《经济学人》前美国国安高官:AI怀疑论者看错了威胁,它其实更
共 1 集
剧集目录
| 集 | 标题 | 时长 |
|---|---|---|
| P1 | 《经济学人》前美国国安高官:AI怀疑论者看错了威胁,它其实更 | 8:04 |
P1 · 《经济学人》前美国国安高官:AI怀疑论者看错了威胁,它其实更能保卫脆弱的现状 [C1] p01 原声 (P1)
字幕摘录
| 时间 | 英文 | 中文 |
|---|---|---|
| 0:07 | Anne Neuberger | 安妮·诺伊伯格 |
| 0:09 | AI sceptics overlook how vulnerable the status quo is and how much the technology can help defend it. | AI怀疑论者忽略了现状的脆弱程度以及技术能帮助捍卫现状的程度. |
| 0:17 | Anne Neuberger is a general partner and the head of global affairs at Andreessen Horowitz. | Anne Neuberger是安德烈森·霍罗维茨的总合伙人和全球事务负责人. |
| 0:23 | She previously served as Deputy National Security Advisor of the United States. | 她此前曾担任美国国家安全副顾问. |
| 0:29 | Much of the debate about artificial intelligence, including a recent leader in The Economist, | 许多关于人工智能的辩论, 包括最近一位《经济学家》的领袖, |
| 0:35 | asks how to prevent a powerful new technology from making the world more dangerous. | 问如何防止强大的新技术使世界更加危险。 |
| 0:41 | There is an equally important, more urgent question. | 有一个同样重要、更紧迫的问题。 |
| 0:46 | How quickly can we use it to make an already dangerous world safer? | 我们如何能利用它使已经危险的世界更安全? |
| 0:52 | The world that large language models entered was very vulnerable. | 大语言模型进入的世界非常脆弱。 |
| 0:57 | In 2017, NotPetya, a Russian cyber attack aimed at Ukraine, | 2017年,NotPetya,俄罗斯网络攻击针对乌克兰, |
展开字幕全文(90 条)
| 序号 | 英文 | 中文 |
|---|---|---|
| 1 | Anne Neuberger | 安妮·诺伊伯格 |
| 2 | AI sceptics overlook how vulnerable the status quo is and how much the technology can help defend it. | AI怀疑论者忽略了现状的脆弱程度以及技术能帮助捍卫现状的程度. |
| 3 | Anne Neuberger is a general partner and the head of global affairs at Andreessen Horowitz. | Anne Neuberger是安德烈森·霍罗维茨的总合伙人和全球事务负责人. |
| 4 | She previously served as Deputy National Security Advisor of the United States. | 她此前曾担任美国国家安全副顾问. |
| 5 | Much of the debate about artificial intelligence, including a recent leader in The Economist, | 许多关于人工智能的辩论, 包括最近一位《经济学家》的领袖, |
| 6 | asks how to prevent a powerful new technology from making the world more dangerous. | 问如何防止强大的新技术使世界更加危险。 |
| 7 | There is an equally important, more urgent question. | 有一个同样重要、更紧迫的问题。 |
| 8 | How quickly can we use it to make an already dangerous world safer? | 我们如何能利用它使已经危险的世界更安全? |
| 9 | The world that large language models entered was very vulnerable. | 大语言模型进入的世界非常脆弱。 |
| 10 | In 2017, NotPetya, a Russian cyber attack aimed at Ukraine, | 2017年,NotPetya,俄罗斯网络攻击针对乌克兰, |
| 11 | spread worldwide within hours, deleting data and paralyzing systems, | 在数小时内向全世界传播,删除数据和瘫痪系统, |
| 12 | including those of Maersk, which carries about a fifth of global shipping. | 包括梅尔斯克的船,它承载着大约五分之一的全球航运。 |
| 13 | In 2023, Iranian-linked hackers broke into American water utilities, | 2023年,与伊朗有联系的黑客闯入美国供水公司, |
| 14 | forcing operators to switch to manual control. | 强制操作员切换到手动控制. |
| 15 | More worrying still are the attacks not yet triggered. | 更令人担忧的是,袭击尚未启动。 |
| 16 | Chinese state hackers have planted footholds in American power, water and transport systems, | 中国的州黑客在美国的电力、水和交通系统 扎下了根基 |
| 17 | ready to strike silently during a conflict. | 准备在冲突期间无声攻击。 |
| 18 | Few of these attacks relied on brilliant new exploits, | 这些攻击中很少有依靠 辉煌的新利用, |
| 19 | code that weaponizes systems flaws to cause harm. | 将系统缺陷武器化以造成伤害的代码。 |
| 20 | They succeeded because the defenders could not find, fix or fence off | 他们成功是因为维权者找不到、修补或封锁 |
| 21 | known weaknesses fast enough across sprawling aging systems. | 已知的弱点 足够快地跨过扩展的老化系统。 |
