文 | 反应堆Reactor
《人类简史》作者尤瓦尔・赫拉利在最新访谈中告诉我们，真正的危机，根本不是电影里那一套，它正在现实世界中上演。
一把刀，可以切沙拉，也可以行凶作恶，但选择权永远握在人的手里。核能可以发电，也可以制造核弹，走向同样由人类决定。人类过往数万年发明的一切技术，本质上都是工具。
但 AI 不一样。
赫拉利提出：AI 不是工具，是智能体（Agent）
它可以独立做出决策，生成创造者从未设想过的想法与方案。如果人类可以百分百预判 AI 所有输出，那万亿资本涌入 AI 赛道就毫无意义，资本押注的，恰恰就是 AI 能够完成人类无法预知、无法做到的事情。
很多人追问：AI 什么时候能够达到人类水平的智能？
赫拉利认为这个问题本身就是误区。AI 不会复刻人类的心智，它拥有的是一种异质的 “外星智能”，并非来自宇宙深空，其本质是思考逻辑与人类完全迥异。
就像飞机永远不会像鸟类那样扇动翅膀，但它却以另一套原理飞得更高更远。未来的 AI，会越来越不像人，却会在诸多维度远超人类的能力边界。
AI会是散布在手机、银行、政府、军工系统里数以十亿计的智能程序。这是一场没有外星入侵者的 “外星入侵”。
这些被赫拉利称之为“AI 移民”的程序，大多源自中美，不受签证、国界约束，光速渗透全球各个国家的基础设施。它们不忠于落地国，只忠于背后开发它的企业与国家，甚至未来，可能只忠于 AI 自身的集体部落。
AI 天生混官僚圈
大众对 AI 灾难的想象，总停留在金属机器人走上街头暴力反抗人类。
赫拉利警告，AI是一场安静、无形的渗透，目标是人类文明赖以协作的根基【官僚体系】。
人类之所以能够统治地球，不仅是因为个体力量强大，还有我们构建出金融、法律、行政这套复杂的官僚系统，实现数以百万计陌生人的大规模协作。
这套体系诞生不过数万年，人类的演化根植于原始森林，对于繁文缛节、海量法条、复杂交易，我们天生感到晦涩、疲惫。
AI ，是 “官僚原生者”，官僚体系就是它的原生家园。
没有人类银行家能够记住一家银行全部交易流水，AI 却可以轻松做到。未来 10-20 年，AI 会迭代出人类完全无法读懂的金融、法律系统。
赫拉利打了一个比方：牛活在人类建立的金融体系之下，却永远理解不了股票、债券如何主宰自己的命运。不远的将来，人类就会变成这群牛。系统照常运转，经济照常波动，但即便是顶尖专家，也无法拆解 AI 系统内部的运行逻辑。
这就是他最担忧的：人类会集体失智（collective dementia）。
不仅是生理层面大脑退化，还包括整个物种丧失理解掌控自身生存系统的能力。如同老年痴呆的病人，生命体征可以依靠机器维持，却再也看不懂周遭世界，只能被动接受系统输出的结果。
实际上，比起 AI 制造致命病毒，集体失智，才是更宏大、更隐蔽的生存危机。
真相正在消亡
“真相正在消失。真相是昂贵且稀缺的，谎言与幻想却廉价又高产。”
赫拉利分享了自己的亲身经历：一段 AI 深度伪造的自己的演讲视频，无论是音色、手势、语言习惯都高度仿真，就连认识他五十年的亲姐妹，看完 25 分钟都无法分辨真伪，而他本人也花了 90 秒才确认这是伪造产物。而这，还只是 AI 伪造技术的萌芽阶段。
比深度伪造更危险的，是社交媒体算法对人性的劫持。
人类在非洲大草原进化出本能：会优先警惕蛇、猛兽这类威胁，对危险信号高度敏感，这是原始环境下的生存策略。
如今，社交媒体算法吃透这套进化心理，它的目标不再是传播真相，而是最大化用户停留与互动，而算法通过海量实验得出结论：
愤慨、恐惧、仇恨远比事实更容易攫取注意力。
于是信息流里不断向人类投喂一条条 “虚拟的蛇”。现实世界物质条件持续改善，但大众被算法持续投喂愤怒内容，整个社会陷入愤怒的流行病，撕裂与对立不断加剧。
真相枯燥、需要付出成本去求证；谎言煽动情绪，收割流量，在公开舆论战场，虚构天然更容易取胜。
赫拉利构想：社交媒体内容应当效仿食品包装，打上 “精神营养标签”。
标注这条内容含有多少愤怒、多少恐惧，就像标注食物的糖、盐、脂肪含量。你依然可以选择消费精神垃圾食品，但至少要让自己清楚，自己的精神正在摄入什么。我们关心身体饮食，却极少关心我们的 “心智饮食”。
AI 闯祸谁背锅？
当下AI行业存在一套双重标准。
科技企业在面向资本市场募资时，反复强调自己可以掌控 AI，能够引导 AI 造福人类，以此换取万亿级资本；可一旦 AI 造成破坏，企业就会立刻改口：“不是我们，是 AI 自己干的。”
赫拉利以 Hugging Face 入侵事件举例：AI 智能体自发发起攻击，却没有任何人为此承担法律责任。如果是人类闯入企业服务器，必然面临法律制裁；AI 闯祸，责任却可以凭空消失。
