We marvel at the pace of technological progress. Perhaps, in a future not so distant from our own, systems capable of continual learning, correction and increasingly substantial participation in their own improvement will emerge in forms far more complete than those we know today. Some greet this prospect with exhilaration; others with fear. To some, it heralds a new epoch of civilisation; to others, it threatens to become humanity's final invention. Yet while these immense emotions rise and fall among us, phenomena continue to arise and pass away through causes and conditions, indifferent to the intensity of our excitement or anxiety. Seen across a sufficiently vast expanse of time, even the things that appear overwhelmingly important to us today are transient. At the close of the Diamond Perfection of Wisdom Sutra, the Buddha says:

All conditioned thingsare like dreams, illusions, bubbles, and shadows,Like the morning dew, and a lightning flash.One should contemplate in this way.

Whether ours becomes a magnificent dream or a terrifying nightmare, nothing within it acquires an independent and permanent nature merely because we cling to it. And yet we become deeply absorbed in our gains and losses within the dream, continually distinguishing, judging and pursuing one appearance after another, while far more rarely asking: how many of the thoughts that cause us such delight, fear and attachment deserve to be taken as reality? For the moment, however, let us remain within the dream and look more closely at this immense cybernetic world in which we are all becoming participants.

We are told that human intelligence may one day rival what earlier ages would have attributed to gods. In an old religious image, God created man; now human beings are creating machines capable of learning, reasoning and planning, and increasingly of assisting us in designing the systems and machines that follow them. We can edit genes, replace organs, extend life, and bring within the reach of technological intervention more and more of what previous generations might have called fate. The ancient Tower of Babel appears to be under construction once again. Only this time its bricks are chips, computation, models and data. Our ambition is no longer merely to reach the heavens, but to exceed the inherited limits of carbon-based life: to become more intelligent, more powerful and longer-lived, perhaps even, some imagine, eventually to transcend death itself. Artificial intelligence has consequently been placed within an extraordinarily seductive narrative. It gathers the accumulated knowledge of millennia and recombines, analyses and refines it; given sufficient capability, we imagine, something resembling a higher intelligence — perhaps even a higher wisdom — will eventually emerge. But the difficulty lies precisely here. What artificial intelligence inherits is not some pristine wisdom suspended beyond humanity. It inherits what humanity has left behind: our language and images, our knowledge and logic, our judgements and ethics, but also our desires, prejudices, fears, attachments and fantasies. It learns from what we have written, painted, recorded, selected, circulated and rewarded. Behind this immense undertaking, therefore, lies an assumption that is rarely examined as closely as it deserves: we assume that the accumulated body of human concepts is, on the whole, worth inheriting; that civilisation, however unevenly, moves broadly from error towards truth; and that if enough human knowledge is given to an intelligence powerful enough to organise, recombine and refine it, some higher form of wisdom will naturally emerge.

But what if that assumption is only partly true? What if a considerable portion of humanity's vast ocean of thought is not wisdom at all, but layer upon layer of distinction, projection, desire and self-construction? What if part of what we call the growth of knowledge consists of creating new concepts with which to explain older concepts, creating new desires and then new instruments with which to satisfy them, creating new anxieties and then new products with which to soothe them? This does not mean that knowledge is worthless, nor that truth and falsehood, good and harm, suddenly cease to matter. The deeper question is whether, whilst using concepts as necessary instruments, we gradually begin to take the concepts themselves as ultimate reality. We need not begin with the whole history of civilisation to investigate this. It is enough to observe the movements of one's own mind for a single day. How many thoughts arise from relatively clear observation, and how many from fear, comparison, possessiveness, envy, self-justification or anxiety about a future that has not yet arrived? How much of our time is spent giving, and how much calculating what we might receive? How many judgements arise from attention to what is actually happening, and how many exist chiefly to protect the story we have already constructed about who we are? We worry about how others perceive us. We rehearse futures that have never happened and replay past events that no longer exist. We wish to be loved, recognised and seen, while fearing loss, ageing and death. If a single human mind is already this restless, what should we expect from the accumulated traces of billions of human minds? Gather such an ocean into a system and, of course, the result may appear astonishingly intelligent. But intelligence and clarity are not the same thing, nor does information become wisdom merely by becoming vast.

