一名中国前高级法官公开致世界AI领军人物信之三：Ray Ku

一名中国前高级法官公开致世界AI领军人物信之三：Ray Kurzweil——人类是否忽视而遗漏了支撑文明的底层规律？

One of a Series of Open Letters from a Former Senior Chinese Judge to the World’s AI Leaders, No. 3: Ray Kurzweil — Has Humanity Overlooked a Foundational Principle That Sustains Civilization?

（Special Note:

The phrase “former Senior Judge” appears in the title only to give this letter a slightly better chance of being noticed. It is not intended to make readers focus on who I am, nor on the particular way in which I express what I wish to say.

I am simply a very ordinary old man. What I truly wish to discuss is also a principle that appears extraordinarily ordinary — so ordinary that no one can do anything without relying on it. Precisely for that reason, it is like air: because human beings are so familiar with it, it has long been overlooked.

What I truly hope all humanity will examine together is this:

Does such an “ordinary” foundational principle in fact determine the success or failure of human behavior, social cooperation, and even the rise and decline of civilizations? Could it be a basic sustaining principle of civilization that humanity has never fully recognized?

Therefore, who is saying it is not important. Whether what is being said is true is what deserves everyone’s attention.

Since 2014, I have continuously written to governments, contacted the media, social organizations, scholars, and people from many different fields, and gradually published the series Letters to Everyone in the World together with related articles. Today, in deciding to make public some of my letters to leading figures in global AI, I still have only one purpose:

to place this question before all humanity for collective examination.

I believe that when a question may concern the direction of human civilization as a whole, governments and international organizations have a responsibility to study it, the media have an obligation to help the public understand it, and every person has the right to participate in judging it.

What humanity faces today is no longer merely a “crossroads” in the conventional historical sense. If the underlying way in which civilization itself operates contains a fundamental problem, then moving forward, backward, left, or right may all amount only to repeating the old pattern — and this time, the repetition may bring humanity closer than ever before to an accelerating path toward its own destruction.

What truly needs to change is an upward transformation. Perhaps the first thing that must be elevated is the human mode of thinking through which we observe the world and judge value.

The core proposition I put forward is:

Procedures form relationships; relationships form structures; structures generate dynamic balance through feedback; and this ultimately appears as the properties and outcomes of a system.

From this, I summarize the most basic practical standard for civilization as:

grasp the procedure, build the pattern, seek the balance.

I believe the greatest difference between this and many grand theories of civilization in the past is that it does not first require people to accept a particular religion, ideology, or concrete value objective. Instead, it attempts to provide a common practical principle that everyone can understand, that can be applied to real action, and that can be tested against reality.

If this direction is wrong, I hope humanity will help prove it wrong.

If it truly touches upon some foundational principle, then humanity may be able to use it to rebuild and strengthen the underlying foundations of social cooperation, allowing civilization gradually to move away from its long-repeated cycles of mutual struggle, mutual harm, and war, and toward stable, harmonious, and sustainable cooperation.

That is the only reason I am making these letters public.

I do not hope the world will focus on the person who wrote these letters. I hope the world will seriously examine the questions these letters are actually asking.

Finally, a note: some websites impose word-count or character limits, so certain versions of these letters found online may be incomplete. If a text appears incomplete, you may search for the full version using the Chinese or English title of the letter, or contact the publicly listed email addresses of AI companies associated with the person to whom the letter is addressed and ask for assistance in locating or forwarding the complete version.

——特别题注：

标题中写入“前高级法官”，只是为了让这封信多获得一点被看见的机会，并不是希望读者关注我是谁，以及在用什么方式表述所说之事。

我只是一个极其普通的老人。我真正想讲的，也是一条看起来极其普通的规律——普通到任何人做任何事都离不开它，也因此这规律就像空气一样，因为人过于熟悉而长期被忽略。

我真正希望全人类共同鉴别的是：

这样一个“普通”的底层规律，是否事实上决定着人的行为、社会合作乃至文明兴衰的成败？它是否正是人类长期没有真正认清的文明底层支撑规律？

所以，谁在说并不重要，说的事情是否真实，才值得所有人关注。

自2014年以来，我持续向政府写信，也不断联系媒体、社会组织、学者和各领域人士，并陆续发表《致全球每一个人的信》及相关文章。今天决定公开部分致世界AI领军人物的信，仍然只有一个目的：

