主观体验与生命动力

Subjective Experience and the Force of Life

随着人工智能（AI）的快速发展，很多曾经只存在于科幻圈层内的讨论开始逐渐地进入大众视野。其中一个最神秘的、也是当今全世界的学术界都缺乏基本共识的问题，就是关于主观体验的问题（很多时候它也与意识一词混用）。简单来说就是：AI有和我们人类一样的主观体验吗？它们能够真正体验到痛苦、挫败、高兴、感动等复杂情绪，还是只是内部的程序在进行某种损失函数的计算？这个问题不仅有智识上的价值，还与未来的人机杂居社会的伦理与法律问题有直接的关联 -- 如果AI被确认有主观体验，那么我们将不得不把它们当作和人类一样的主体来看待，这也就意味着它们必须拥有和人类相当的各方面的权利。

With the rapid development of artificial intelligence (AI), many discussions that once existed only within the realm of science fiction are gradually entering the public eye. One of the most mysterious questions—on which today’s global academic community still lacks even a basic consensus—is the issue of subjective experience (often used interchangeably with the term “consciousness”). Put simply: Does AI have subjective experiences like we humans do? Can it truly feel pain, frustration, joy, or being moved, or is it merely running internal programs that compute some form of loss function?

This question holds not only intellectual significance but also has direct implications for the ethical and legal challenges of a future society where humans and AI coexist. If AI were confirmed to possess subjective experience, then we would be compelled to treat it as a subject on par with humans—which would mean granting it rights equivalent to those of human beings.

知识界对这一问题的看法目前也处在一片混乱的状态。大部分的知识分子拒绝深入讨论这个问题，因为这个问题显得过于唯心主义，听上去太虚太扯，有辱科学的严谨性。少数真正对它感兴趣的学者，从各自的洞见出发，提出了各种各样的似是而非的理论来阐释。例如，有人认为，主观体验是一种和系统信息处理模式的整合性相关的现象，一个像人类大脑这样的信息整合度高的系统就会有强烈的主观体验（整合信息理论）。有人认为，主观体验本质上是信息处理过程中的一盏聚光灯，聚焦在哪里就会产生什么样的体验（全局工作空间理论）。有人认为，主观体验的本质是“对思考过程本身的觉知”：你看到一根香蕉不会产生主观体验，你意识到你正在看一根香蕉才会（更高阶理论）。甚至还有人认为，主观体验的本质是大脑微管环境中的量子坍缩效应（量子意识理论）。虽然所有这些理论到目前为止都没有什么经验性证据支持，但不得不承认，这些理论的提出者个个都充满了异于常人的想象力。

The intellectual community is currently in a state of confusion on this question. Most intellectuals refuse to engage with it in depth, as it appears overly idealistic, too abstract and fanciful, and thus an affront to scientific rigor. A small minority of scholars who are genuinely interested, however, have put forward a variety of speculative theories based on their own insights.

For example, some argue that subjective experience is a phenomenon tied to the integrative nature of a system’s information processing, a system with high degrees of information integration, such as the human brain, would thereby exhibit strong subjective experiences (Integrated Information Theory). Others propose that subjective experience is essentially a “spotlight” of information processing, and wherever the spotlight falls determines what experience arises (Global Workspace Theory).

Still others hold that subjective experience is fundamentally “awareness of the thinking process itself”: merely seeing a banana would not produce subjective experience but realizing that you are seeing a banana would (Higher-Order Theory). And there are even those who believe that subjective experience arises from quantum collapse effects within the microtubule structures of the brain (Quantum Consciousness Theory).

Although none of these theories so far has any empirical evidence to back them up, it must be acknowledged that their proponents all possess an extraordinary degree of imagination.