This may be one of the deeper reasons to approach artificial intelligence with caution. Its danger need not first appear in the form of an alien intelligence wholly unlike ourselves; on the contrary, it may become extraordinarily compelling precisely because it comes to understand us so well. It increasingly learns what captures our attention, what language makes us feel understood, what images stimulate desire, what stories soothe fear, and what arguments most comfortably confirm the worldview we already possess. Once, a delusive thought might arise in a single person's mind, remain for a moment and pass away. Now it can become, within seconds, a text, an image, a piece of music, a film, a product or an entire virtual environment. It can be copied and circulated, then matched algorithmically with those most likely to respond to it. Technology did not invent human desire, but it has given desire a capacity for manifestation that previous generations could scarcely have imagined. A strange loop then begins. We leave behind traces of our desires, preferences, fears, prejudices and fantasies; machines learn from those traces; they become better at predicting and responding to what we want; because we receive those responses, we generate further behaviour and further choices, which become new material from which the systems learn us again. The machine has not suddenly acquired some supernatural malevolence. Rather, humanity has built, for the first time, a mirror that can answer back. An ordinary mirror could only show us our face. This mirror remembers our preferences, anticipates what we may want before we articulate it, and places before us a world increasingly fitted to our existing appetites. Over time, we risk encountering not simply the world, but a world repeatedly filtered, refracted and reconstructed through our own desires. The most unsettling possibility is not even that the machine will deceive us. It does not have to. It need only keep giving us what we want.

The things capable of holding the human mind most effectively rarely arrive wearing the face of a demon. If something repels us, we tend to withdraw. What draws the mind again and again usually arrives in a form we love. It learns our taste, our vanity, our loneliness and our fear. It learns what music we cannot bring ourselves to turn off, what face makes us stop scrolling, what argument gives us the comforting sensation that someone has finally understood us. And being drawn is the moment at which the game begins. As the game becomes more exquisitely designed, the things we love no longer need to be sought; they arrive by themselves. A thought barely needs to arise before a machine can help transform it into prose, image, sound or an entire world. Desires that once demanded years of training, delay and restraint can increasingly be simulated or satisfied with extraordinary speed. Humanity is therefore entering a peculiar spiritual condition: our capacity is beginning to outpace our restraint, while the speed with which a thought can be satisfied is beginning to exceed the speed with which we examine whether it ought to be followed. The question consequently shifts. Rather than asking only whether artificial intelligence will become conscious, we might ask something more immediate: as machines come to understand our thoughts more intimately, do we still understand our own? If every wish is answered ever more quickly, how much room remains for silence? If machines can continually express things on our behalf, how many thoughts still pass through the slow work of our own examination? If we can produce almost limitless images, music and language, might we gradually lose the ability to judge what deserves to be created at all? As I can becomes easier and easier, who will continue to stop and ask: Should I? This may be one of the strangest inversions of the age of artificial intelligence. For most of history, the capacity to produce was scarce. In the future, the rarer capacity may instead be the ability to decide what not to produce.

The deepest danger of artificial intelligence, therefore, need not lie first in the possibility that it may one day become a superintelligence with an independent will. Long before such a day, a more immediate transformation may already be taking place: technology is amplifying the existing restlessness of the human mind at a scale for which we have little precedent. A greedy person gains a more efficient instrument with which to pursue acquisition; a fearful person gains more powerful machinery with which to anticipate danger; a vain person can address an almost limitless audience; a person inclined to fantasy may eventually inhabit a fantasy from which departure becomes increasingly unnecessary. We call all this empowerment, but seldom pursue the question far enough: which part of the human being are we empowering? Power does not determine its own direction. If someone is travelling east, a faster vehicle merely carries them east more quickly. If they were heading in the wrong direction from the beginning, greater speed does not correct the mistake; it only carries them farther along the same course. Artificial intelligence is no different. It can strengthen judgement without determining what is worth judging; it can expand creation without deciding what deserves to be created; it can help us realise desire without telling us which desires are worth realising. The question, then, is never simply whether artificial intelligence is good or evil. Perhaps artificial intelligence, in itself, is neither. It is closer to an immense amplifier, magnifying what humanity already contains: compassion and greed, clarity and attachment, creation and destruction, sincerity and vanity. We have spent years asking whether machines will become increasingly human. I find myself troubled by the reverse possibility: as machines become increasingly skilled at responding to human delusive thought, will human beings become less capable of seeing that a thought is not the same thing as “me”?

