一串六个大规模枪击事件在不到两周的时间里，加利福尼亚州发生了30人死亡，19人受伤的事件，重新引发了美国解决枪支暴力问题的呼吁。

总统乔·拜登本月早些时候推在全国范围内禁止突击步枪;而反对这一措施的共和党人在袭击发生后基本上保持沉默。针对其他大规模枪击事件，共和党人呼叫改善精神健康服务。

国会的僵局和明显无效加州健全的州枪支法让人们寻找替代方案。一个相对较新的潜在解决方案，使用人工智能增强的安全性，因其在开枪前逮捕枪手的承诺而引起了人们的兴趣。

人工智能安全行业吹捧能够识别携带武器在学校外游荡的嫌疑人的摄像头，能够发现隐藏枪支的高科技金属探测器，以及分析信息以标记潜在大规模枪击者的预测算法。

开发人工智能增强安全摄像头的公司官员表示，这项技术可以纠正容易出错的安全官员，他们表示，这些安全官员经常难以监控多个视频源，并找出新出现的威胁。相反，公司官员表示，人工智能可以在攻击者准备攻击时可靠地识别他们，为安全官员节省宝贵的几分钟或几秒钟，并可能挽救生命。

“这是保守得最好的秘密，”人工智能安全公司ZeroEyes的联合创始人萨姆·阿莱莫告诉美国广播公司新闻。“如果学校外面有突击步枪，人们会想知道更多信息。如果拯救了一条生命，那就是胜利。”

然而，批评者质疑产品的有效性，称公司未能提供关于准确性的独立验证数据。他们补充说，即使人工智能有效工作，该技术也引起了对隐私侵犯和潜在歧视的主要担忧。

“如果你打算用你的隐私和自由来换取安全，你需要问的第一个问题是:你得到了一笔好交易吗?”美国公民自由联盟演讲、隐私和技术项目的高级政策分析师杰伊·斯坦利告诉美国广播公司新闻。

人工智能安全市场

随着学校、零售商和办公室考虑采用人工智能安全，该行业正蓄势待发。根据研究公司的数据，探测隐藏武器的产品市场预计将从2022年的6.3亿美元增加到2031年的12亿美元未来市场洞察。

这种乐观情绪部分归因于安全摄像头的日益普及，允许人工智能公司销售软件，增强许多建筑中已经使用的系统。

截至2017-18学年，83%的公立学校表示他们使用安全摄像头国家教育统计中心找到了。该组织的调查显示，这一数字与1999-2000学年相比有显著上升，当时只有19%的学校配备了安全摄像头。

“我们与现有的监控系统合作，”人工智能安全公司Scylla的销售副总裁克里斯·格雷纳(Kris Greiner)告诉美国广播公司新闻。“我们只是给它一个大脑。”

致力于人工智能安全以防止枪击事件的公司

格雷纳说，总部位于德克萨斯州奥斯汀的公司Scylla于2017年推出，它提供的人工智能不仅可以帮助安全摄像头检测隐藏的武器，还可以检测可疑的活动，例如规避安全措施或挑起争斗。

他说，当全自动系统识别出武器或可疑行为者时，它会通知学校或企业的官员，并指出大规模射击者通常会在进入设施前拔出枪。他说，该系统还可以设置为立即拒绝进入并锁门。

格雷纳说:“在分秒必争的时刻，它很可能会产生重大影响。”

他补充说，该公司已经在33个国家进行了大约300次安装，允许客户机构克服安全官员的常见缺点。

“想象一下，一个人坐在指挥中心看视频墙，这个人只能看4到5个摄像头4到5分钟，然后他就会开始错过东西，”Greiner说。“人工智能可以观看的内容没有限制。”

另一家人工智能安全公司ZeroEyes提供类似的人工智能增强视频监控，但目的更窄:枪支检测。

ZeroEyes的联合创始人阿莱莫表示，该公司由前海豹突击队(Navy Seals)在2018年成立，其创始人之一意识到，安全摄像头提供了事后证明大规模枪击事件的证据，但在防止暴力方面几乎没有做什么。

“在大多数情况下，枪手在扣动扳机前已经把枪露出来了，”阿拉莫说。“我们想得到那把枪的图像，并用它警告第一反应人员。”

与Scylla的产品一样，ZeroEyes AI可以跟踪实时视频，并在检测到枪支时发出警报。然而，ZeroEyes的警报会传到内部控制室，公司员工在那里确定情况是否构成真正的威胁。

“我们有一个人在循环中，以确保客户永远不会得到误报，”阿拉莫说，并补充说，从警报到验证再到与客户沟通，整个过程只需三秒钟。

人工智能安全中的准确性

美国公民自由联盟的斯坦利说，人工智能增强的安全性在理论上听起来像是一个潜在的救命突破，但产品的准确性仍然不确定。“如果没有效果，就没有必要讨论隐私和安全，”他说。“谈话应该结束了。”