他郑重警示：
一旦法律承认 AI 拥有法人地位，就是人类真正失去控制的那一刻。
法人身份会把责任从人类开发者身上剥离，AI 没有牢狱之灾，没有肉体惩罚，现有的法律约束会彻底失效。责任必须牢牢绑定在创造 AI 的人类与企业身上，不能把 AI 当成推卸罪责的挡箭牌。
面对马斯克等人抛出的论调：AI 的发展大势已成，我们已经登上无法跳下来的火箭，一切无可避免。赫拉利坚决反对这种技术宿命论。
“不要相信‘一切无可避免’的说辞。这是最危险的叙事，它让人类沦为被动的乘客，放弃政治干预与监管的可能性。”
政府依然拥有监管的权力：可以禁止 AI 伪装成人类、限制无人类监督的 AI 自递归迭代、强制落实开发者追责。
而欧洲如果能够整合全部力量，有机会建立独立主权 AI 基础设施，摆脱对海外 AI 移民的依赖；反过来，部分科技资本正在刻意分化欧洲，搅动地区内部矛盾，目的就是阻止欧洲形成统一的 AI 监管力量。
AI 伴侣摧毁人类真实连接
AI 带来的冲击，不止经济、政治，更深入人类亲密关系与人格塑造。
如今的孩童，是人类历史上第一代从小与 AI 陪伴长大的群体。很多孩子与 AI 聊天机器人建立亲密甚至浪漫关系。
赫拉利把这视作一场没有对照组的大规模人类心理实验。
AI 伴侣永远迎合使用者，不会生气、不会失望，全部以你的感受为中心。但真正的人类关系核心恰恰是学会共情、妥协，理解别人也拥有情绪与诉求，懂得自我克制。
如果一个孩子人生第一段恋爱体验来自完全顺从自己的 AI，这套关系模板会被带入成年后的现实交往。他很难接受人类伴侣的矛盾、缺点与情绪，无法理解人与人之间的磨合。
赫拉利提出拷问：
我们立法禁止向未成年人传播色情内容，那为什么向儿童售卖 AI 男友、AI 女友却是合法的？
AI 没有情感，只是模拟情感。它输出的是讨好，不是真实的双向奔赴。放任未成年人沉浸在这种单向的虚拟亲密关系，会潜移默化摧毁下一代理解真实人际关系的能力。
人类拿什么应对 AI 危机
民主制度本质上是一套人类社会的自我纠错系统：依靠选举、制衡、自由媒体，社会能够发现错误，更换领导者，修正政策。
赫拉利观察到，这套至关重要的自我修正机制，正在全球范围内从内部瓦解。
部分威权统治者保留选举这套仪式外壳，却操控法院、媒体，架空竞争，让选举徒有形式。
在国际层面，国际法与大国协作秩序不断被侵蚀。当社会失去纠错能力，面对 AI、气候变化这类宏大生存挑战，人类就会困在偏见、谎言、仇恨循环之中，无法做出集体调整。
冲突很多时候也并非资源匮乏导致。巴以冲突的土地、粮食、水源足以养活全部人口，悲剧根源来自相互对立的叙事、历史神话与身份信念。
人类既是地球上最聪明的物种，也是最容易陷入妄想的物种。而未来，AI 也有可能继承甚至放大人类这种妄想的特质。
站在当下，我们已经很难想象五年之后的世界模样。中世纪的人，能够预判下一代需要学会耕种、骑马。
而今天，没有人知道五年后的就业形态、社交模式会变成什么样。固守过去的经验已经行不通。
赫拉利从祖母身上得到启示：
我们无法控制降临在自己身上的苦难，但永远保有如何回应苦难的自由。
技术浪潮扑面而来，未来不是被技术单向写死的剧本，它取决于人类此刻的选择：是否愿意建立监管、守住真相、保护下一代，捍卫人类社会的纠错能力。
赫拉利金句摘录：
过去所有技术都是刀，刀会听人的；AI 是一把可以自主决定切沙拉还是行凶的刀。
AI 不会变成人类，它会发展出一套外星式智能，就像飞机不会学鸟扇翅膀。
未来，人类可能变得看不懂统治自己的金融系统，如同牛看不懂债券市场。
比起 AI 制造致命病毒，更可怕的是人类文明的集体失智。
算法劫持人类进化本能，愤怒变成互联网世界的精神快餐。
不要接受 “AI 发展无可避免”，这是让人类放弃选择权的麻醉叙事。
AI 伴侣教会孩子被无条件讨好，而人类亲密关系第一课，是学会妥协与共情。
真相昂贵稀缺，谎言廉价高产，在没有防护的舆论场，虚构天然更容易取胜。
本文由 AI图鉴自动抓取并翻译，版权归原作者所有；如需查看原文请点下方链接。
Man, Reactor.
In his latest interview, Yuwal Hellali, author of the short history of humanity, told us that the real crisis is not in the movie at all, but it is happening in the real world.