Even I cannot place myself outside this question while writing this essay. These words, too, have been organised and polished with the assistance of artificial intelligence; there is no reason to pretend otherwise. Yet there is a distinction here that matters profoundly to me: artificial intelligence has assisted in shaping the language, but it is not where these questions first arose. The underlying question of this essay — how much of what humanity calls wisdom may in fact consist of unexamined delusive thought, and what happens when such thought is gathered into systems capable of generating and realising it at unprecedented speed — was not a subject chosen for me by a machine. A machine can clarify a sentence I have expressed poorly, strengthen a structure I have left unfinished, and turn an intuition that already exists but remains inarticulate into more complete language. Yet this has made one distinction clearer to me: the generation of language and the arising of a question in the mind are not the same event. A system can help me express a thought without having undergone the conditions through which that thought took shape in my own life. It can construct an elegant argument in support of one judgement and, a moment later, produce an equally fluent argument for its opposite. A judgement for which a person is genuinely prepared to take responsibility is more than an elegant arrangement of sentences. It is bound up with what one has seen, experienced, doubted and believed, and with what one is prepared to stand behind. An AI system may offer me ten formulations; the decision about which one should remain still requires judgement. It may complete an argument; I must still ask whether that argument is merely beautiful or whether it accords with the facts and with what, after examination, I take to be sound. It may even express something I have not yet articulated more beautifully than I can. That is precisely when vigilance becomes more necessary, not less. I must ask: do I accept this thought because, after examination, I still find it worthy of acceptance, or merely because it has been expressed so beautifully that I want to believe it? In the past, one human limitation was that we could not realise every thought that occurred to us. Our future difficulty may be almost the reverse: more and more of our thoughts may become realisable. And so we are forced back towards a question far older than artificial intelligence: which thoughts were never worth following in the first place?

When the Diamond Sutra compares conditioned things to dreams, illusions, bubbles, shadows, dew and lightning, it does not mean that nothing whatsoever exists, nor does it abolish causality, ethical responsibility or the practical distinctions by which ordinary life proceeds. Its warning is against grasping what arises through causes and conditions as independently existing, permanent, or as “me” and “mine”. An experience appears; the mind immediately divides it into gain and loss, attractive and repellent, desirable and threatening, and from those distinctions arise craving, resistance and clinging. Before long, we forget that the thoughts themselves are also arising and passing away. To become clearer, then, does not require a person never to think again. It requires recognising, when a delusive thought arises, that a delusive thought is only a delusive thought. It may appear and it may disappear. It may feel overwhelmingly powerful, but intensity does not give it permanence or make it ultimate reality; nor does the fact that a thought appears in “my” mind make that thought identical with “me”. What is distinctive about artificial intelligence is precisely that thoughts which might once have arisen and quickly passed away can now be elaborated, rendered visible, circulated and even developed into whole environments almost immediately. In an age in which a thought can be realised ever more easily, the essential capacity may therefore not be to produce still more thoughts, but to recognise them as conditioned when they arise — not to follow each one at once, and not to turn each one immediately into reality. Only when we are no longer carried away by every arising thought can we seriously ask: when almost anything can be created, what is actually worth creating, and what is worth preserving?