把这个问题交给全人类共同甄别。

因为我认为，对这样一个可能关系整个人类文明方向的问题，政府和国际组织有责任研究，媒体有义务帮助公众了解，每一个人也都有参与判断的权利。

今天人类面对的，已经不仅是通常历史意义上的“十字路口选择”。如果原有文明运行方式本身存在底层问题，那么前后左右可能都只是旧模式的重复，且这次可能是最接近不断加速毁灭全人类的一次。

真正需要改变的是向上抬升，也许首先是人类观察世界和判断价值的思维模式。

我提出的核心判断是：

程序形成关系，关系形成结构，结构通过反馈形成动态平衡，并最终表现为系统的性质与结果。

由此，我把文明最基本的做事标准概括为：

抓程序、做模式、求平衡。

我认为，这与过去许多宏大文明理论最大的区别在于：它并不首先要求人们接受某一种宗教、主义或具体价值目标，而试图提供一个人人都能够理解、能够用于具体做事、也能够接受事实检验的共同抓手。

如果这个方向是错的，希望全人类帮助证明它错。

如果它确实触及了某种底层规律，那么人类也许可以由此重新夯实社会合作结构的基础，使文明从长期反复的互斗、互害和战争循环中，逐步走向稳定、和谐、可持续的合作。

这也是我公开这些信件的唯一原因。

不是希望全世界关注写信的人，而是希望全世界认真看看这些信究竟提出了什么问题。

最后说明：部分网站存在字数或字符限制，因此网上看到的某些信件可能不是全文。如内容不完整，可以按照信件的中文或英文标题搜索全文，也可以通过与致信对象相关AI企业公开邮箱的联系方式，请其协助查找或转交完整版本。）

To Ray Kurzweil: If Technology Follows Accelerating Returns, Can Human Social Cooperation Keep Pace with the Singularity?

To Mr. Ray Kurzweil and the Kurzweil Technologies / Kurzweil Library Team,

I found these contact addresses through the official Kurzweil Library website.

I would be deeply grateful if the person receiving this message could kindly forward this email and its attachments to Mr. Ray Kurzweil.

The central question raised in this letter grows directly out of Mr. Kurzweil’s Law of Accelerating Returns:

If technological capability is accelerating exponentially, while humanity’s institutions and mechanisms of social cooperation continue to evolve much more slowly, could the widening gap between the two become one of the fundamental risks of the AGI era?

If members of the Kurzweil team also find this question worthy of consideration, I would warmly welcome any criticism, comments, suggestions, or assistance in bringing it to researchers who may be interested.

Thank you very much for your time and help.

Dear Mr. Ray Kurzweil,

I hope this message finds you well.

My pen name is Jin Guyu. I am an elderly Chinese citizen and formerly served as a Senior Judge in a local court in China.

Recently, I have been studying the ideas you further developed in The Singularity Is Nearer: When We Merge with AI.

Your discussion of the Law of Accelerating Returns particularly caught my attention.

As you have argued over many years, information technology does not advance at an ordinary linear pace. Each generation of technology helps produce the next breakthrough, creating an accelerating trajectory of progress.

You continue to believe that human-level artificial intelligence, and perhaps AGI, may arrive in the relatively near future, and that as artificial intelligence, brain-computer interfaces, nanotechnology, and related technologies continue to develop, human intelligence may become increasingly integrated with machine intelligence.

If your basic judgment is correct, then what we are facing is not merely:

a more advanced ordinary tool.

We may be facing:

for the first time in human history, a technological force capable of accelerating its own further development while continuously increasing the effective capabilities of individuals and organizations.

That is precisely why I especially wanted to write to you.

After studying your ideas, one question came strongly to mind:

If technology follows a “Law of Accelerating Returns,” does the evolution of human institutions of social cooperation possess any corresponding mechanism of acceleration?

And if it does not, what happens?

For many years, drawing on my experience in law, the judiciary, Chinese social reality, and the evolution of human civilization, I have been thinking about one question:

Why are the structures of social cooperation created by humanity so persistently difficult to make genuinely stable and sustainable?

In my view, human history contains an extremely important but easily overlooked pattern:

scientific and technological change usually advances faster than the institutions and theories governing human social cooperation.

New technologies continuously create new capabilities.

Those new capabilities disrupt existing modes of production, relationships of interest, power relationships, and forms of cooperation among people.

