Eyes evolved independently dozens of times across animal lineages. Each invention compressed a vast amount of environmental information into neural patterns — and each, in turn, enlarged the nervous systems behind them. This site traces that history from the first light-sensitive patch on a single-celled organism to the visual cortex, the camera eye, and the silicon sensors that now read the world for us. 在动物谱系中,眼睛独立演化过数十次。每一次发明都把环境中的大量信息压缩成神经模式——又反过来扩展了支撑它的神经系统。本站追溯这段历史:从单细胞生物上最初的感光斑,到视觉皮层、相机型眼睛,再到如今替我们阅读世界的硅基传感器。
Eyes did not appear in one leap. They accreted: a light-sensitive pigment, then a clustered patch, then a cup, then a pinhole, then a lens-bearing camera. Each stage works on its own; each is found in living organisms today; each conferred a survival advantage at its scale. Below: the canonical sequence first sketched by Lamarck and Darwin, sharpened by Nilsson & Pelger's 1994 simulation, and grounded in modern molecular biology. 眼睛并非一蹴而就。它是逐层堆积的:先是感光色素、再到色素斑、凹陷、针孔,最后是带透镜的相机型眼。每个阶段单独工作有效;每个阶段在今天的活体生物中都能找到;每一阶段在其尺度上都赋予了生存优势。下文为达尔文与拉马克最初勾勒、Nilsson & Pelger 1994 年的模拟所锐化、并由现代分子生物学所支撑的标准序列。
Light is electromagnetic radiation. Vision evolved at the narrow band of the spectrum where the Sun emits most strongly and water transmits cleanly — roughly 380–750 nanometers. This is not coincidence; it is selection acting where the photons actually were. Below: the spectrum, the visible band, and the four physical operations every eye must perform. 光是电磁辐射。视觉演化于太阳辐射最强、水体也最透明的窄段——约 380–750 纳米。这并非巧合,而是在光子实际所在之处发生的选择。下方为电磁谱系、可见光带,以及每只眼睛都必须完成的四项物理操作。
A retina is the only piece of central nervous tissue you can see in a mirror. It performs the first stages of visual processing — contrast, motion, edges — before the signal even reaches the brain. The visual cortex then layers on object recognition, depth, attention, and prediction. About a third of the human cortex is devoted to vision; in some primates, more than half. 视网膜是镜子里唯一能看到的中枢神经组织片段。它在信号尚未抵达大脑之前,即完成视觉处理的初级阶段——对比度、运动、边缘检测。视觉皮层之后再叠加对物体的识别、深度、注意与预测。人类皮层约有三分之一专司视觉;在某些灵长类中超过一半。
The Cambrian explosion (~540 Mya) coincides closely with the appearance of complex eyes. Once predators could see, prey had to hide, run, or signal. Once prey could see, predators had to ambush, mimic, or out-pattern. The simulator below scores six structural outcomes for a population given predator/prey eye sophistication, environmental clarity, and selection pressure. 寒武纪大爆发(约 5.4 亿年前)与复杂眼睛的出现在时间上密切重合。一旦捕食者能看,猎物便须隐藏、奔跑或发信号;一旦猎物能看,捕食者便须埋伏、拟态或压过对方的图案识别。下方模拟器在给定捕食/猎物的眼睛复杂度、环境清晰度与选择压力时,对一个种群的六项结构性输出打分。
Evolution rarely arrives at the same answer twice. Below: six lineages with completely different optical architectures, each optimal for a different ecological niche. None is "more advanced" than the others — that framing is one of the most stubborn errors in popular biology. 演化很少两次得出相同答案。下文为六个具有完全不同光学架构的谱系,各自在不同生态位上最优。没有一个"比另一个更高级"——这一框架是大众生物学中最顽固的错误之一。