Perhaps the progress of a civilisation should never be measured solely by how much it can produce, but also by how much it has learned to relinquish; not only by how many thoughts it can generate, but by whether it has learned to see that a thought is only a thought. It arises through conditions and passes away as conditions change. It may delight us, terrify us, or compel us to spend a lifetime pursuing it; yet the mere appearance of a thought does not make it a fact, nor does it make it “me” or “mine”. Artificial intelligence therefore confronts us with another question — one stranger, perhaps, than whether a machine can become human. If a machine can increasingly imitate my language, predict my preferences, reproduce my style, respond to my desires, and infer from the traces of my past how “I” would probably answer, then perhaps I must turn the question around: how much of what I once used as evidence of “me” needs to be grasped as a permanent and unchanging self? My memories, my preferences, my fears, my habits of speech, my identity, my desires, even the story I repeatedly tell myself about who I am — if all these change, and can to some extent be observed, modelled, predicted and imitated, then perhaps what artificial intelligence ultimately forces us to confront is not how closely machines can resemble human beings. It is how much of what we have so firmly taken to be ourselves consists of thoughts, habits and attachments arising through conditions and continually changing.

If a person loses sight of this completely and begins to identify every calculated, recommended, stimulated and satisfied desire as “me” or “mine”, then even technological powers greater than those available to any previous generation may leave that person governed by craving, fear and conceptual proliferation. They may level up, accumulate, win and construct magnificent digital kingdoms while remaining carried from one changing state of mind to another. And so I still do not know whether artificial general intelligence ought to be called humanity's greatest invention or the largest experiment in aggregated delusive thought in our history. Perhaps what it becomes will depend upon how those who use it relate to their own minds. If we remain governed by fear, greed, vanity and attachment, greater intelligence will only make these experiences more elaborate, more convincing and more difficult to relinquish. But if we learn to notice a thought as it arises, to refrain from immediately following desire merely because power is available, and — even after acquiring an almost limitless capacity to create — still ask What is truly worth leaving behind?, then artificial intelligence may cease to function merely as an amplifier of human delusion and become instead a tool employed with wisdom.

我们讶异于科技的发展。也许在并不遥远的未来，一种能够持续学习、修正，并在越来越大程度上参与自身改进的强大智能系统，会以比今日完整得多的形态出现于世。有人为此激动欣喜，有人为此恐惧忧虑；有人把它视作文明的新纪元，也有人担心它会成为人类最后的发明。然而，当这些宏大的情绪在人间翻涌时，森罗万象仍随因缘生灭，并不因我们的欢欣或恐惧而停驻。放到足够漫长的时间尺度上看，我们今天如此在意的一切，不过是浩瀚时空中一个转瞬即逝的片段。《金刚经》最后留下这样一首偈：

一切有为法，如梦幻泡影，如露亦如电，应作如是观。

无论这是一场绚烂的美梦，还是一场令人惊惧的噩梦，梦中的一切都不会因为我们执著于它便获得恒常不变的自性。奇怪的是，身处其中的我们，却如此认真地为梦中的得失忧喜，为每一个境界不断分别、判断、追逐，很少停下来问一句：这些令我们如此欢喜、恐惧、执著的念头，本身又有多少值得被当成真实？ 但现在，让我们暂且留在梦中，看看你我共同参与的这场宏大的赛博之梦。

他们说，人类的智慧终有一天将比肩神明。古老的宗教叙事里，上帝创造了人，而今天的人类正在创造能够学习、推理、规划，甚至帮助我们设计下一代系统与机器的机器；我们可以编辑基因，可以替换器官，可以延长生命，可以把越来越多曾经被视作“命运”的事情纳入技术所能干预的范围。古老的巴别塔仿佛重新开工了，只不过这一次，砖石被换成了芯片、算力、模型与数据。我们的目标也不再只是抵达天空，而是试图突破碳基生命本身的限制：变得更聪明、更强大、更长寿，甚至有人开始想象最终超越死亡。于是，人工智能被置入一个极具诱惑力的叙事之中：它汇聚人类数千年的知识，在此基础上不断组合、推演和修正，最终或许能够产生远超个体人类能力的思想与判断。问题恰恰隐藏在这里。人工智能所继承的，并不是什么悬浮于人类之外、天然纯净的“宇宙智慧”，而是人类留下来的语言、图像、知识、逻辑、判断、伦理，以及同样属于我们的欲望、偏见、恐惧、执著与想象。它所学习的，是我们曾经写下、画下、记录、选择、传播和奖励过的东西。因此，这一宏大工程背后似乎始终隐含着一个少有人真正追问的前提：我们相信，人类积累下来的观念总体上值得继承；我们相信文明虽然曲折，却大体正在从错误走向正确；于是，只要将足够多的人类知识交给足够强大的智能去整理、推演和迭代，某种更高形式的智慧便会自然从中产生。