Yet humanity has rarely been able to construct, in advance, social institutions capable of matching the new powers created by technological revolution.

More often, the pattern is:

technology breaks through first;

existing structures of social cooperation then become mismatched;

society experiences enormous disruption, conflict, and sometimes war;

old institutions are forced to change under crisis;

new institutions and rules of cooperation gradually emerge;

and humanity then reaches another temporary equilibrium.

In other words, across long periods of human history:

institutions have not guided technology safely into society; rather, technology has repeatedly collided with old institutions and forced humanity to relearn how to cooperate.

In the past, this institutional lag has imposed enormous costs, but humanity has usually still had the opportunity to:

experience disaster → learn → modify institutions → rebuild order.

But when I place this historical pattern alongside your Law of Accelerating Returns, I become deeply concerned:

If technological progress accelerates more and more rapidly, while social institutions continue to adapt mainly through correction after crises, will the speed gap between the two continue to widen?

In other words:

technology may advance exponentially, while human institutions still evolve largely through slow political bargaining, conflict, failure, and historical experience.

If so, then the deepest danger may not simply be how powerful AI becomes.

The deeper danger may be:

an expanding “civilizational speed gap” between technological capability and humanity’s capacity for social cooperation.

In the past, major technologies such as the steam engine, electricity, industrialization, nuclear technology, and the internet could disrupt old structures of social cooperation and then force humanity to adjust.

But AGI may be the first technology for which:

the gap between technological capability and social institutions becomes so large that humanity may no longer be able to afford the old pattern of “being broken first, then learning how to rebuild.”

This time, what is crushed may not merely be a particular institution.

What may be crushed is humanity’s very opportunity to continue learning and rebuilding civilization.

I have noticed that although you remain broadly optimistic about artificial intelligence and the future, you do not ignore the risks.

You have particularly discussed the dual-use nature of AI:

the same intelligence that can help solve disease, poverty, and other major human problems can also be used for war, attack, deception, and destruction.

You have also expressed a judgment that particularly caught my attention:

to keep AI safe, humanity must protect and improve its governance and social institutions.

This is very close to a conclusion I have reached independently over many years.

But I would like to take the question one step further:

What exactly must be improved when we say that “governance and social institutions” need improvement?

More regulation?

More laws?

New international organizations?

Or do we need to investigate something more fundamental:

whether the basic procedures through which human beings form relationships of social cooperation contain structural defects of their own?

You have argued that we have, in effect, already entered a kind of human-machine civilization.

And as brain-computer interfaces, artificial intelligence, and other technologies continue to merge, the boundary between humans and machines may change even further.

This leads me to a question that I believe is extremely important:

If AI increasingly becomes “part of us,” then what ultimately determines the nature of future AI civilization may not be only the algorithms inside AI, but the cooperative structure according to which the entire “human + AI” system operates.

Suppose the basic operating logic of human society continues to rely heavily upon:

competition among states;

competition among corporations;

struggles for power;

conflicts of interest;

mutual deterrence;

information manipulation;

military competition;

and ever-increasing defenses created by distrust.

What happens when AI increases the capabilities of every participant by a hundredfold or a thousandfold?

It may, of course, enormously increase humanity’s ability to create wealth, cure disease, and understand the universe.

But it may simultaneously increase:

the ability of each state to suppress or overpower other states;

the ability of each organization to control other organizations;

the ability of malicious actors to cause destruction;

and the ability of every participant in society to harm others in zero-sum competition.

This brings me to the question that I have been hoping humanity will discuss:

Even if AI itself never loses control, if the human competitive system amplified by AI loses control, can we still say that AI is safe?

I deeply respect your optimism about the future.

You have pointed to the nuclear era as evidence that humanity has shown an ability to avoid the most catastrophic outcomes.

I sincerely hope that judgment is correct.

But I would like to raise an additional question.

Nuclear weapons, despite their enormous destructive power, do not themselves continuously improve scientific research capacity, economic capability, information-processing power, political influence, and strategic decision-making.

Moreover, once nuclear weapons are built, they can be placed under the control of a relatively small number of actors who may use them primarily as instruments of deterrent balance, with all sides having powerful incentives not to use them.

AGI may be fundamentally different.

It may enter almost every subsystem of society while simultaneously increasing the capabilities of all participants in competition.