| 22 | The cost of that status quo belongs in any honest accounting of AI risk. | 维持现状的代价在于对AI风险的任何诚实的核算。 |
| 23 | I have worked on both sides of cyber conflict, | 我为网络冲突双方工作过 |
| 24 | conducting intelligence operations at America's National Security Agency, | 在美国国家安全局进行情报活动, |
| 25 | then running its cyber security directorate to turn those lessons into defenses. | 然后管理网络安全局 把这些教训变成防御 |
| 26 | Later at the White House, I pushed through the first mandatory cyber security rules | 后来在白宫 我推了第一个强制性的网络安全规则 |
| 27 | for America's pipelines, railways and airports. | 为了美国的管道,铁路和机场。 |
| 28 | The experience taught me something important. | 经历教会了我一些重要的事情. |
| 29 | AI can help defenders even more than it can help attackers. | 大赦国际能够帮助维权者比帮助攻击者还要多。 |
| 30 | True, AI makes the challenge harder in some ways. | 的确,大赦国际在某些方面使挑战变得更加困难。 |
| 31 | It lowers the skill needed by attackers | 它降低了攻击者所需的技能 |
| 32 | and lets them jiggle every digital doorknob continuously and at scale. | 让他们不停地和大规模地 抖动每一个数字门把手 |
| 33 | It is already speeding up the discovery of new flaws, | 它已经在加速发现新的缺陷, |
| 34 | including in software foundational to the Internet. | 包括互联网基础软件。 |
| 35 | But the old rule of cyber conflict, that defenders must be strong everywhere | 但网络冲突的旧规则 维权者必须强壮 |
| 36 | while attackers need to find only one weakness, | 攻击者只需要找到一个弱点 |
| 37 | rested on defenders lacking the time, people and money to cover everything. | 维权者没有时间、人和金钱来应付一切。 |
| 38 | AI alleviates that problem. | 大赦国际缓解了这一问题。 |
| 39 | AI is already making defenders' response to these threats stronger. | 大赦国际已经在加强维权者对这些威胁的反应。 |
| 40 | They own their code and can task AI to preemptively analyze it line by line | 他们拥有自己的代码,可以让AI先发制人地逐行分析 |
| 41 | from all possible angles of attack. | 从所有可能的角度攻击。 |
| 42 | Rather than only probing it from the outside, as attackers must do. | 而不是像攻击者必须做的那样,只从外部探测。 |
| 43 | They see their own network traffic, | 他们看到自己的网络流量, |
| 44 | so AI can spot the subtle signs of an intruder using stolen credentials | 这样AI就能发现入侵者的微妙迹象 用偷来的证件 |
| 45 | and legitimate tools, as China's hackers do, which older tools miss. | 和合法工具,像中国的黑客所做的, 旧的工具错过了。 |
| 46 | Using AI available today, utilities can turn operational data | 使用今天可用的AI,公用设施可以转换操作数据 |
| 47 | into virtual replicas of power grids and water systems. | 变成电网和水系的虚拟复制品。 |
| 48 | These digital twins let operators test defenses | 这些数码双胞胎让操作员测试防御 |
| 49 | without putting essential services at risk. | 而不危及基本服务。 |
| 50 | Attackers working from outside can build only crude approximations. | 从外部工作的攻击者只能建立粗糙的近似. |
| 51 | Crutely, AI can fix flaws after finding them. | 谨慎地说,AI在找到缺陷后可以弥补缺陷. |
| 52 | In DARPA's AI Cyber Challenge last year, | 去年在DARPA的AI网络挑战中, |
| 53 | AI systems scanning open-source software of the kind that underpins critical infrastructure | AI系统扫描支持关键基础设施的开源软件 |
| 54 | patched 43 planted vulnerabilities and uncovered 18 previously unknown flaws, | 修复了43个埋设的弱点,并发现了18个以前未知的缺陷, |
| 55 | fixing 11 of them. | 修理11个 |
| 56 | For aging equipment that cannot safely be patched, | 对于无法安全补齐的老旧设备, |
| 57 | AI can give each flaw a specific workaround, | AI可以给每个缺陷一个具体的工作, |
| 58 | such as a firewall rule or a settings change, | 例如防火墙规则或设置更改, |
| 59 | tested on a digital twin before it goes live. | 测试在数字双胞胎开始前。 |
