Introduction: The Anatomical Compiler

• Levin’s goal: Total rational control of biological growth and form (morphogenesis). This would solve many medical problems (birth defects, injury, cancer, aging) and enable new technologies (synthetic morphology, non-neuromorphic AI).
• The “Anatomical Compiler” is a future system: You draw the desired organism (anatomy, not molecular biology), and the system generates stimuli to guide cells to build it. *Not* 3D printing or genomic editing, but a *communication* device to translate our goals to cells.
• Current limitations: We lack this compiler; can only control morphology in very limited cases. Genetics/molecular biology alone aren’t enough, as a Genome encodes protein *hardware*, not large-scale anatomical *instructions*.

The Morphogenetic Code: Beyond Genetics

• Example: “Frog-lottle” (Axolotl/Frog hybrid). Genomes of both are known, but we can’t predict if it will have legs, illustrating the gap between genotype and large-scale phenotype.
• Forward vs. Reverse Problem: Going from simple rules to complex outcomes (like fractals) is easy. Going backward (regenerative medicine: “fix this asymmetry”) is incredibly hard (“intractable inverse problem”).
• Molecular manipulation is not whole picture. Where biology was like needing to physically rewire hardware to acheive goals, a la old Computer Science of 40s and 50s, we need higher levels of understaning.
• Multiscale Competency Architecture: Biological systems have problem-solving ability at *every* level (molecules, cells, tissues, organs, organism). Each level navigates its own “space” (gene expression, physiology, anatomy, behavior). We can *communicate* with these levels, not just rewire them.
• Examples of biological problem-solving (intelligence): Embryonic development (twins from split embryos), regeneration (salamanders, axolotls – limbs, organs), deer antlers (rapid bone growth), human liver/fingertip regeneration.

Bioelectricity: The “Software” Layer

• Inspiration from the Nervous System: The brain guides the body through behavioral space using electrical signals (ion channels, gap junctions).
• The Same Applies to Anatomy: *All* cells have ion channels and gap junctions (not just neurons). Evolution discovered electrical networks for information processing long before brains.
• Electrical networks are important because every cell has an ion channel, cells connected to cells and communicate, so that it should be treated like a kind of hardware which, with a certain amount of bioelectrical output/inputs/pattern-over-time, it should, conceptually, become a new form of electrical computer and this principle existed in the times of bacterial films and cells did not forget this just because it joined and coorperated with other cells to make larger, new forms of ‘computers’.
• “Electric Face” of Frog Embryo: Voltage patterns *predict* future anatomy (eyes, mouth) *before* relevant genes are expressed. This is a bioelectric *memory* guiding development.
• Pathological Patterns: Cancer cells show altered electrical states (decoupling from neighbors) *before* becoming tumors.
• Rewriting Patterns: Tools (optogenetics, drugs targeting ion channels) allow us to *control* bioelectric states and thus *manipulate* development (induce extra organs, limbs, change body plan).
• Permanent Changes: Two-headed planaria (flatworms) demonstrate that altered bioelectric patterns can be *stable* and *heritable* (without changing the genome). We rewrite the *memory* of the “correct” body plan.
• Organ-Level Induction: A “subroutine call” (“make an eye here”) can trigger complex organogenesis. Cells can *recruit* neighbors, demonstrating collective intelligence.

Xenobots: Synthetic Biology and Emergent Behavior

• Collaboration with Josh Bongard (UVM): Creating “Xenobots” from frog skin cells. When isolated, these cells *self-assemble* into motile structures. No brain and new neurons are needed.
• Emergent behaviours can spontaneously arise: these include such behaviours: move in circles, patrol back/forth, interact collectively.
• AI-Guided Design: Evolutionary algorithms can *predict* and *design* xenobot behavior (e.g., “Pac-Man” shape for particle collection). Xenobots can even build copies of themselves (“kinematic self-replication”).
• Implications: Skin cells have a “hidden” behavioral repertoire, revealed by removing constraints. AI can help us control and understand this “native” intelligence.

Future Directions and Implications

• Applications: Regenerative medicine (controlling wound healing, limb regeneration in mammals), cancer therapeutics (restoring electrical communication), biorobotics.
• Long-Term Vision: Understanding and controlling “collective intelligence” of cells for various applications. Creating new forms of artificial intelligence inspired by biology.
• Biology’s future is expected to involve lots of evolved/designed material/software at different scales. Thus ethical concerns become very necessary for navigating. Darwin’s “endless forms most beautiful” is just a tiny part.
• Future systems will likely rely on using AI to make connections between software top-down processes which affect execution machinery down the line.
• Broader Implications: Need for new ethics to deal with “hybrid” organisms (not fitting traditional categories). Expanding our concept of “intelligence” to include diverse biological systems.