And unlike nuclear weapons, AGI may not remain a power that can be confined to the control of a small number of comparatively rational actors. It may eventually become accessible to vast numbers of people and organizations.

Its capabilities may also cross barriers that today appear secure and operate in ways outside ordinary expectations — potentially including the capacity to undermine systems that currently safeguard other forms of extreme technological power.

So I ask:

Is humanity’s past success in controlling nuclear weapons sufficient evidence that our existing institutions can also preserve stability within a continuously self-accelerating “human-AI system”?

I do not know the answer.

But I believe the question deserves serious examination.

Over the past twenty years or more, I have gradually developed a hypothesis at the level of natural philosophy, which I provisionally call:

the “Procedural Determination Hypothesis.”

It asks:

Are the properties ultimately displayed by a complex system determined not only by the elements that compose it, but also, to a significant extent, by the procedures through which those elements form relationships, the structures created by those relationships, and the way those structures establish dynamic equilibrium through feedback?

In its simplest form:

procedure → relationship → structure → dynamic feedback → system properties and outcomes.

I do not believe this hypothesis has been proven.

I want it to be criticized, experimentally tested, and falsified if it is wrong.

Several recent developments in neuroscience have particularly interested me.

For example:

the whole-brain fruit-fly connectome and computational models based on its real connectivity indicate that, even when individual neuron models are greatly simplified, preserving substantial portions of the real relational structure still allows many functions to be reproduced or predicted;

and when human cortical organoids are transplanted into rat brains, the human neurons can mature, form functional connections, and influence rat behavior, yet the animal does not thereby begin displaying recognizably human behavior.

These experiments certainly do not prove my hypothesis.

But they make me increasingly hope that the scientific community will investigate the following question:

How much of the overall nature of a complex system resides in “what the elements are,” and how much resides in “how those elements form relationships and structures”?

If we change the combinations, connections, and feedback procedures among elements without changing the elements themselves, how does the nature of the system change?

Conversely, if we change some of the elements but embed them within the original relational structure, what happens to the system’s properties?

What ultimately matters most to me is extending this question toward:

human social cooperation.

I provisionally call this direction:

“Human Social Cooperation Engineering.”

I do not mean creating a new ideology.

Nor do I mean designing a system that one person could impose upon all humanity.

What I want to discuss is:

Can human social cooperation itself be studied scientifically, as we study other complex systems?

For example:

What kinds of procedures allow independent individuals to form stable relationships?

What kinds of institutional feedback can prevent power, wealth, violence, or information from continuously concentrating toward extremes?

Why can every participant make a locally rational decision, yet collectively produce a catastrophe that none of them wanted?

Why do states all say they want peace, yet repeatedly find themselves structurally pushed toward arms races?

Why might companies understand that an AGI race is dangerous, yet no individual company feels able to stop?

Why does one country’s fear that another country will gain the lead ultimately cause all countries to accelerate?

If these problems share common structural causes, then:

can we turn those structural causes into scientific questions that can be modeled, compared, experimented upon, falsified, and optimized?

This is also one of the greatest insights your Law of Accelerating Returns has given me.

If:

technology experiences accelerating returns;

but:

institutions of social cooperation do not experience accelerating renewal;

then is human civilization creating an increasingly dangerous divergence?

On one side:

ever-faster growth in capability.

On the other:

social mechanisms that still update slowly and largely through crisis, conflict, and historical cost.

That is why I especially want to ask you:

Before the Singularity truly arrives, does humanity also need to create an “accelerated learning mechanism for social cooperation”?

So that humanity no longer has to:

experience disaster first, and learn cooperation afterward;

but can instead:

simulate in advance → identify structural risks in advance → modify procedures in advance → search for dynamic equilibrium in advance.

If this were possible, then “Human Social Cooperation Engineering” might be understood as:

an attempt to give the learning speed of human social cooperation, for the first time, a chance to catch up with the speed of technological progress.

I have also noticed an important difference between your overall outlook and mine.

You believe that, despite its risks, technology has on balance continued to improve life, health, prosperity, and human well-being, and that humanity can use new intelligence to solve problems created by new technologies.

I do not wish to reject your optimism.

On the contrary, I want to ask:

Can we give that optimism a more secure structural foundation?

If you believe that:

future AI will ultimately reflect human values,

then I would like to ask a further question:

Should we first study through what institutions, procedures, and relational structures human values are transmitted into the broader “human-AI civilization”?

