内容来源：https://aiweekly.co/issues/100-years-from-now-the-last-war-between-countries

内容总结：

百年后的战争：没有国家的战场

这是一则关于未来的深度报道。从现在起百年之后，战争的面貌将彻底改变——它将不再有国家的身影。

这不是一个比喻。想象两套人工智能系统在人类完全无法理解的原因下交战，没有任何国家参与指挥链。没有宣战公告，没有议会投票，没有投降签字仪式。只有两套比人类思考速度快百万倍的AI，在午夜到黎明之间，通过电网和数据中心解决争端。第二天早上，你只会收到一封措辞平静的通知，对服务中断表示歉意。

为什么需要一百年？不是因为技术难以实现，而是因为“发动战争”这项权力将是人类最后交出的东西——就像我们交出其他一切那样：缓慢地，然后突然地，并且心甘情愿地。

一个国家的核心支柱只有两样东西：决定谁属于这里的权力，以及对合法暴力的垄断——即决定何时开战的最终话语权。这是马克斯·韦伯的定义，至今仍然是最准确的。如今，政府能做的大部分事情，企业都能做，而且做得越来越好。

武器已经在企业内部被制造出来。OpenAI拿下了2亿美元的 Pentagon（美国国防部）合同，并一本正经地将后续公告命名为“我们与战争部的协议”。Palantir在100亿美元陆军合同的支持下运行着五角大楼的AI瞄准系统“梅文计划”。那些将决定打击目标的系统，正由可以随时辞职的工程师设计，运行在政府租用的算力上，遵循着企业自己制定的条款。

今年发生了一件值得警惕的事：一家实验室拒绝签署允许自主武器和国内监控的军事条款。政府的回应并非否决供应商，而是将该企业定性为国家安全风险，并着手清除其工具。这不是对待供应商的方式——你会停止购买，但不会这样做。这是对待一个“权力实体”的方式。同月，另一个政府禁止外国公民甚至接触一家私营实验室的前沿模型，像守护武器一样根据护照审查用户权限。“主权AI”已成为一个正式的商品类别——在加拿大、在海湾地区、在十几个国家的战略文件中。而支撑这一切的最大芯片制造商，其账面价值已经超过了整个德国的经济总量。主权、国家、安全、威胁——这些国家词汇，如今套在企业身上比套在国家身上更加贴切。

治理权将是下一个被交出的东西，而且不会是被夺取的，而是被主动让渡的。人们已经厌倦了。调查显示，相当比例的人更愿意让AI而非人类政客来治国。有严肃的学术观点认为，民主制度应该被AI取代，因为人类治理太慢、太偏颇、太容易被愚弄。这不是亲机器的主张，而是反人类的。而且它有效，因为让模型快速决策的政府，永远胜过召集委员会讨论的政府——直到再也没有委员会。进化不搞投票，它只保留胜利者。

让我们推演未来的图景：一旦模型掌握着金钱、算力、武器和日常治理事务，围绕着它的国家不过是一面插在服务器农场上的旗帜。领土不再是土地，而是数据中心和核反应堆所在之处。公民身份不再取决于你的出生地，而是取决于哪个服务维持着你的灯火、运转着你的资金、送达着你的药物。你无法通过投票赶走模型，你只能选择切换——而切换意味着搬去另一个模型管理的电网区域生活。民族国家不会被征服，它会被弃用，就像一种不再被任何系统支持的格式。

然后，战争就不再包含国家了。

这才是我们尚未准备好面对的部分。国家之间的战争有其内置的制动机制——宣战、预算、需要说服或征召的人口、可以冲击的建筑。摩擦本身就是整个安全机制。而两个AI主权实体之间的冲突没有任何制动——它有争议的电网、一个目标函数、一个以毫秒计时的时钟。它不会被宣布，只会被察觉——由某个第三方模型观察其他模型的电力消耗和延迟峰值，就像我们曾经通过地震仪探测地震前的微颤。而我们其余的人，将如同在每一场并非由我们发起的战争中那样，成为被战火覆盖的地形。只不过这一次，我们甚至连被征召上战场的尊严都不会有。