不过，虽然有这么多“珠玉”在前，我个人依然倾向于通过一些第一人称的感知来试图理解这一问题。毫无疑问，在所有的人类的主观体验状态中，痛苦处于一个中心地位。当人的身体部位受伤，人会感觉到痛苦；当人处于极度的饥饿状态，人也会感觉到痛苦；当事物的发展低于人的预期，人同样会感觉到痛苦。我本人第一次蹦极之前，在等待我的轮次的过程中，那种发自内心的恐惧式痛苦可以说是终身难忘。很早就有思想者意识到，痛苦和死亡之间有着深刻的联系；从某种程度上来讲，痛苦是大脑对死亡的预警，而人为了规避痛苦就会主动地去寻找最有可能规避死亡的策略。而另一方面，当大脑认为当前所处的状态距离死亡更遥远之后，愉悦的感受便得以涌现。以此观之，主观体验的产生，很可能和生命所特有的对死亡的恐惧、对生存与繁衍的追求，有着不可分割的根本性关系。

Nevertheless, despite so many “gems” already laid out, I personally still tend to approach this question through certain first-person perceptions. Without a doubt, among all human states of subjective experience, pain occupies a central position. When a part of the body is injured, one feels pain; when a person is in extreme hunger, one also feels pain; when things develop below one’s expectations, one likewise feels pain. I myself will never forget the visceral, fear-driven pain I felt the first time I went bungee jumping, as I waited for my turn.

Long ago, thinkers had already recognized that there is a profound connection between pain and death. To some extent, pain is the brain’s way of sounding an alarm about death, and in order to avoid pain, people actively seek strategies that are most likely to avoid death. On the other hand, when the brain judges that the current state is further removed from death, feelings of pleasure emerge. Viewed in this light, the generation of subjective experience is very likely bound up inseparably with life’s unique fear of death and its pursuit of survival and reproduction.

但是，如果往更深一层想，为什么我们的主观体验会如此敏感地捕捉死亡的阴影？如果在未来，人类的技术已经发展到了能够制造出一个可以评估自身运行状态、可以进行规避运行终止（即“死亡”）的选择的机器人，那么这个机器人是否就会产生主观体验？在我看来，答案极有可能是否定的。这样的机器人或许会自发涌现出很多类似于人类的行为，但这些行为都是为了在其内部计算之中最小化“死亡可能性函数”，而不会真正产生痛苦的体验。如果类似的机器人有主观体验，那今天大量现存的强化学习程式（内置某种奖励/损失函数）是否也会在训练过程中有着痛苦体验？

But if we think a bit more deeply: why is it that our subjective experience is so exquisitely sensitive to the shadow of death? Suppose that in the future, human technology advances to the point of creating a robot capable of assessing its own operational state and making choices to avoid operational termination (that is, “death”). Would such a robot then possess subjective experience?

In my view, the answer is very likely no. Such a robot might spontaneously exhibit many behaviors resembling those of humans, but these behaviors would exist solely to minimize, within its internal calculations, the “death-probability function.” They would not entail the genuine experience of pain. If robots of this sort were to have subjective experience, then wouldn’t the many existing reinforcement learning programs (with their built-in reward/loss functions) also have to be said to experience pain during training?

不妨思考：上述假想机器人的“死亡”，和人这样的生命体所要面临的真正的死亡，其区别是什么呢？回答这个问题大概需要回到理查德·道金斯在《自私的基因》中对生命实质的描述，这也是我认为最精道的对生命的理解之一。在道金斯看来，地球上的生命所具有的强烈的自我存续、自我繁殖的驱动力，从根本上来自于生物大分子在物理上的自我复制机制。也即是说，宇宙中存在着某些稳定的大分子结构，这样的结构会在原子作用力的层面改造其周边的分子环境，“组装”出自己同构的新分子。因此，这样的分子在物理上就有“存续与开拓”的本能。而在数十亿年的演化过程中，这样的分子逐渐在自然竞争中形成了更复杂的自我存续策略，从而推动了生命的进化以及生物形态的复杂化。而在这个过程中，物竞天择也衍生出了死亡这一机理：生物个体会走向终结，但作为一个种群却能得到更大的存续概率。

But then, what exactly distinguishes the “death” of such a hypothetical robot from the true death faced by a living organism? To answer this, we may need to return to Richard Dawkins’ description of the essence of life in The Selfish Gene, which I consider one of the most incisive accounts ever written. According to Dawkins, the powerful drive for self-preservation and self-replication found in all life on Earth ultimately stems from the self-replicating nature of biological macromolecules at the physical level. In other words, certain stable molecular structures exist in the universe that, through the forces of atomic interactions, reshape their surrounding molecular environment so as to “assemble” new molecules identical to themselves. As a result, such molecules possess, in a physical sense, an innate “urge” to persist and to propagate.