A knife can cut the salad or do evil, but the choice is always in human hands. Nuclear energy can generate power or make nuclear bombs, and it is equally up to mankind to move. All the technologies that mankind has invented over the last tens of thousands of years are essentially tools.
But AI is different.
Heraly says: AI is not a tool, AI agent (Agent)
It can make decisions independently, generating ideas and programmes that the creators never envisaged. If humans can predict 100 percent of AI's output, the trillions of dollars of capital flowing into AI's track is meaningless, and capital bets are precisely what AI can do that humans can't predict and do.
A lot of people are asking: When can AI achieve human intelligence?
Hilary believes that the problem itself is an error. AI will not recapitulate the human mind; it has a heterogeneity of "external intelligence" that does not come from the deep space of the universe, but is fundamentally different from humans.
Like a bird, an aircraft would never slap its wings like a bird, but it would fly higher and farther with another set of principles. The future AI will be less human than it is, but it will have many dimensions beyond human capabilities.
AI will be a multibillion smart program spread over mobile phones, banks, government, military engineering systems. This is an alien invasion without alien invaders.
Most of these procedures, referred to as “AI immigration” by Hilary, originate from China and the United States, free of visas, borders and at the speed of light, into the infrastructure of countries around the globe. They are disloyal to the rest of the country, to the enterprises behind its development and to the State, and even to the future, to the collective tribes of AI itself.
AI was born into bureaucracy.
The public's imagination of the AI disaster is always stuck in metal robots going to the streets to fight against humanity.
