諾獎得主Wilczek：通向自我複製的機器之路_風聞

撰文 | Frank Wilczek

翻譯 | 胡風、梁丁當

中文版

或許很快，我們就能模仿生命體的功能，製造出能自我複製的機器。

縱觀歷史長河，富有創造力的人類工程師不斷從生物世界中汲取靈感。萊昂納多·達芬奇（Leonardo da Vinci）受到鳥類、魚類和烏龜的啓發，分別設計了飛行器、潛艇和坦克。如今，受動物神經系統啓發而研發的計算機構架——人工神經網絡，已成為機器學習的前沿技術。但這些應用都未觸及生物學的深層結構。而這，或將成為未來創造的燈塔。

諾 貝 爾 生 理 學 或 醫 學 獎 獲 得 者 保 羅 · 納 斯（Paul Nurse）在他的新書《生命是什麼？》（What is Life?）中指出，生命的深層結構指的是細胞或有機體這樣的基本單元，它們能夠自我繁殖，並允許微小的變異。繁殖與變異共同通過自然選擇推動物種的演化，從而形成多樣化的生物種羣。它們不僅能夠在變化的環境中存活，還能夠利用新的機會。而那些成功適應環境的單元就能繼續繁殖後代。

類似的機制在不同的尺度上都發揮着作用，構成了眾多關鍵生物過程的基礎。胚胎從單細胞發育為成熟有機體的過程中，會經歷好幾個生長階段（人類有幾十個），每個都與前一個略有不同。最終，受精卵繁衍出各種不同的細胞，包括心臟、肝臟和腦部的細胞。在一個微型的演化過程中，局部物理、化學環境中的信號會誘導“正確”的細胞形成。當幹細胞在應對損傷，或者是皮膚、腸道和血液細胞由於磨損而死亡時，這樣的微型演化機制也會被激活。

約翰·馮·諾依曼（John von Neumann）和達芬奇一樣，也是一位有遠見的工程師，只不過他表達的方式不是通過藝術，而是方程式與圖表。他創建了博弈論，以及以程序和隨機存儲器為基本特徵的“馮諾依曼體系”——這幾乎是所有現代計算機的基礎。他早期將量子力學與信息理論聯繫起來的一些觀點直到“第二次量子革命”時才被廣泛認可。

馮·諾依曼於1957年去世。去世前，他正在研究一個新項目，其未完成的手稿後來被收入了自我複製自動機理 論 》（Theory of Self-Reproducing Automata）一書。其中，他精確地設計了一個被稱為“通用複製器”的數學模型。它包括三個基本組成部分：機器A是一台可以根據指令整合資源並進行組裝的機器 ；程序B能夠指揮機器A；主程序C可以指揮A來製造A+B+C。

對於這種能在簡單化的現實世界中運行的複製系統，馮·諾依曼做了嚴格、詳細的設計。從技術上講，它是一台元胞自動機，可以從周圍隨機散落的碎片中獲取零件。原則上，根據他的設計，你可以用現代技術造出一個3D打印與計算機的混合系統，它能夠收集材料來製作你想要的東西或複製其自身。通過精心設計故意犯錯的程序或寬鬆的質量控制，我們也能解鎖生命的另一個秘密——變異。

用現成的3D打印機、計算機和原材料所構建的系統肯定是笨拙和低效的。如果有一天，科學家們能夠從生物中學到如何根據編碼在DNA上的指令來製造分子機器，那麼馮 · 諾依曼的願景將更接近實用。要記住，他早期的電腦模型是在真空電子管時代提出的，這同樣超越了當時的技術。

能自我複製的機器可以釋放指數增長的魔力，或許能讓一些大膽的工程成為現實。它們或許能讓將其他天體地球化的科幻夢想變得觸手可及。而最深刻的或許是，通過呈現生物學的深層結構，生命與非生命的界限終將變得模糊。

英文版

Advances in technology will soon allow us to build machines that replicate themselves and evolve like living beings.

Throughout history, creative human engineers have taken inspiration from artifacts of the biological world. Leonardo da Vinci designed flying machines, submarines and tanks with birds, fish and tortoises in mind. Today, artificial neural nets, a computer architecture directly inspired by animal nervous systems, are the cutting edge of machine learning. But none of those applications get to the deep structure of biology-likely a beacon of future creativity.

As the Nobel biologist Paul Nurse explains in his recent book “What is Life?,” the deep structure of life is the existence of physical units (cells or organisms) that can reproduce themselves, allowing small variations. Those ingredients-reproduction and variation-together drive evolution by natural selection. They generate a diverse population that can survive changes and exploit new opportunities. Those that succeed will be those that breed.

Remarkably similar tricks, working on different scales, underlie many other key biological processes. Embryos develop from single cells into mature organisms after several stages of growth (in humans, a few dozen), where each stage differs a little from the previous. Thus, the fertilized egg’s diverse progeny eventually includes heart, liver and brain cells. The “right” kind of cell emerges in response to signals in its local physical and chemical environment, in a kind of guided miniature evolution. Less specialized stem cells can re-ignite this mini-evolution in response to injury or, in the case of skin, gut and blood cells, death by wear-and-tear.

Though he worked in equations and diagrams rather than artistic renderings, John von Neumann was a visionary modern engineer on the level of da Vinci. He gave us game theory and the so-called “von Neumann architecture,” featuring stored programs and random-access memory, that is the foundation for almost all present-day computers. Some of his early ideas connecting quantum mechanics with information theory are only now becoming widely appreciated, in the “second quantum revolution.”

At the time of his death in 1957, at the age of 54, von Neumann was well into a major new project. His unfinished manuscript, edited by Arthur Burns into the book “Theory of Self-Reproducing Automata,” is monumentally impressive. In it, he gives precise designs for mathematical models of objects he called “universal replicators.” They consist of three basic parts: a machine A that can gather resources and assemble things following a program, a program B that instructs A how to make desired products, and a master program C that instructs A how to make A + B + C.

Von Neumann provided a rigorous, detailed design for a system of this kind operating within a recognizable simplification of the real world. Technically, it would be a cellular automaton within a bath of randomly scattered pieces that it can scavenge for parts. Exploiting modern technologies, you could in principle elaborate on his designs to make a hybrid 3D printer/computer system that collects material to build something you want plus a copy of itself. It wouldn’t be hard to incorporate life’s other secret ingredient-that is, variation—either by deliberate programming or by loose quality control.

A system built with off-the-shelf 3D printers, computers, and the materials they require would be unwieldy and inefficient, to be sure. But as scientists master the art of making molecular machines according to plans encoded in DNA, von Neumann’s vision will get closer to practicality. It’s worth remembering that his early computer designs, which date from the era of vacuum tubes, likewise outstripped available technology.

Self-reproducing machines could unleash the power of exponential growth, thus enabling audacious engineering projects. They might bring the science-fiction dream of terraforming astronomical bodies within reach. Most profoundly, by embodying biology’s deep structure, they would blur the distinction between life and non-life.

Frank Wilczek

弗蘭克·維爾切克是麻省理工學院物理學教授、量子色動力學的奠基人之一。因發現了量子色動力學的漸近自由現象，他在2004年獲得了諾貝爾物理學獎。

本文經授權轉載自微信公眾號“蔻享學術”。本文紙質版發表在《環球科學》2021年10月刊，編輯：黃琦。