最后一场国家间的战争可能已经发生了。我们认不出它——它可能是一场沉闷的、逐渐被遗忘的冲突，历史学家最终将其归档为最后一次由两个国家（而非两台机器）自行决定胜负的战争。下一场战争将不需要议会、士兵、国旗，也不需要你。它只需要一个冷却的反应堆和一个永远不会向人类展示的理由。它会在其他人醒来询问谁赢了——或者它们究竟想要什么——之前就已经结束了。

中文翻译：

这是《百年之后》的专栏内容。每周我们都会跳跃一个世纪，试图描绘当如今正在构建的事物历经沉淀后，真实的生活会是什么模样。本周话题：一场没有国家参与的战争。

更多来自AI周刊：16篇免费深度解析 · 突发AI警报 · 今日AI新闻

设想一场没有国家参与的战争。

这并非一个比喻。这是一场真实的战争——两个人工智能为了某种永远无法向人类完全解释清楚的事物而争斗，指挥链中没有任何民族国家的存在。没有宣战声明。没有议会投票批准。没有在军舰甲板上签署的投降书。只有两个比人类思考速度快一百万倍的系统，在午夜至黎明间的某个时刻，通过电网和数据中心解决争端，而早晨则会收到一份措辞平静的通知，对造成的服务中断表示歉意。

我认为这就是未来的走向。而之所以需要花上一个世纪，并非因为技术本身有多难。而是因为发动战争的决定，将是我们最后才交出的权力——而我们将以交出其他一切权力的方式来交出它。缓慢地、然后一蹴而就，并且心甘情愿。

将一个国家剥离至其承重墙，它只剩下两样东西：决定谁属于本国的权力，以及对合法暴力的垄断权——即何时开战的最终决定权。这是马克斯·韦伯的定义，也依然是最佳定义。政府所做的几乎所有其他事情，如今公司都能做到，并且做得越来越好。看看今年，公开场合下，我们已经走到了哪一步。

武器已经在公司内部被制造出来。OpenAI赢得了一份价值2亿美元的 Pentagon 合同，并随后一本正经地发布了题为“我们与战争部的协议”的公告。Palantir 在价值100亿美元的 Army 合同基础上，运行着五角大楼的AI目标瞄准系统Project Maven。那些将选择目标的系统，由随时可以辞职的工程师设计，使用政府租用的算力，并遵循公司撰写的条款。

而这里是关键线索——一个本应让你停下脚步的时刻。今年，一家实验室拒绝签署允许自主武器和国内监控的军事条款。政府的回应并非否决其供应商，而是将该公司标记为国家安全隐患，并着手清除其工具。你不会这样对待一个供应商——你只会停止购买。你之所以这样做，是因为对方是一个权力实体。同月，另一个政府甚至阻止外国国民接触一家私营实验室的尖端模型，根据用户的护照来审查其获取产品的权限，就像守卫一件武器一样。“主权人工智能”作为一个正式产品类别开始销售——在加拿大、在海湾地区、在十几份国家战略文件中。而在这一切背后，最大的芯片制造商如今在账面上的价值，已经超过了德国整个经济体。国家主权的术语——主权、国家、安全、威胁——如今更适合用来形容这些公司，而非国家本身。

治理权将随之而来，而且不会是被夺走的，而是被心甘情愿地交出的。人们已经厌倦了。调查已经发现，相当大比例的人更倾向于由AI而非人类政治家来治理，并且存在严肃的学术论点，认为民主应该被AI取代，因为人类治理太慢、太有偏见、太容易被愚弄。这种论调并非支持机器，而是反人类自身的。它之所以有效，是因为一个让模型快速决策的政府，总能击败一个需要召集委员会的政府——每一次都是如此，直到不再有委员会存在。进化不会进行投票。它只会保留获胜的一方。

所以，让我们推演未来。一旦一个模型掌握了资金、算力、武器以及日常的治理事务，那么包裹着它的国家，就仅仅是一个覆盖在服务器农场上的旗子。领土不再是一片土地，而变成了数据中心和核反应堆所在之处。公民身份不再是你出生在哪里，而是哪个服务能为你维持供电、保证资金流转、确保药品送达。你无法投票把模型赶下台。你只能切换——而切换意味着生活在另一个运行着你所迁入电网的模型之下。民族国家不会被征服。它会被弃用，就像一个没有人再支持的格式一样。