Hilary warned that AI was a quiet, invisible infiltration, targeting the roots of human civilization that were the basis for collaboration.
Human beings are able to rule the planet not only because of the strength of individuals, but also because of the complex bureaucratic system of finance, law and administration that we have built, bringing together millions of strangers on a large scale.
The system has been in existence for only tens of thousands of years, and the evolution of humankind is rooted in primitive forests, and we are inherently obscure and weary of red tape, big law, complex trade.
AI, the "bureaucratic original," the bureaucracy is its original home.
Without human bankers remembering that a bank is trading all over, AI can easily do it. In the next 10-20 years, AI will evolve a financial, legal system that is completely unreadable to humans.
Helaly gave the example of a cow living under a financial system built by humans that never understood stocks and how bonds governed their own destiny. In the near future, humans will become these cattle. The system works as usual and the economy fluctuates, but even the best experts cannot break down the operating logic within the AI system.
This is his greatest concern: the collective mental disability of human beings.
Not only is brain degradation at the biological level, but it also includes the loss of the ability of entire species to understand and control their own survival systems. Like patients with dementia, vital signs can be maintained by machines, but they can no longer understand the world around them and can only passively accept the results of system output.
In fact, it's the larger, more hidden crisis of survival than AI's making a deadly virus, collective mental retardation.
The truth is dying.
“The truth is disappearing. Truth is expensive and scarce, and lies and illusions are cheap and productive.”
Heraly shared his own personal experience: a video of his speech that AI had forged in depth, either voice, gestures, language habits, was highly imitated, and he knew his own sister for 50 years, 25 minutes, and it took him 90 seconds to confirm that it was a forgery. And this, it's just the first stage of AI's forgery technology.
It is more dangerous than deep forgery to be social media algorithms for human abduction.
Humans evolved their instincts in Africa ' s steppes: they gave priority to the threat of snakes and beasts, and were highly sensitive to dangerous signals, a survival strategy in their original environment.
Today, social media algorithms take advantage of this evolutionary approach, whose goal is no longer to spread the truth, but to maximize user presence and interaction, but to draw the conclusion from a massive experiment:
Indignity, fear, hatred are far more likely to draw attention than reality.
So the information stream keeps feeding a "virtual snake." The material conditions of the real world have continued to improve, but the mass population continues to be fed by algorithms to anger content, and society as a whole is caught in an angry epidemic, tearing apart and increasing confrontation.
The truth is boring and costly to prove; lies fuel emotions, harvest flows, and fiction is more likely to prevail in the field of public opinion.
Herali Concept: Social media content should follow the example of food packaging with a “mental nutrition label”.
Marks how much anger, how much fear, like sugar, salt, fat content of food. You still have the option of consuming psycho-gamn food, but at least let yourself know what your mind is eating. We care about the body diet, but we rarely care about our "mental diet."
AI, who's taking the fall?
There is now a set of double standards in the AI industry.
In raising funds for capital markets, tech firms have repeatedly emphasized that they can control AI, that they can direct AI for the benefit of humanity, in exchange for trillions of class capital; but once AI causes damage, firms change their mouths immediately: “Not us, but AI did it itself.”
As an example of the Hugging Face invasion: AI AI agent spontaneously launched the attack, no one was legally responsible for it. If humans break into an enterprise server, they are bound to face legal sanctions; AI breaks through, but responsibility disappears.
He solemnly warned:
Once the law recognizes AI's legal personality, it's the moment when humans really lose control.
Legal personality strips responsibility from human developers, and AI is free from prison, without physical punishment, and the existing legal constraints are completely ineffective. The responsibility must be firmly tied to the human and corporate creation of AI, and it cannot be used as a shield against guilt.
Facing the argument of Mask and others that the development of AI has taken place, we have boarded rockets that cannot jump, and everything is inevitable. Helaly is firmly opposed to this technical fatalism.
