS8 in November. “If that means that officer is going to get back out on the street quicker, it’s super important.”Still, law enforcement’s usage of AI has long sparked controversy. Many of those concerns have centered on the often-troubling biases programmed into AI platforms. But as AI platforms have become more integrated into everyday life, other worries have emerged, like whether the tendency of AI platforms to flatter and confirm the priors of their users can negatively impact police investigations.
事实上，已有大量案例表明人工智能平台在处理警务相关任务时频频出错，甚至导致无辜民众蒙受损失，尤其是在面部识别软件的应用方面。
Already, there are scores of examples of AI platforms bungling police-related tasks, leading to real-life consequences for innocent individuals , particularly when it comes to reliance on facial recognition software .
随着对Flock等人工智能监控摄像头的抵制声浪日益高涨，上述担忧更是被进一步放大。目前圣地亚哥市已布设了大量此类摄像头。去年，当地一名男子因Flock摄像头的错误识别而被误判为犯罪嫌疑人，最终在狱中关押了近一个月。
All of these concerns have been sent into overdrive by increasing pushback against AI-enabled surveillance cameras like Flock, of which San Diego has a robust network . Last year, one local man spent nearly a month behind bars after being arrested for a crime he didn’t commit thanks to a misidentification from a Flock camera.
利用人工智能撰写警务报告的做法也遭到了不少反对者与专家的批评。多项研究指出，从技术层面而言人工智能生成的文本质量确实更高，但未必能真正为警员节省时间。
The use of AI to create police reports has been panned by some opponents and experts . Studies have found that AI can produce writing that is better, at least technically, but may not actually save officers time.
最近一项研究显示，由人工智能生成的警方报告在准确性方面“明显逊色于人工撰写的报告”。这项研究由前警察、犯罪学教授伊恩·亚当斯主持。
One recent study found that AI-generated police reports were “rated substantively and significantly worse on accuracy,” according to Ian Adams , a former police officer and criminology professor who ran the study.
亚当斯在文章中写道：“这份报告的最终读者包括决定是否提起公诉的检察官、寻找报告漏洞的辩护律师，以及数月后需要还原事件真相的法官与陪审团。因此，准确性是撰写此类报告的最低要求。”该研究所使用的平台由Axon公司开发，这家企业专门为各警察局提供各类技术解决方案。事实上，圣地亚哥警局配备的随身摄像设备也由该公司生产，这样一来，若警方选择与Axon合作，就能实现设备与平台的顺畅对接乃至自动上传功能。
“For a document whose downstream readers are prosecutors deciding to charge, defense attorneys probing for inconsistencies, and judges and juries reconstructing events months later, accuracy is the threshold condition,” meaning it is a minimum requirement, Adams wrote in a post on The platform used in that study was created by Axon, a company that creates all sorts of technology for police departments. In fact, the body-worn cameras used by San Diego police officers are made by the company, allowing for easy integration and even automatic uploads should the department opt to contract with the corporation.
该平台还设有多项安全机制：所有书面报告均需人工审核确认，系统甚至会自动插入一些无实际意义的语句，以此确保报告确实经过了警员核对。
It also includes safety elements, like a feature that requires human input in all written reports and even manually inserts nonsense phrases to ensure reports have been reviewed by officers.
乔治华盛顿大学法学教授安德鲁·弗格森专门研究警方监控技术以及人工智能融入警务工作所带来的影响。在2025年发表于某法律期刊上的一篇关于人工智能生成警方报告的论文中，他写道：“对刑事司法体系而言，最大的风险在于算法会篡改事实叙述，进而导致案件审理过程中出现‘事实失真’的现象。”不过，弗格森并不认为自己是人工智能技术的彻底反对者——目前人工撰写的警方报告质量也远称不上完美，毕竟人类同样会犯错。但他确实持怀疑态度。
Andrew Ferguson is a law professor at George Washington University who specializes in police surveillance and the impacts of the integration of AI into policing. In a 2025 law review article on AI-produced police reports, he wrote “the danger for the criminal legal system is the digital poisoning of the facts in criminal trials by algorithmically altering the narrative.”Still, Ferguson doesn’t view himself as an out-and-out critic. The current reality of human-produced police reports is far from optimal. After all, humans produce errors as well. But he does count himself as a skeptic.
弗格森指出：“纵观警方技术的发展历程，大多数情况下都是先引入新技术，随后出现‘糟糕，我们制造了新问题’的情况，接着引发丑闻与媒体报道，最后才被迫撤回相关技术。”
“Because most of the history of police technologies has usually been one of adoption, ‘Oops, we created a problem,’ scandal, news report and then walking it back,” Ferguson said.
在他看来，人工智能必然会愈发深入地融入警务工作，这一点毋庸置疑。事实上，这一进程的推进速度已经超出了我们的预期。不过，相关官员与公众都必须明白：将原本由人类承担的工作转交给机器来完成，必然存在诸多潜在风险。
To him, there’s no question that AI will only become more integrated into policing. It’s already happening quicker than we realize. But officials and the public needed to understand there are inherent dangers in “moving a very human job to a less human mechanism.”
“问题在于我们会失去哪些人性化的要素？我们又该采取何种措施来确保技术不会出现实质性错误或问题呢？”弗格森说道，“最令我感到无奈的是，在引入相关技术之前，我们似乎从未认真探讨过这些关键问题。倘若我们事先能充分思考这些问题，或许就能在动用纳税人的钱来采购相关技术之前，避免一些显而易见的失误。” 本文最初由《圣迭戈之声》刊发，后经美联社合作转载。
“The question is what human aspect are you losing, and what do you do to make sure that there aren’t real errors or problems in the technology?”Ferguson said. “The thing that I find the most frustrating is that we never seem to ask these hard questions before we roll out the technology. If we were really asking the questions beforehand, we might have avoided some of the more obvious errors before using tax dollars to pay for this stuff.”This story was originally published by Voice of San Diego and distributed through a partnership with The Associated Press.