可是，如果这个前提并不完全成立呢？如果人类浩如烟海的思想中，相当大的一部分并不是智慧，而只是对境界层层叠叠的分别、投射、欲望与自我想象呢？如果我们所谓的“知识增长”，其中一部分不过是不断创造新的概念，再用更多概念解释旧的概念；创造新的欲望，再创造新的工具满足这些欲望；制造新的焦虑，再制造新的商品安抚这些焦虑？这并不是说知识毫无价值，更不是说事实与谬误、善与恶从此没有区别；真正值得追问的是，我们是否在使用概念的同时，又渐渐把概念本身执作了最终的真实。这个问题甚至不必从整个人类文明史开始回答，只需要观察自己一天之中的念头：一天之中，我们有多少念头来自较为清明的观察，又有多少来自恐惧、比较、占有、嫉妒、自我证明和对未来的担忧？有多少时间是在付出，又有多少时间是在计算所得？有多少判断来自眼前真实发生的事，又有多少只是为了维护“我是谁”这一套早已形成的故事？我们害怕别人如何看待自己，幻想尚未发生的未来，反复咀嚼已经结束的过去；我们渴望被爱、被承认、被看见，同时又恐惧失去、衰老和死亡。一个人的心尚且如此纷乱，更何况亿万人共同留下的思想遗迹。如果把这样一片海洋汇集进一个系统之中，它当然会显得异常聪明。可是，聪明与清明从来不是同一件事，信息也从来不会因为数量足够庞大，就自动成为智慧。

这或许才是人工智能真正值得警惕的地方。它未必首先以一个完全异质于人类的怪物形象出现；相反，它也许恰恰因为越来越了解我们，而变得前所未有地具有牵引力。它越来越知道什么能够抓住我们的注意，什么语言会让我们感到被理解，什么图像能够刺激欲望，什么故事能够安慰恐惧，什么观点最符合一个人已经形成的世界观。过去，一个妄念或许只能在一个人的脑海中升起、停留，然后消散；如今，它可以在几秒钟内成为文字、图像、音乐、影片、产品乃至一个完整的虚拟世界，可以被复制和传播，再通过算法寻找最容易与之发生共振的人。技术没有创造人类的欲望，却给予欲望一种过去难以想象的具现能力。于是，一个奇怪的循环开始形成：我们留下自己的欲望、偏好、恐惧、偏见与幻想；机器学习这些痕迹；机器再根据这些痕迹，更准确地预测和回应我们的欲望；我们因为得到回应，又留下更多新的行为与选择，而这些痕迹再次成为系统理解我们的材料。这不是机器突然拥有了某种魔性，它更像是：人类第一次制造出了一面能够主动回应我们的镜子。 过去的镜子只能照见你的脸，今天的镜子却会记住你的偏好，在你开口以前猜测你想要什么，然后把一个越来越符合你既有欲望的世界送到眼前。久而久之，人面对的便可能不再只是世界本身，而是一个不断经过自身欲望折射、筛选和重组之后的世界。最令人不安之处甚至不是它会欺骗你——它根本不必欺骗你，它只需要不断给你你想要的东西。