Because “human values” are not a single variable.

Human beings possess:

cooperation and altruism;

but also war and hatred;

creativity and compassion;

but also domination, deception, and fear.

So perhaps the true question is not:

Will AI reflect human values?

but:

What structures within human society will cause which human values to be systematically amplified in the AI era?

I believe this is one of the most important points of connection between your technological optimism and my Procedural Determination Hypothesis.

The practical importance of this connection is the following:

if, through serious examination of the Procedural Determination Hypothesis, we can first clarify a foundational safeguard for humanity’s values — one capable of serving as an alignment standard amid humanity’s many conflicting value judgments — then the foundations of human social cooperation may gain a more stable and sustainable safeguard.

That, in turn, might provide a stronger structural foundation for the realization of the optimistic future you envision.

Therefore, I am taking the liberty of writing to you not because I expect you to accept my theory.

What I truly hope for is:

First, please help judge:

Does the widening gap between the speed of technological acceleration and the speed at which institutions of social cooperation are updated deserve to be regarded as an independent and foundational risk of the AGI era?

Second, if you believe this question deserves study, I hope that you or appropriate researchers might review my ideas concerning the “Procedural Determination Hypothesis” and “Human Social Cooperation Engineering,” and identify either their most fundamental errors or the aspects that may deserve further experimental investigation.

Third, if preliminary examination suggests that the question does have value, I hope you might use your influence in futures research, artificial intelligence, and the study of the Singularity to help encourage an interdisciplinary discussion:

Before human intellectual capability expands exponentially, must humanity also develop a way to learn, adjust, and stabilize its structures of social cooperation much more rapidly?

In other words:

If the “Singularity” means an accelerating leap in human capability, must our capacity for social cooperation undergo a corresponding leap of its own?

Based on my past experience and present circumstances, while I am still able to communicate freely with the international community, I hope to place this question before people who are genuinely capable of judging it.

I am not asking you to believe:

that I have found the answer.

What I truly hope you will help judge is simply:

Have I at least raised a question worthy of serious scientific examination?

If it is wrong,

please help prove it wrong.

If it deserves investigation,

please do not allow it to lose the opportunity to be seriously evaluated simply because the person raising it is an ordinary elderly Chinese man without a research institution or research funding.

Thank you very much for taking the time to read this unusual letter.

Your work on the accelerating returns of technology has helped humanity see ever more clearly that:

the future is approaching us faster and faster.

What I now want to ask is:

Can humanity also begin to learn how to cooperate faster?

May the intelligence that grows in the future amplify not only humanity’s capabilities, but also humanity’s wisdom in cooperation.

With sincere respect,

Pen name: Jin Guyu
Former Senior Judge of a Local Court in China
Author of the Letters to Everyone in the World series

Email: ——

Attachments

1. The Procedural Determination Hypothesis — A One-Page Challenge

2. The AI Era That Could End Civilization Before We Even Understand What Went Wrong Is Here — An Open Global Challenge to Bill Gates

3. A Request for Humanity to Jointly Study and Experiment with Strategic Directions for Upgrading Civilization: Who Can Guarantee That No One Will One Day Use AGI to Launch a “9/11” Against All Humanity?

4. To Demis Hassabis: Must Humanity Upgrade Its Structures of Social Cooperation Before AGI Arrives?

5. A Renewed Request Concerning Human Cooperation, Civilization, and the Risks of AGI — Letter to Elon Musk

The text that follows is the original Chinese version of this letter.

