陶哲轩最近在讨论 AI 与数学的 misalignment 时，提出了一个我很喜欢的问题：当 AI 可以越来越容易地给出答案，我们究竟为什么还要做数学？在他的 Mathematics in the Age of AI 里，他列出的数学目标远不只有解决未解问题，还包括发展新的理论与结构、理解世界、建立共同的数学知识，以及创造具有持久价值的作品。 我从中读到的一个意思是：答案本身不是数学，答案是对于真正理解了这片景观的人的奖赏。
我想进一步说，科学本就是对于理解神谕景观的人的奖赏。
如果把实验者比喻为登山者，那么实验就是对于山林景观的一次探索。一个冒险家先望向远处，提出一个目的地：那里是不是有一座山？山后面是什么？沿着这条河往上走，会不会抵达一个从来没有人去过的地方？然后他背起行囊，真正进入森林，沿途留下坐标，检查自己的路线，在一次次走错、返回与重新寻找之后归来，把他所见过的景观分享给我们。
科学也是如此。一个科学家先提出问题，再用严谨、可检验、可证伪的方法走进去。实验、证明、统计与重复验证，是他留下的坐标。它们重要，不是因为坐标本身就是景观，而是因为只有依靠这些坐标，我们才有理由相信：这个人真的去过那里，他带回来的不是幻觉，不是一张想象中的地图。科学的重要，因而既在于科学家选择了什么目的地，也在于他用怎样严谨的手段，把一片过去不可见的景观带到了我们面前。
这也是为什么，当我们聊 AI 会怎样改变科学时，我总觉得应该先回头看看艺术。艺术其实已经经历过一次又一次这样的技术冲击。十九世纪摄影出现以后，绘画原本承担的一部分功能——准确地记录人的面孔、城市、战争和现实——第一次被机器大规模接管。此后电影、摄影以及机械复制继续扩大图像生产的边界。如果绘画的价值只是“把眼前的世界准确画下来”，那么照相机当然是一种近乎残酷的替代。
但后来留下来的，并不是一场关于“谁画得更像”的竞赛。印象派开始追逐光线、空气、瞬间的颜色与观看本身：莫奈可以面对同一座鲁昂大教堂，在不同时间反复画它，因为重要的已经不再只是“教堂长什么样”，而是一个人在某一个下午究竟看见了什么。再往后，我们留下梵高，留下塞尚，留下无数并没有因为摄影能够更准确复制世界就停止工作的艺术家。我们记住的不只是他们最后生产出来的图像，也记住他们怎样看世界，怎样选择，怎样挣扎，以及他们用一生走过的那条路。
所以技术并没有简单地“消灭艺术”。它消灭的，往往只是某一种过去被误认为艺术本身的稀缺能力。当准确复制现实不再稀缺，我们才更加清楚地发现：艺术从来不只是复制现实。类似的事情，也可能正在科学里发生。当证明、代码、论文甚至一个问题的答案开始变得廉价，我们才被迫重新问：过去我们真正珍视的，究竟是这些产物，还是产生这些产物背后的理解、判断、问题与探索？
贡布里希在《艺术的故事》开头写过一句后来非常著名的话：“There really is no such thing as Art. There are only artists.” 他后来解释，这句话本来就是要把那个抽象的、大写的“Art”重新拉回具体的人：艺术不是一个脱离人的神秘实体，真正存在的是一个个面对具体问题、掌握具体技艺、做出具体选择的艺术家。
我越来越觉得，科学或许也是如此。
我们今天习惯把科学想象成一个巨大的知识机器：论文、定理、数据、benchmark、实验结果不断被加入进去，于是 Science 自己仿佛在向前运动。但真正提出一个值得追问的问题的人，真正决定往哪座山走的人，真正第一次看见一片新景观的人，从来都是具体的人。论文只是他归来以后留下的地图，答案只是抵达某个山顶之后获得的奖赏。
科学不存在，只有科学家存在。
Tao recently raised a question I very much like while discussing the misalignment between AI and mathematics: when AI can produce answers with ever greater ease, why do we still do mathematics? In his Mathematics in the Age of AI, the goals he sets out for mathematics go far beyond solving open problems. They include developing new theories and structures, understanding the world, building a shared body of mathematical knowledge, and creating works of lasting value. One idea I take from this is that the answer itself is not mathematics; the answer is the reward for someone who has truly come to understand the landscape.
I want to go one step further: science itself is the reward for those who come to understand the oracular landscape.
If we imagine the experimenter as a mountaineer, then an experiment is an exploration of a mountain landscape. An adventurer first looks into the distance and chooses a destination: is there a mountain over there? What lies beyond it? If I follow this river upstream, will it lead to a place no one has ever reached before? Then he shoulders his pack and actually enters the forest, leaving coordinates along the way, checking his route, getting lost, turning back, and searching again, before finally returning to share with us the landscape he has seen.
Science is much the same. A scientist first asks a question, then enters it by rigorous, testable, and falsifiable means. Experiments, proofs, statistics, and replication are the coordinates he leaves behind. They matter not because the coordinates are themselves the landscape, but because only through them do we have reason to believe that this person truly went there - that what he brought back was not a hallucination, not an imagined map. The importance of science therefore lies both in which destination the scientist chose and in the rigor of the means by which he brought a previously invisible landscape before us.
This is also why, when we talk about how AI may change science, I often feel that we should first look back at art. Art has already lived through technological shocks of this kind again and again. After photography emerged in the nineteenth century, one of the functions painting had long performed - accurately recording faces, cities, wars, and reality - was, for the first time, taken over by machines on a large scale. Film, photography, and mechanical reproduction then continued to expand the boundaries of image production. If the value of painting were simply to "draw the world before our eyes accurately," the camera would of course be an almost brutal substitute.
Yet what endured was not a competition over "who could paint more accurately." The Impressionists began to pursue light, air, the colors of a fleeting moment, and the act of seeing itself. Monet could paint the same Rouen Cathedral again and again at different times, because what mattered was no longer merely "what does the cathedral look like?" but what, exactly, a person saw on a particular afternoon. Later came Van Gogh, Cezanne, and countless other artists who did not stop working simply because photography could reproduce the world more accurately. We remember not only the images they ultimately produced, but also how they saw the world, how they chose, how they struggled, and the path they spent a lifetime walking.
Technology, then, did not simply "destroy art." What it often destroyed was only a scarce ability that had previously been mistaken for art itself. Once accurate reproduction of reality ceased to be scarce, we could see more clearly that art had never been merely the reproduction of reality. Something similar may now be happening in science. When proofs, code, papers, and even the answer to a question begin to become cheap, we are forced to ask again: what did we truly value all along - these outputs, or the understanding, judgment, questions, and exploration that gave rise to them?
At the beginning of The Story of Art, Gombrich wrote what later became a famous sentence: "There really is no such thing as Art. There are only artists." He later explained that the point was to bring the abstract, capital-A "Art" back down to concrete human beings: art is not some mysterious entity detached from people. What truly exists are individual artists facing specific problems, mastering specific skills, and making specific choices.
I increasingly feel that science may be the same.
Today we are accustomed to imagining science as a vast machine of knowledge: papers, theorems, data, benchmarks, and experimental results are continuously added to it, and Science itself seems to move forward. But the person who asks a question truly worth pursuing, the person who decides which mountain to walk toward, the person who sees a new landscape for the first time, has always been a particular human being. A paper is only the map he leaves behind after returning; an answer is only the reward received upon reaching a summit.
There is no such thing as science. There are only scientists.