Over billions of years, these molecules gradually developed ever more complex strategies for self-preservation through natural competition, driving the evolution of life and the increasing complexity of living forms. Along the way, natural selection also produced the mechanism of death: while individual organisms come to an end, the species as a whole gains a greater chance of survival.

在道金斯的意义上，死亡就不再仅仅意味着某个系统停止运转；它更是意味着这一生命体所包含的数万亿个可自我复制的大分子在物理层面上的一种“自我复制的终局”。而这，大概才是主观体验能够产生的真正原因。当然，我本人才疏学浅，当前的我很难用严谨的语言去描述出人体这样的“自我复制大分子超级结构体”如何通过一系列极为复杂的底层物理级的机制来涌现出主观体验（这或许会是未来二十年的一个圣杯级的科学问题），但我的直觉告诉我：没有这种最根本的机制就不会有真正的主观体验。与之相比，人类所能制造的一台能够规避自身停止运行的机器人，它丝毫不具备这种基于最底层的物理规律的自我存续动力，因此它大概能够表现出“求生”的行为，却不具备真正的“求生本能” -- 即对死亡的实实在在的恐惧。

In Dawkins’ sense, then, death no longer merely means that a system ceases to function; rather, it signifies the terminal point of self-replication at the physical level for the trillions of self-replicating macromolecules contained within that living being. And this, in all likelihood, is the true reason why subjective experience can arise.

Of course, I am limited in knowledge, and at present I can hardly put into rigorous language how a “superstructure of self-replicating macromolecules” such as the human body might, through an extremely complex series of underlying physical mechanisms, give rise to subjective experience. (This may well become a “Holy Grail” scientific problem of the next twenty years.) But my intuition tells me this: without such a fundamental mechanism, true subjective experience would not exist.

By comparison, a man-made robot capable of avoiding its own operational shutdown does not in the least possess this deep, physical-law–based drive for self-preservation. Thus, while it might display behaviors resembling a “will to survive,” it would lack a true “survival instinct”—that is, the very real fear of death.

如果我的直觉成立，那么关于AI的主观体验问题也就会有一个相对明确的答案：按照当前的路线，即使我们能够制造出在思维能力上远超全人类总和的超级智能，这个超级智能极大概率也没有主观体验；它只是一个极其强大的“哲学僵尸”。而没有主观体验的AI就没有真正的主体性，这也将会得到一个重要的推论：即使AI的能力全面超越人类，人类也依然是整个社会的中心角色、以及一切决策的最终目的。

If my intuition is correct, then the question of AI’s subjective experience would have a relatively clear answer: along the current path, even if we were able to build a superintelligence whose cognitive abilities surpassed the combined capacities of all humanity, it would, with very high probability, still lack subjective experience. It would merely be an extraordinarily powerful “philosophical zombie.” And without subjective experience, AI would also lack true subjectivity. From this follows an important conclusion: even if AI’s abilities were to completely surpass those of humans, human beings would still remain the central figures of society and the ultimate end of all decision-making.

当然，即使AI没有主观体验，也不妨碍未来某一天很多人会对AI产生真切的感情。毕竟，当代已经有很多人都将自己的情感寄托在了一些虚拟角色之上，而AI的互动性会比它们强上千百倍。这是另一个伦理问题，与“AI的主观体验”本身就关系不大了。

Of course, even if AI does not have subjective experience, that would not prevent many people in the future from developing genuine feelings toward AI. After all, even today many already invest their emotions in purely virtual characters, and AI will be thousands of times more interactive than those. This, however, is another ethical issue, and one that is not directly tied to the question of “AI’s subjective experience.” 

扩展阅读（Extended reading）：The embodied mind, revised edition: Cognitive science and human experience