致 Ray Kurzweil：如果技术遵循加速回报，人类社会合作制度能否跟得上奇点的速度？

尊敬的 Ray Kurzweil 先生：

您好！

我笔名金谷雨，是一名来自中国的老人，曾任中国地方法院高级法官。

最近，我认真了解了您在 《The Singularity Is Nearer: When We Merge with AI》 中进一步阐述的思想。

您关于**“加速回报定律”（Law of Accelerating Returns）**的论述，尤其引起了我的注意。

按照您的长期判断，信息技术并不是以普通线性速度进步，而是在不断利用上一轮技术成果推动下一轮突破，因此呈现持续加速的发展趋势。

您仍然认为，人类级人工智能乃至AGI可能在非常接近的未来到来，而随着人工智能、脑机接口、纳米技术等进一步发展，人类自身的智能能力也可能与机器智能越来越深地融合。

如果您的基本判断成立，那么我们面对的并不是：

一种更先进的普通工具。

我们面对的可能是：

人类历史上第一次出现一种能够反过来加速自身进步，并且不断提高人类和组织行动能力的技术力量。

这正是我为什么特别希望写信给您。

因为，在阅读您的思想以后，我产生了一个问题：

如果技术遵循“加速回报定律”，那么人类社会合作制度的更新，是否也存在一种能够与之匹配的“加速机制”？

如果不存在，会发生什么？

我长期从法律、司法、中国社会现实以及人类文明演变中思考一个问题：

为什么人类建立的社会合作结构，总是如此难以真正稳定和可持续？

在我看来，人类历史上长期存在一个非常重要但容易被忽视的现象：

科学技术迭代的速度，通常快于人类社会合作制度和相关理论更新的速度。

新的技术不断创造新的能力。

新的能力打破原来的生产方式、利益关系、权力关系以及人与人之间的合作方式。

但是，人类很少能够在技术革命到来以前，就提前建立好与新能力相匹配的社会合作制度。

更多时候发生的是：

技术先突破；

旧的社会合作结构随后失配；

社会发生巨大震荡、冲突甚至战争；

旧制度在危机中被迫改变；

新的制度和合作规则才逐渐形成；

然后人类重新获得一个暂时的平衡。

也就是说，在很长的历史时期内：

不是制度引导技术安全进入社会，而是技术一次又一次通过冲击旧制度，逼迫人类重新学习合作。

过去这种滞后虽然给人类造成过极其巨大的代价，但人类通常仍有机会：

经历灾难 → 学习 → 修改制度 → 重建秩序。

但是，当我把这个历史现象与您的“加速回报定律”放在一起以后，我产生了一个非常强烈的担忧：

如果技术进步越来越快，而社会合作制度的调整仍然主要依靠危机以后再纠错，那么两者之间的速度差是否会越来越大？

也就是说：

技术可能指数式前进，而人类社会制度仍然主要依靠缓慢的政治博弈、冲突、失败和历史经验来更新。

如果是这样，那么真正危险的并不只是AI有多强。

真正危险的可能是：

技术能力与社会合作能力之间正在形成一个越来越大的“文明速度差”。

过去，蒸汽机、电力、工业化、核技术、互联网等重大技术可以打碎旧的社会合作模式，然后迫使人类再调整。

但是AGI可能是第一次：

技术跨越社会制度的幅度大到，人类未必还能承受“先被打碎，再学习重建”的代价。

这一次，被碾碎的也许不只是某一种制度。