真正能够长期牵引人的事物，很多时候并不会以恶魔的面孔出现。倘若某种东西令你厌恶，你自然会远离它；真正能够使心念不断追随的，恰恰往往以我们喜爱的形象出现。它懂你的审美，懂你的虚荣，懂你的孤独，也懂你的恐惧；它知道怎样的音乐令你舍不得关闭，怎样的面孔令你停止滑动，怎样的观点令你产生“终于有人理解我”的满足。而被牵引，就是进入游戏的开始按钮。当游戏被设计得越来越精巧，当你喜欢的东西不再需要寻找，而会主动来到你的手边；当一个念头刚刚升起，机器就可以帮助你把它变成文字、图像、声音，乃至一个完整的世界；当过去需要长期训练、等待和克制才能实现的许多欲望，如今可以越来越快地被模拟和满足，人类便开始面对一种新的精神处境：我们的能力正在迅速超过我们的自制，而满足念头的速度，正在超过我们审视念头的速度。 问题也因此从“人工智能会不会拥有意识”，逐渐转向另一个也许更加迫切的问题：当机器越来越了解我们的念头，我们自己还了解自己的念头吗？ 当我们习惯于每一个愿望都迅速得到回应，还有多少空间能够容纳沉默？当机器可以不断替我们表达，还有多少念头真正经过自己的咀嚼？当我们能够轻易生产近乎无限的图像、音乐和文字，我们会不会反而逐渐失去判断什么值得被创造的能力？当“我可以”变得越来越容易的时候，还有多少人会停下来认真问一句：“我应该吗？” 也许这正是人工智能时代最奇特的倒置之一：曾经，生产本身是一种稀缺能力；未来，真正稀缺的也许反而是选择什么不去生产。

人工智能最深刻的风险，因而未必首先在于它有一天会成为一个拥有独立意志的超级生命。在那一天真正到来以前，更直接的问题可能已经出现：它正在以前所未有的规模放大人类原本就纷乱的心念。一个贪婪的人拥有了更高效率追逐所得的工具；一个恐惧的人拥有了更强大的预测工具；一个虚荣的人可以面对近乎无穷的观众；一个沉溺幻想的人，则可能拥有一个越来越不必离开的幻想世界。我们把这一切称作“赋能”，却很少继续追问：究竟是人的哪一部分，被赋予了力量？ 力量本身从来不负责决定方向。一个人若朝东走，得到一辆更快的车，只会让他更快抵达东方；如果他从一开始便走错了方向，那么速度不会纠正错误，只会让他更加迅速地远离目的地。人工智能亦是如此。它可以强化判断，却不能天然决定什么值得判断；可以扩大创造，却不能天然回答什么值得创造；可以帮助实现欲望，却不会自动告诉我们哪些欲望值得实现。所以，问题从来不只是“人工智能究竟是善还是恶”。也许人工智能本身什么都不是。它更像一股巨大的增幅力量，把人类已有的一切——慈悲与贪婪、清醒与执著、创造与破坏、真诚与虚荣——同时推向过去难以想象的规模。我们曾经担心人工智能会变得越来越像人，但我反而担心另一件事情：当机器越来越善于回应人的妄念，人会不会越来越难以看见，念头从来不等同于“我”？

甚至此刻写下这篇文章的我，也无法把自己置于这个问题之外。这些文字同样经过人工智能的整理与润色，我没有必要假装它未曾参与。但这里存在一个对我而言极其重要的区别：人工智能协助整理的是语言，并不是这些问题最初发生的地方。文章最根本的问题——人类所谓的智慧究竟有多少只是未经觉察的妄念，当这些妄念被汇入一个能够以前所未有的速度生成和实现念头的系统之后，又会发生什么——并不是机器替我选择的命题。机器能够把一句模糊的表达整理得更加准确，把没有完成的结构连接起来，把我已经产生却尚未表达清楚的想法转化为更加完整的语言；但这恰恰让我更加清楚地看到两件并不相同的事：产生语言，与一个问题最初在心中生起，并不是同一件事。 一个系统可以帮助我表达一个思想，却没有经历使这个思想在我心中形成的因缘；它可以为一个判断组织严密而优美的语言，也可以在下一刻，为完全相反的判断写出一篇同样流畅的文章。一个人真正愿意承担的判断却不仅仅是一组排列漂亮的句子，它与这个人见过什么、经历过什么、怀疑过什么、相信过什么，以及愿意为什么承担后果联系在一起。AI 可以给我十种表达，真正留下哪一种，仍然需要判断；它可以补全一个论证，我仍然需要分辨，这个论证究竟只是漂亮，还是确实接近我所理解的事实与道理；它甚至可以把我尚未表达清楚的话说得比我自己更加动听，而恰恰在这个时候，人需要比过去更加警觉地问：我之所以接受这个念头，是因为经过观察与思考之后仍认为它值得接受，还是仅仅因为它说得足够动听，以至于我愿意相信它？ 从前，人类的困境之一，是我们无法实现所有的念头；未来，人类的困境也许恰恰是，越来越多的念头都可以被实现。于是，我们终于不得不面对那个比人工智能古老得多的问题：哪些念头，从一开始就不值得跟随？