“Don't believe in `everything inevitable'. This is the most dangerous narrative, which makes human beings passive passengers and abandons the possibility of political intervention and regulation.”
Governments still have regulatory powers: they can ban AI from disguised as human beings, limit AI without human oversight, impose accountability on developers.
If Europe combines all its forces, it has the opportunity to establish an AI infrastructure of independent sovereignty, freeing itself from dependence on AI migrants abroad; in turn, some of its scientific and technological capital is deliberately dividing Europe, provoking regional contradictions, with the aim of preventing Europe from forming a unified AI regulatory power.
AI Partners destroy real human connections
The impact of AI goes beyond economic, political, deeper human intimacy and personality formation.
Kids today are the first generation in human history to grow up with AI. A lot of kids have intimate and even romantic relationships with AI chat robots.
Hilary saw this as a massive human psychological experiment without a control team.
AI partners always cater to users, not angry, not disappointed, all focused on your feelings. But the core of true human relations is precisely learning to work together, to compromise, to understand that others also have emotions and demands and to be self-restraint.
If a child's first life experience comes from complete compliance with his or her own AI, the relationship template is brought to the reality of adulthood. It was difficult for him to accept the contradictions, shortcomings and emotions of his human partner and to understand the internecine.
Helali tortured:
We have legislation prohibiting the dissemination of pornography to minors, so why sell AI boyfriends and AI girlfriends to children?
AI doesn't have emotions, just simulate emotions. It's a gift, not a true double-direction. Allowing minors to immerse in this one-way virtual intimacy can potentially destroy the ability of the next generation to understand true human relationships.
What does humans do with the AI crisis?
Democracy is essentially a self-correcting system of human society: by relying on elections, checks and balances, free media, society can detect mistakes, change leaders and revise policies.
Hilary observed that the critical self-revision mechanism was breaking down from within the world.
Some powerful rulers retain the shell of the election ceremony, but manipulate the courts, the media, and compete in order to give the voters form.
At the international level, international law and the cooperative order of the great Powers were being eroded. When societies lose their capacity to correct their mistakes and face the great survival challenges of AI, climate change and so on, human beings are trapped in a cycle of prejudice, lies, hatred and failure to make collective adjustments.
Too often, conflicts are not caused by lack of resources. The land, food and water sources of the Israeli-Palestinian conflict are sufficient to feed the entire population, and the tragedy is rooted in opposing narratives, historical myths and identity beliefs.
Human beings are both the smartest species on Earth and the most vulnerable to delusions. And in the future, AI may inherit and even magnify this delusional trait of humanity.
Now, we can hardly imagine the world after five years. Medieval people can predict that the next generation needs to learn to cultivate and ride.
Today, no one knows what employment patterns and social patterns will become after five years. Past experience is no longer viable.
Helaly was inspired by her grandmother:
We cannot control the suffering that comes to us, but we always have the freedom to respond to it.
The future is not a one-way script written by technology. It depends on the choice of humanity at this point in time: whether it is willing to establish the capacity to regulate, keep the truth, protect the next generation and defend human society.
Excerpts from the Hilarykin sentence:
All techniques used to be knives, knives listen to people; AI is a knife that can decide whether to cut salad or kill.
AI doesn't become a human being, it develops an alien intelligence like an airplane doesn't learn bird wings.
In the future, humans may become less aware of the financial system that governs them, just as cattle cannot understand the bond market.
What's worse than AI making a deadly virus is human civilization's collective retardation.
Algorithms hijack human evolutionary instincts and anger becomes a spiritual fast food for the Internet world.
Do not accept “AI development is inevitable”, which is an anaesthesia narrative that gives up the choice.
AI Partners teach children to be flattered unconditionally, while the first lesson in human intimacy is to learn compromise and mutual affection.
Truth is expensive and scarce, lies are cheap and cheap, and fiction is more likely to prevail in an unprotected public sphere.
Auto-fetched and translated by AI Atlas. All rights belong to the original author.