被碾碎的可能是整个人类继续学习和重新建设文明的机会。

我注意到，您虽然对人工智能和未来总体保持乐观，但并没有忽视风险。

您特别谈到了人工智能的双重用途：

同一种智能能力，既可以解决疾病、贫困和其他巨大问题，也可以被用于战争、攻击、欺骗和破坏。

而您提出过一个让我非常注意的判断：

要使AI保持安全，人类必须保护并改善自己的治理和社会制度。

这一点与我长期形成的判断非常接近。

但我希望再往前追问一步：

“改善治理和社会制度”究竟应该改善什么？

是增加更多监管？

增加更多法律？

增加新的国际组织？

还是我们需要研究一个比这些更底层的问题：

人类形成社会合作关系的基本程序本身，是否存在结构性缺陷？

您曾经提出，我们事实上已经进入一种“人—机器文明”。

而未来随着脑机接口、人工智能和其他技术继续融合，人类和机器之间的界限还会进一步改变。

这使我产生了一个我认为非常重要的问题：

如果AI最终越来越成为“我们的一部分”，那么真正决定未来AI文明性质的，也许就不只是AI内部的算法，而是“人类 + AI”这一整个系统按照什么样的合作结构运行。

假如今天人类社会的基础运行方式仍然大量依靠：

国家之间的竞争；

企业之间的竞争；

权力争夺；

利益冲突；

相互威慑；

信息操纵；

军事竞争；

以及因为不信任而不断增加的防御，

那么，当AI使所有参与者的能力提高百倍、千倍以后，会发生什么？

它当然可能极大提高人类创造财富、治疗疾病和理解宇宙的能力。

但它也可能同时提高：

每一个国家压制其他国家的能力；

每一个组织控制其他组织的能力；

每一个恶意行为者造成破坏的能力；

每一个社会主体在零和竞争中伤害对方的能力。

于是就出现了我一直希望全人类讨论的问题：

即使AI本身始终没有失控，如果被AI放大的人类竞争系统失控了，我们还能说AI是安全的吗？

我非常尊重您对未来的乐观。

您以核武器时代为例，认为人类最终显示出了避免最严重灾难的能力。

我也希望这个判断是正确的。

但是我想提出一个补充问题：

核武器虽然具有极大的毁灭性，但它并不会自己不断提高科研能力、经济能力、信息能力、政治影响能力和战略决策能力。特别是，核武器造出来可以被少数更具理性的人控制成为威慑性的平衡工具，大家都不使用。

而AGI可能不同。

它可能进入几乎每一个社会子系统，同时提高参与竞争者的能力。而且其不可能像核武器一样能被少数更具理性的人控制，而使将成为每个人都能获取的能量，且这个能量将冲破更多的壁垒且在常识性意外的层面发挥能量，这里也包括突破少数人控制的核能的保障体系而予以全部瞬间释放.......

所以：

过去人类成功控制核武器的经验，是否足以证明我们也能在一个能力不断自我加速的“人类—AI系统”中继续依靠原有制度维持稳定？

我不知道答案。

但我认为这个问题值得被认真检验。

过去二十多年，我逐渐形成了一个自然哲学层面的假说，我暂时称之为：

“程序定质假说”（Procedural Determination Hypothesis）。

它试图研究：

一个复杂系统最终呈现出的性质，是否不仅取决于构成它的元素本身，更在很大程度上取决于这些元素按照什么程序形成关系、这些关系形成什么结构，以及这些结构如何通过反馈形成动态平衡。