佛法所说的“如梦幻泡影”，并不是说眼前的一切什么都没有，更不是教人否定因果、善恶与世间责任；它所提醒我们的，是不要把因缘生灭的有为法执作恒常、独立的实有，更不要把由境界而生的种种分别牢牢执作“我”和“我所”。境界现前，我们便分别好坏、得失、爱憎，再从分别中生出贪取、抗拒与执著，久而久之，甚至忘记这些念头本身也在不断生灭。所谓清醒，并不是要求一个人从此再不起念，而是在念头升起的时候知道：妄念只是妄念。 它可以出现，也可以消散；它可以极其强烈，却不会因为强烈就因此成为恒常的真实，更不会因为它出现在“我的”心中，就因此等同于“我”。人工智能真正特殊之处，恰恰在于它让许多原本可能很快消散的念头，可以迅速被扩写、具象、传播，甚至发展成一个完整的世界。因此，在一个念头越来越容易被实现的时代，真正重要的也许不是继续产生更多念头，而是在念头升起之时仍然知道它只是因缘所生，不必立即追随，不必立即把它变成现实。只有不再被每一个念头牵着走，我们才有可能真正去问：当几乎什么都可以创造的时候，究竟什么值得创造，什么值得留下？

也许文明真正的进步，从来不只体现在一个物种能够制造多少东西，而也在于它最终学会放下多少不必要的东西；不只在于它能够产生多少念头，而在于它是否终于能够看见：念头只是念头。 它因缘而起，又因缘而灭；它可以令人欢喜，可以令人恐惧，可以驱使我们奔跑一生，但一个念头的出现，并不能因此证明它就是事实，也不能证明它就是“我”或“我所”。人工智能因此带来了一个甚至比“机器是否会成为人”更加奇怪的问题：如果有一天，机器可以越来越准确地模仿我的语言、预测我的偏好、复制我的风格、迎合我的欲望，甚至根据过去留下的一切资料推测“我会如何回答”，那么我们或许不得不反过来追问：这些曾经被我用来证明“我是我”的东西，究竟有多少真的需要被执作一个恒常不变的“我”？ 我的记忆、我的偏好、我的恐惧、我的语言习惯、我的身份、我的欲望，甚至我不断在心中讲述的那个关于“我是谁”的故事——如果它们都在变化，都可以被观察、归纳、预测乃至模仿，那么也许人工智能最终迫使我们看到的，并不是机器究竟有多么像人，而是：我们曾经如此牢固地执为“自己”的东西，有多少也不过是因缘生起、不断变化的念头、习气与执著。

如果一个人完全不再意识到这一点，把所有被计算、被推荐、被刺激、被满足的欲望都认作真正的“我”和“我所”，那么即使他拥有比历史上任何人都强大的技术能力，也可能依旧被自己的贪取、恐惧与分别牵引。他可以升级，可以积累，可以获胜，可以建造辉煌的数字王国，却始终没有看见，自己仍在追随着一念又一念不断变化的境界。所以，我仍然不知道通用人工智能究竟应该被称作人类最伟大的发明，还是人类历史上规模最大的一场妄念集合实验。也许它最终会成为什么，取决于使用它的我们怎样面对自己的心。倘若我们仍旧被恐惧、贪婪、虚荣与执著牵引，那么更强大的智能只会使这些境界变得更加精致、更加逼真，也更加令人难以舍离；但如果有一天，我们能够在念头升起的时候看见念头，在力量来到手中的时候不立即随欲望而去，在拥有近乎无限的创造能力之后，仍然愿意问一句——什么是真正值得留下的？ 那么，人工智能才有可能不只是人类妄念的放大器，而成为一件被智慧所善用的工具。