Scylla的Greiner表示，该公司的人工智能在识别枪支等武器方面的准确率为99.9%——“有很多9”。但他没有说明该系统在识别可疑活动方面的准确性，并表示该公司尚未对该系统的准确性进行独立验证。

“对我们来说，寻找第三方来做这件事-我们还没有这样做，”Greiner说，并补充说，该公司允许客户在购买前测试产品。

ZeroEyes的Alaimo表示，该公司将员工验证作为其警报流程的一部分，可以消除误报。但他拒绝透露人工智能系统向员工提供误报的频率，或者员工是否在评估警报时出错。

斯坦利说:“透明度是关键，因为如果社区打算尝试就他们是否希望在这些公共场所使用这些技术做出有希望的民主决定，他们需要知道这是否值得。”

对人工智能的其他担忧

抛开这些系统的功效不谈，批评者对侵犯隐私以及人工智能可能带来的歧视表示担忧。

首先，超过30个州允许人们公开携带手枪，让这些人成为人工智能增强安全的潜在目标。

“现在在这个国家的大多数地方，携带枪支是合法的，”纽约大学研究人工智能伦理的法学教授巴里·弗里德曼告诉美国广播公司新闻。“很难知道你要以不侵犯人们权利的方式搜索什么。”

阿拉莫说，在ZeroEyes，人工智能在个人合法持有枪支的情况下向客户发出“非致命警报”，让客户意识到武器，但没有紧急反应。

注意到其他隐私问题，斯坦利说，安全官员从来不看目前记录的绝大多数监控录像，除非可能的犯罪已经发生。然而，人工智能的算法会扫描每一分钟的可用镜头，在某些情况下，会观察被认为可疑或不寻常的活动。

“这真让人毛骨悚然，”斯坦利说。

鉴于面部识别系统评估中发现的种族歧视，斯坦利警告说，人工智能可能会遭受同样的问题。弗里德曼补充说，这个问题有在更广泛的刑事司法系统中复制种族不平等的风险。

“当我们没有准备好使用这些工具时，使用它们的代价是人们的生活将被撕碎，”弗里德曼说。"人们会在不应该成为执法目标的时候被认定为目标。"

Greiner和Alaimo表示，他们的人工智能系统不会评估安全信息中显示的个人种族。“我们不以种族、性别、民族来识别个人，”格雷内尔说。"我们从字面上把人定义为拿着枪的人."

阿拉莫表示，如果美国放弃人工智能解决方案，可能会面临不必要的悲剧，特别是因为其他修复措施需要更长的时间。

“我们可以也应该继续谈论心理健康。我们可以也应该继续讨论枪支法，”阿莱莫说。“我关心的是今天——不是一年后，也不是10年后，那时我们可能会找到那些更困难问题的答案。”

“我们现在做的是一个解决方案，”他说。

Can artificial intelligence help stop mass shootings?

A string of sixmass shootingsin California over less than two weeks, which left 30 dead and 19 injured, has reignited calls that the U.S. address gun violence.

PresidentJoe Bidenearlier this monthpushedfor a nationwide ban on assault rifles; while Republicans, who oppose such a measure, have remained largely silent in the aftermath of the attacks. In response to other mass shootings, Republicans havecalledfor improved mental health services.

The congressional gridlock andapparent ineffectivenessof California's robust state gun laws have left people searching for alternatives. One relatively new potential solution, the use of artificial intelligence-enhanced security, has drawn interest for its promise of apprehending shooters before a shot is fired.

The AI security industry touts cameras that identify suspects loitering outside of a school with weapons, high-tech metal detectors that spot hidden guns, and predictive algorithms that analyze information to flag a potential mass shooter.

Company officials behind the development of the AI-boosted security cameras say the technology corrects for fallible security officers, who they said often struggle to monitor multiple video feeds and suss out emerging threats. Instead, company officials say, AI reliably identifies assailants as they ready for an attack, saving security officials precious minutes or seconds and possibly saving lives.

"This is the best-kept secret," Sam Alaimo, co-founder of an AI security company called ZeroEyes, told ABC News. "If there's an assault rifle outside of a school, people want to know more information. If one life is saved, that's a victory."

Critics, however, question the effectiveness of the products, saying companies have failed to provide independently verified data about accuracy. Even if AI works effectively, they add, the technology raises major concerns over privacy infringement and potential discrimination.

"If you're going to trade your privacy and freedom for security, the first question you need to ask is: Are you getting a good deal?" Jay Stanley, senior policy analyst at the ACLU Speech, Privacy, and Technology Project, told ABC News.

The market for AI security

As schools, retailers and offices consider adopting AI security, the industry is poised for growth. The market for products that detect concealed weapons is expected to nearly double from $630 million in 2022 to $1.2 billion by 2031, according to research firmFuture Market Insights.

The optimism owes in part to the increased prevalence of security cameras, allowing AI companies to sell software that enhances the systems already in use at many buildings.