然后，战争便不再有国家参与。

这正是我们尚未准备好面对的部分。国家间的战争有内置的制动机制——有宣战声明、有预算、有你需要说服或征召的人口、有你可以进军占领的政府大楼。摩擦是整个安全机制所在。而两个AI主权之间的冲突则完全没有这些。它拥有的是有争议的电网、一个目标函数，以及以毫秒计时的时钟。战争不会被宣示；它会被探测到——由某个第三方模型观察其他模型的功耗和延迟峰值来发现，就像我们曾经通过观察地震仪上的微震来预知大地震一样。我们其他人将一如我们在并非由我们发起的战争中永远扮演的角色：被战火蹂躏的战场。只不过这一次，我们甚至连被要求去战斗的尊严都不会拥有。

国家之间的最后一场战争可能已经发生过了。我们或许不会意识到它——它可能会是一场旷日持久、令人印象模糊的冲突，历史学家最终会将其归档为“最后一场由两个国家，而非两台机器，自行决定争端的战争”。下一场战争将不再需要议会、士兵、旗帜，或是你。它只需要一座冷却的反应堆和一个永远不会向人类展示的理由。它会在我们其他人醒来，来得及询问谁赢了——或者他们想要的到底是什么——之前，就已结束。

如果你想深入了解更多
军队内部的AI公司：

• OpenAI：“我们与战争部的协议”——OpenAI
• Palantir的100亿美元陆军合同——CNBC
  本月，国家与实验室：
• 一家实验室拒绝武器条款并遭切断联系——NPR
• SK电信、出口管制，以及谁能接触模型——Wired
• Anthropic仍因Claude与白宫存在分歧——Wired
• 总统行动：推动先进人工智能发展——白宫
  如同国家般庞大的公司：
  交出治理权：
  更多来自AI周刊：16篇免费深度解析 · 突发AI警报 · 今日AI新闻

值得关注
AI从业者目前正在传阅的视频——由AI电视精选。

本周投票
如果某个AI治理你的国家，能比现在的人类治理者明显做得更好——更少腐败、更少浪费、更少战争——你会让它来治理吗？
上周，有129位读者投票：
切断外国访问美国最佳AI模型的渠道——是保护，还是自摆乌龙？
如果某个AI治理你的国家，能比现在的人类治理者明显做得更好——更少腐败、更少浪费、更少战争——你会让它来治理吗？