| 60 | And AI is cheap enough that small, understaffed utilities can use it too. | 而AI也足够便宜,小的,人手不足的公用事业也可以使用. |
| 61 | None of this happens by default. | 这一切都不是默认发生的。 |
| 62 | AI favors defenders only if they move with urgency. | 只有在维权者急于行动时,大赦国际才给予他们优惠。 |
| 63 | Government should start with the infrastructure operators and shared suppliers, | 政府应从基础设施运营商和共享供应商开始, |
| 64 | whose failure would cause the greatest disruption, | 他们的失败将造成最大的破坏, |
| 65 | enlisting AI labs to find flaws in their code or firmware | 招募人工智能实验室寻找其代码或固件中的缺陷 |
| 66 | and produce fixes by set deadlines. | 并按规定的期限作出修正。 |
| 67 | Government should share the cost for smaller operators. | 政府应分担较小经营者的费用。 |
| 68 | Success should be measured by how quickly threats are contained | 成功与否应当以控制威胁的速度来衡量 |
| 69 | and whether essential services survive an attack. | 以及基本服务是否在攻击中幸存。 |
| 70 | Critical infrastructure operators could address the highest priority systems within a year, | 关键的基础设施运营商可在一年内处理最优先的系统, |
| 71 | judging by how quickly they met the cybersecurity standards issued by the Biden administration. | 从他们达到拜登政府发布的网络安全标准的速度来看 |
| 72 | Findings should be shared with trusted partners using the coordinated disclosure practices | 利用协调的披露做法,与信任的伙伴分享调查结果 |
| 73 | that keep fixes from tipping off attackers before systems are patched. | 在系统被补丁之前 避免向攻击者告密 |
| 74 | The aim should be for a discovery in one country to strengthen every country running the same technology. | 目标应该是在一个国家进行探索,以加强每一个使用相同技术的国家。 |
| 75 | Such practical partnerships are more urgent than AI treaties, and the foundations exist. | 这种实际的伙伴关系比大赦国际的条约更为紧迫,而且存在各种基础。 |
| 76 | NATO's annual Locked Shields exercise brings together thousands of participants from dozens of countries | 北约每年的锁盾演习 聚集了来自数十个国家的数千名参与者 |
| 77 | to defend power grids, water systems and other networks against live attacks. | 以防御电网、供水系统和其他网络的现场攻击。 |
| 78 | The exercises could be expanded to include using AI to generate patches and controls, | 这些练习可以扩大,包括利用AI生成补丁和控制, |
| 79 | sharing the results across the alliance. | 分享整个联盟的成果。 |
| 80 | Here, too, there is precedent. | 这里也有先例。 |
| 81 | In 2021, America formed the International Counter Ransomware Initiative, now comprising nearly 70 countries. | 2021年,美国组建了国际反兰素软件倡议,现由近70个国家组成. |
| 82 | Members share threat warnings and forensic techniques, | 成员分享威胁警告和法医技术, |
| 83 | including how to trace ransoms laundered through the crypto ecosystem. | 包括如何追踪通过密码生态系统清洗的赎金。 |
| 84 | They coordinate policy, for example by pledging not to pay ransoms. | 它们协调政策,例如保证不支付赎金。 |
| 85 | Alongside law enforcement takedowns of gangs and their money laundering networks, | 在执法部门打击帮派及其洗钱网络的同时, |
| 86 | the effort has squeezed the business. | 努力挤压了企业。 |
| 87 | After peaking at over $1 billion in 2023, ransom payments have fallen each year since. | 在2023年达到顶峰,超过10亿美元之后,赎金支付自此逐年下降. |
| 88 | The time has come to use AI as aggressively on defence as adversaries will use it in attack. | 现在是时候用AI来防御了 因为对手会用它来攻击 |
| 89 | A serious international agenda for AI safety should explain not only how it will contain the risks of a new technology, | 一项严肃的AI安全国际议程不仅应解释它将如何遏制新技术的风险, |
| 90 | but how it will keep the lights on for the world's citizens, and how soon. | 但它将如何保持 灯为世界的公民, 以及多久。 |
该视频共有字幕 90 条。解锁更多字幕为会员功能,请移动到 价格
![《经济学人》前美国国安高官:AI怀疑论者看错了威胁,它其实更能保卫脆弱的现状 [C1] p01 原声 (P1)](https://1zimu.com/subtitle/imgs/20261117/dhey6x24owaxiuuvpwqe1zet_1794886041010.jpg)