As of the 2017-18 school year, 83% of public schools said they use security cameras, theNational Center for Education Statisticsfound. The figure marked a significant uptick from the 1999-2000 school year, when just 19% of schools were equipped with security cameras, the organization's survey said.

"We work with an existing surveillance system," Kris Greiner, vice president of sales at AI security company Scylla, told ABC News. "We just give it a brain."

Companies working on AI security to prevent shootings

Scylla, an Austin, Texas-based company launched in 2017, offers AI that aids security cameras in not only detecting concealed weapons but also suspicious activity, such as efforts to circumvent security or start a fight, Greiner said.

When the fully automated system identifies a weapon or suspicious actor, it notifies officials at a school or business, he said, noting that mass shooters often draw their gun prior to entering a facility. The system can also be set to immediately deny access and lock doors, he said.

"At a time when every second counts, it's quite possible that it would have a heavy impact," Greiner said.

The company, which has performed about 300 installations across 33 countries, allows client institutions to overcome the common shortcomings of security officers, he added.

"Imagine a human sitting in a command center watching a video wall, the human can only watch four to five cameras for four to five minutes before he starts missing things," Greiner said. "There's no limit to what an AI can watch."

Another AI security company, ZeroEyes, offers similar AI-enhanced video monitoring but with a narrower purpose: Gun detection.

The company, launched by former Navy Seals in 2018, entered the business after one of its founders realized that security cameras provided evidence to convict mass shooters after the fact but did little to prevent violence in the first place, said Alaimo, ZeroEyes Co-founder.

"In the majority of cases, the shooter has a gun exposed before squeezing the trigger," Alaimo said. "We wanted to get an image of that gun and alert first responders with it."

As with Scylla's product, the ZeroEyes AI tracks live video feeds and sends an alert when it detects a gun. However, the alert at ZeroEyes goes to an internal control room, where company employees determine whether the situation poses a real threat.

"We have a human in the loop to make sure the client never gets a false positive," Alaimo said, adding that the full process takes as little as three seconds from alert to verification to communication with a client.

Accuracy in AI security

AI-enhanced security sounds like a potentially life-saving breakthrough in theory, but the accuracy of the products remains uncertain, said Stanley, of the ACLU. "If it isn't effective, there's no need to get into a conversation about privacy and security," he said. "The conversation should be over."

Greiner, of Scylla, said the company's AI is 99.9% accurate -- "with a whole lot of 9s" -- in identifying weapons, such as guns. But he did not say how accurate the system is in identifying suspicious activity, and said the company has not undergone independent verification of the system's accuracy.

"For us to search for a third party to do it -- we haven't done that yet," Greiner said, adding that the company allows customers to test the product before purchasing it.

Alaimo, of ZeroEyes, said the company's use of employee verification as part of its alert process allows for the elimination of false positives. But he declined to say how often the AI system presents employees with false positives or whether employees make mistakes in assessing the alerts.

"Transparency is key, because if communities are going to try to make hopefully democratic decisions about whether or not they want these technologies in these public places, they need to know whether it's worth it," said Stanley.

Other concerns about AI

Setting aside the efficacy of the systems, critics have raised concerns over privacy infringement as well as potential discrimination enabled by AI.

To start, more than 30 states allow people to openly carry handguns, leaving such people as potential targets of AI-enhanced security.

"Carrying a gun is lawful in most places in this country now," Barry Friedman, a law professor at New York University who studies AI ethics, told ABC News. "It's very hard to know what you're going to search for in a way that doesn't impinge on people's rights."

At ZeroEyes, the AI sends out a "non-lethal alert" to clients in cases where an individual is lawfully holding a gun, making the client aware of the weapon but stopping short of an emergency response, Alaimo said.

Noting additional privacy concerns, Stanley said security officials never watch the vast majority of surveillance footage currently recorded, unless a possible crime has taken place. With AI, however, an algorithm scans every minute of available footage and, in some cases, watches for activity deemed suspicious or unusual.

"That's pretty spooky," Stanley said.

In light of racial discrimination found in assessments made by facial recognition systems, Stanley cautioned that AI could suffer from the same issue. The problem risks replicating racial inequity in the wider criminal justice system, Friedman added.

"The cost of using these tools when we're not ready to use them is that people's lives will be shredded," Friedman said. "People will be identified as targets for law enforcement when they should not be."

For their part, Greiner and Alaimo said their AI systems do not assess the race of individuals displayed in security feeds. "We don't identify individuals by race, gender, ethnicity," Greiner said. "We literally identify people as a human who is holding a gun."

Alaimo said the U.S. could face unnecessary tragedy if it forgoes AI solutions, especially since other fixes operate on a longer time horizon.

"We can and should keep talking about mental health. We can and should keep arguing gun laws," Alaimo said. "My concern is today -- not a year from now, not 10 years from now, when we might have answers to those more difficult questions."

"What we're doing is a solution right now," he said.