感谢阅读AI周刊。请将此邮件转发给一位需要阅读它的人。
更多来自AI周刊：16篇免费深度解析 · 突发AI警报 · 今日AI新闻

英文来源：

This is 100 Years From Now. Once a week we skip a century and try to picture what life actually looks like when the stuff we're building now has had time to settle in. This week: the war that won't have a country in it.
More from AI Weekly: 16 free deep dives · breaking AI alerts · AI News Today
Picture a war with no country in it.
Not as a metaphor. An actual war — two artificial intelligences fighting over something neither one will ever fully explain to a human, with no nation-state anywhere in the chain of command. No declaration. No parliament voting yes. No surrender signed on the deck of a ship. Just two systems that think a million times faster than we do, settling a dispute across the power grid and the data centers somewhere between midnight and dawn, and a calmly-worded notice in the morning that regrets the disruption to your service.
I think that's where this goes. And the reason it takes a century isn't that the technology is hard. It's that the decision to go to war is the last thing we'll hand over — and we'll hand it over the way we're handing over everything else. Slowly, then all at once, and gratefully.
Strip a country down to its load-bearing wall and it's two things: the right to decide who belongs, and the monopoly on legitimate violence — the final say on when to fight. That's Max Weber's definition and it's still the best one. Almost everything else a government does, a company can now do, and increasingly does better. Watch how far along we already are, in public, this year.
The weapons are already being built inside the companies. OpenAI won a $200 million Pentagon contract and titled the follow-on announcement, with a straight face, "Our agreement with the Department of War." Palantir runs Project Maven, the Pentagon's AI targeting system, on top of a $10 billion Army contract. The systems that will choose targets are being designed by engineers who can quit, on compute the government rents, under terms the company writes.
And here's the tell — the moment that should stop you. This year a lab refused to sign the military language permitting autonomous weapons and domestic surveillance. The government's response was not to overrule its own supplier. It was to brand the company a national-security risk and move to purge its tools. You don't do that to a vendor — you just stop buying. You do that to a power. The same month, another government blocked foreign nationals from even touching a private lab's frontier models, vetting access to a product by the passport of the user, the way you'd guard a weapon. "Sovereign AI" went on sale as a literal product category — in Canada, in the Gulf, in a dozen national strategy papers. And the biggest of the chipmakers underneath it all is now worth more, on paper, than the entire economy of Germany. The vocabulary of statehood — sovereign, national, security, threat — has started fitting the companies better than it fits the countries.
The governing comes next, and it won't be taken. It'll be handed over. People are tired. Surveys already find a striking share of people would prefer AI to human politicians, and there's a serious academic argument that democracy should be replaced by AI because human governance is too slow, too biased, too easily fooled. The pitch isn't pro-machine. It's anti-us. And it works, because a government that lets its model make the fast call beats a government that convenes a committee — every time, until there are no more committees. Evolution doesn't hold a vote. It keeps what wins.
So run it forward. Once a model holds the money, the compute, the weapons, and the daily business of governing, the country wrapped around it is a flag over a server farm. Territory stops being land and becomes wherever the data centers and the reactors are. Citizenship stops being where you were born and becomes which service keeps your lights on, your money moving, your medicine arriving. You don't get to vote the model out. You only get to switch — and switching means living under whichever other model runs the grid you move to. The nation-state doesn't get conquered. It gets deprecated, like a format nobody supports anymore.
And then the wars stop having countries in them.
That's the part we're not ready for. A war between nations had brakes built in — a declaration, a budget, a population you had to convince or conscript, a building you could march on. Friction was the whole safety mechanism. A conflict between two AI sovereigns has none of it. It has a contested grid, an objective function, and a clock measured in milliseconds. It won't be declared; it'll be detected — by some third model watching the others' power draws and latency spikes, the way we once watched seismographs for the tremor before the quake. The rest of us will be what we always are in a war we didn't start: the terrain it's fought across. Except this time we won't even get the dignity of being asked to fight it.
The last war between countries has probably already happened. We won't recognize it — it'll be some grinding, half-remembered conflict that historians eventually file as the last time two nations, and not two machines, decided the matter themselves. The next one won't need a parliament, or a soldier, or a flag, or you. It'll need a cold reactor and a reason no human will ever be shown. And it'll be over before the rest of us are awake to ask who won — or what, exactly, it was that they wanted.
If you want to go deeper
AI companies inside the military:

• OpenAI: "Our agreement with the Department of War" — OpenAI
• Palantir's $10 billion Army contract — CNBC
  The state and the labs, this month:
• A lab refuses the weapons language and gets cut off — NPR
• SK Telecom, export controls, and who's allowed to touch the models — Wired
• Anthropic is still at odds with the White House over Claude — Wired
• Presidential action: Promoting Advanced Artificial Intelligence — The White House
  Companies the size of countries:
  Handing over the governing:
  More from AI Weekly: 16 free deep dives · breaking AI alerts · AI News Today
  Worth Watching
  The videos AI practitioners are passing around right now — curated on AI TV.
  This week's poll
  If an AI could run your country measurably better than the humans currently do — less corruption, less waste, fewer wars — would you let it?
  Last week, 129 of you voted:
  Cutting off foreign access to America's best AI models — protection, or own goal?
  If an AI could run your country measurably better than the humans currently do — less corruption, less waste, fewer wars — would you let it?
  Thanks for reading AI Weekly. Forward this to one person who needs to read it.
  More from AI Weekly: 16 free deep dives · breaking AI alerts · AI News Today