Color is what a nervous system computes from the relative responses of its photoreceptors. A bee sees ultraviolet patterns invisible to humans; a snake images thermal infrared on its facial pits; a mantis shrimp has twelve to sixteen photoreceptor types but appears to use them not for fine discrimination but for fast classification. The phenomenon called color is real; the reds, greens, and blues are constructions. 颜色是神经系统从其感光细胞的相对响应中算出之物。蜜蜂看到人类看不见的紫外图案;蛇用面部颊窝成像热红外;螳螂虾有 12 至 16 类感光细胞,但似乎并非用于精细辨别,而是用于快速分类。"颜色"现象是真实的,但红、绿、蓝皆为构造。
| Species物种 | Photoreceptor types感光细胞类型数 | What it sees you cannot它能看到而你不能 |
|---|
Sighted animals run an inner model of the visual world that is updated continuously and compared with new evidence. Mistakes in that model are interesting: they reveal where the brain is generating versus measuring. Optical illusions, dreams, hallucinations, and blindsight are not bugs — they are windows into how vision is computed. 具备视觉的动物运行一份对视觉世界的内部模型,并不断更新、与新证据比对。该模型的错误是有趣的:它们揭示了大脑何处在生成、何处在测量。视错觉、梦境、幻觉、盲视并非缺陷——而是窥探视觉计算的窗口。
From Chauvet's 30,000-year-old cave paintings, to the writing systems of Sumer and Shang, to perspective drawing in the Renaissance, to the microscope, telescope, photograph, cinema, screen, and now generative-AI imagery — human civilization has steadily handed more of its memory, instruction, and judgement to the visual channel. Below: a sequence of inflection points and what each one made newly possible. 从三万年前的肖维洞穴壁画,到苏美尔与商代的书写系统,到文艺复兴的透视画法,再到显微镜、望远镜、摄影、电影、屏幕,如今至生成式 AI 图像——人类文明持续把记忆、指令与判断更多地交给视觉通道。下文为一系列拐点,每一拐点令何种事物首次成为可能。
Modern computer vision did not solve sight by simulating biology. It solved sight by training very large parameter spaces on very large image collections. The biological and the artificial converge structurally in places (hierarchical features, motion processing, attention) and diverge sharply in others (energy use, sample efficiency, robustness to distribution shift). Below: where the two systems meet and where they part ways. 现代计算机视觉并非通过模拟生物来解决"看"的问题——而是通过在巨量图像上训练巨大的参数空间。生物视觉与人工视觉在某些地方结构性收敛(层级特征、运动处理、注意),在其他地方明显分岔(能耗、样本效率、对分布漂移的鲁棒性)。下文为两者的交汇与分岔。
Retinal implants partially restore sight to the blind by stimulating the optic nerve electrically. Optogenetic therapies can confer light-sensitivity to surviving retinal cells. AR overlays add information to what eyes already see. Brain-machine interfaces can — in principle — bypass the eye entirely. None of these collapse the biological story; they extend it. The honest framing is neither utopian nor dystopian. 视网膜植入物通过电刺激视神经,部分恢复盲人的视觉;光遗传学疗法可使残存的视网膜细胞获得感光能力;增强现实在眼睛已见之物上叠加信息;脑机接口原则上可完全绕过眼睛。这些都不否定生物学故事,而是延长它。诚实的态度,既非乌托邦,亦非反乌托邦。
Five canned responses on vision evolution. Free-text input falls back to a heuristic synthesizer over the same knowledge base. I will not produce design-language teleology, pseudoscience about "higher" species, or simple-stories that flatten the actual complexity. 五个关于视觉演化的预设回答。自由文本输入将回退到同一知识库上的启发式合成器。我不会输出"设计论"目的论、关于"高等"物种的伪科学,亦不会以简化故事抹平实际复杂度。
Biologist · ready生物学家 · 待命
Pick a question above, or type your own. I reason from the evolutionary record, the molecular biology, and the neuroscience — not from teleological language about "purpose" in evolution. 从上方选择问题,或自行输入。我以演化记录、分子生物学与神经科学为依据——不以"目的"等目的论语言推理。