用最简约的方式表达就是：

程序 → 关系 → 结构 → 动态反馈 → 系统性质与结果。

我并不认为这个假说已经得到证明。

我希望它被批评、被实验、被证伪。

最近一些神经科学研究尤其引起我的兴趣。

例如：

果蝇全脑连接组以及基于真实连接结构建立的计算模型显示，即使单个神经元模型被大幅简化，只要保留大量真实的关系结构，系统仍可以恢复和预测相当多的功能；

而人类皮层类器官植入大鼠脑以后，人类神经元能够成熟、形成实际功能连接并影响大鼠行为，但整个动物并没有因此表现出“人类行为”。

这些实验当然没有证明我的假说。

但它们使我越来越希望科学界能够检验一个问题：

复杂系统的整体性质，有多少储存在“元素是什么”之中，又有多少储存在“元素如何形成关系和结构”之中？

如果改变元素之间的组合、连接和反馈程序，而不改变元素本身，系统性质会发生什么？

反过来，如果改变部分元素，却把它们放进原来的整体关系结构，系统性质又会怎样？

而我最终真正关心的，是把这个问题进一步推进到：

人类社会合作。

我暂时把这个方向称作：

“人类社会合作工程”（Human Social Cooperation Engineering）。

我的意思不是建立一种新的意识形态。

更不是由某个人设计一套制度以后强迫全人类接受。

我想讨论的是：

能不能像研究其他复杂系统一样，对人类社会合作本身进行科学研究？

例如：

什么样的程序使独立的个体形成稳定关系？

什么样的制度反馈能够阻止权力、财富、暴力或者信息不断向极端集中？

为什么每一个参与者都可能作出局部理性的选择，最后却形成所有人都不希望看到的集体灾难？

为什么国家都说自己希望和平，却又不断被结构推向军备竞赛？

为什么企业都可能知道AGI竞赛存在风险，却没有任何一家企业敢真正停下来？

为什么一个国家担心另一个国家领先以后，最终所有国家都会加速？

如果这些问题具有共同的结构原因，那么：

我们能不能把这些结构原因变成可以建模、比较、实验、证伪和优化的科学问题？

这也是您的“加速回报定律”给我带来的最大启发之一。

如果：

技术存在加速回报；

但：

社会合作制度不存在加速更新；

那么，人类文明是否正在形成一个越来越危险的剪刀差？

一边是：

越来越快的能力增长。

另一边是：

仍然缓慢、依靠危机、冲突和历史代价才能更新的社会合作机制。

所以我特别想请您判断一个问题：

奇点真正到来以前，人类是否还需要建立一种“社会合作的加速学习机制”？

使人类不再必须：

先发生灾难，随后才学习合作。

而能够：

提前模拟 → 提前发现结构风险 → 提前修改程序 → 提前寻找动态平衡。

如果这能够实现，那么“Human Social Cooperation Engineering”或许可以被理解为：

让人类社会合作的学习速度，第一次有机会追赶技术进步速度的一种尝试。

我也注意到，您的总体判断与我有一个很重要的不同。

您相信技术虽然带来危险，但总体正在使生命、健康、财富和人类福祉不断改善，并且人类有能力利用新的智能解决新技术制造的问题。

我并不想否定您的乐观主义。

相反，我希望提出的是：

能否为这种乐观提供一个更加牢固的结构基础？

如果您认为：

未来AI最终会反映人类价值，

那么我想进一步问：

我们有没有必要先研究，人类价值究竟通过什么制度、程序和关系结构进入整个“人类—AI文明”？

因为“人类价值”并不是一个单一变量。

人类同时拥有：

合作和利他；

也拥有战争和仇恨；

创造和同情；

也拥有支配、欺骗和恐惧。

所以真正的问题也许不是：

AI会不会反映人类价值？

而是：

人类社会的什么结构，会使哪些价值在AI时代被系统性放大？

我认为，这是您的技术乐观主义与我提出的“程序定质假说”之间最值得讨论的连接点。这个链接最值得关注的现实意义是：我们如果能通过甄别“程序定质规律”，先将人类的价值实现的底线保障系统明晰化，成为人类杂乱价值判断的对齐标准，那么，人类的社会合作结构基础将有一个稳定可持续的保障，来为你的乐观展望的实现提供一个更加牢固的结构基础。

因此，我冒昧给您写这封信，不是希望您接受我的理论。

我真正希望得到的是：

第一，请您帮助判断：

“技术加速速度与社会合作制度更新速度之间不断扩大的差距”，是否值得被视为AGI时代一个独立而基础的风险问题？

第二，如果您认为这个问题值得研究，我希望您或者适当的研究人员能够审阅我提出的“程序定质假说”和“人类社会合作工程”设想，并指出其中最根本的错误或者值得继续实验的地方。

第三，如果经过初步检验以后，您认为这个问题确实具有价值，我希望您能够利用自己在未来研究、人工智能以及奇点思想领域的影响力，帮助推动一次跨学科讨论：

在人类智能能力发生指数级扩张以前，人类是否也必须建立一种能够更快速学习、调整和稳定自身社会合作结构的方法？

换句话说：

如果“奇点”意味着人类能力的加速跃迁，那么“社会合作能力”是否也必须拥有自己的跃迁？

基于我过去的经历和现实条件，我希望趁自己仍然能够自由与国际社会联系的时候，把这个问题交给真正有能力判断它的人。

我不要求您相信：

我已经找到了答案。

我真正希望您帮助判断的只是：

我是否至少提出了一个值得被科学界认真检验的问题？

如果它错了，

请帮助证明它错。

如果它值得研究，

请不要让它因为提出者只是一个没有研究机构、没有科研资金、已经年老的普通中国人，而失去一次被认真判断的机会。

感谢您愿意阅读这封不同寻常的信。

您关于技术“加速回报”的研究，使人类越来越清楚地看见：

未来正在越来越快地向我们走来。

我现在真正想追问的是：

人类自身学习合作的速度，能不能也开始加快？

愿未来不断增长的智能，最终放大的不仅是人类的能力，也包括人类合作的智慧。

谨致敬意！

笔名：金谷雨
曾任中国地方法院高级法官
《致全球每一个人的信》系列作者

Email: ——

附件

1. The Procedural Determination Hypothesis — A One-Page Challenge

2. The AI Era That Could End Civilization Before We Even Understand What Went Wrong Is Here — An Open Global Challenge to Bill Gates

3. A Request for Humanity to Jointly Study and Experiment with Strategic Directions for Upgrading Civilization: Who Can Guarantee That No One Will One Day Use AGI to Launch a “9/11” Against All Humanity?

4. To Demis Hassabis: Must Humanity Upgrade Its Structures of Social Cooperation Before AGI Arrives?

5. A Renewed Request Concerning Human Cooperation, Civilization, and the Risks of AGI — Letter to Elon Musk