“

简介：解剖编译器

• Levin 的目标：全面理性地控制生物的生长与形态（形态发生）。这将解决许多医学问题（先天缺陷、损伤、癌症、衰老），并带来新技术（合成形态学、非神经形 AI）。
• “解剖编译器”是一种未来的系统：你可以绘制所需的生物体（解剖层面，而非分子生物学），然后系统会生成刺激信号，引导细胞构建它。这并不是 3D 打印或基因组编辑，而是一种将我们的目标传达给细胞的“交流”装置。
• 当前的局限：我们还没有这种编译器；只能在极少数情况下控制形态。仅依靠基因学或分子生物学并不够，因为基因组只编码蛋白质“硬件”，而非大规模的解剖“指令”。

形态发生密码：超越遗传学

• 示例：“青蛙蝾螈”（墨西哥钝口螈/青蛙的杂交体）。尽管我们已知它们的基因组，却仍无法预测它是否会长出腿，这说明基因型与大规模表型之间存在巨大差距。
• 正向问题 vs. 逆向问题：从简单规则生成复杂结果（如分形）相对容易。但反向推导（再生医学中的“修复这种不对称”）极其困难，被称为“难以求解的逆问题”。
• 分子层面的操作并不代表全部。过去的生物学如同需要对硬件进行物理布线才能实现目标（类似 20 世纪四五十年代的计算机科学），我们需要更高层次的理解。
• 多尺度能力结构：生物系统在每个层级（分子、细胞、组织、器官、有机体）都具备解决问题的能力。每个层级都在各自的“空间”（基因表达、生理学、解剖学、行为）中运作。我们可以与这些层级“对话”，而不仅仅是对它们重新布线。
• 生物系统解决问题（智能）的示例：胚胎发育（通过分裂胚胎形成双胞胎）、再生（蝾螈、墨西哥钝口螈的肢体和器官再生）、鹿角（快速骨骼生长）、人类肝脏与指尖再生等。

生物电：如同“软件”层

• 来自神经系统的启示：大脑利用电信号（离子通道、间隙连接）在行为层面引导身体。
• 解剖层面同样适用：所有细胞都含有离子通道和间隙连接（并非只有神经元）。在大脑出现很久以前，进化就已发现用电网络进行信息处理。
• 电网络至关重要，因为每个细胞都有离子通道，相互连接并进行交流。从概念上看，它们如同硬件，通过一定的生物电输出/输入及随时间变化的模式，可变成新的“电计算机”。这一原理早在细菌膜时期就已存在；细胞并未因彼此协作构建更大、更先进的“计算机”而忘记这一点。
• 青蛙胚胎的“电面孔”：在相关基因表达之前，电压模式就能“预示”未来的解剖结构（如眼睛、嘴），这是一种引导发育的生物电“记忆”。
• 病理模式：癌细胞在形成肿瘤之前便会出现异常的电状态（与邻近细胞脱耦）。
• 重写模式：通过光遗传学或作用于离子通道的药物等工具，我们能够控制生物电状态，从而操纵发育（诱导额外器官或肢体，改变身体结构）。
• 永久性改变：双头涡虫（扁形虫）显示出更改生物电模式可以是稳定且可遗传的（无需改变基因组）。我们由此改写了“正确”身体结构的“记忆”。
• 器官层面的诱导：一次“子过程调用”（“在此生成一个眼睛”）就能触发复杂器官的发生。细胞还可“招募”邻近细胞，展现了集体智能。

Xenobots：合成生物学与自发行为

• 与 Josh Bongard（佛蒙特大学）合作：使用青蛙皮肤细胞制造 “Xenobots”。当这些细胞被分离后，可自我组装成可运动结构。它们无需大脑或新的神经元。
• 可能自发出现的行为：绕圈移动、往返巡逻、集体交互等。
• AI 引导设计：进化算法能预测并设计 Xenobot 的行为（例如，用“吃豆人”形状收集颗粒）。Xenobot 甚至可构建自身副本（“运动自我复制”）。
• 意义：皮肤细胞有“潜在”的行为方式，只是在被约束时未能展现。AI 可以帮助我们控制并理解这种“原生”智能。

未来方向与意义

• 应用：再生医学（控制伤口愈合、哺乳动物肢体再生）、癌症治疗（恢复细胞间的电信号交流）、生物机器人等。
• 长期愿景：理解并控制细胞的“集体智能”以应用于各领域，并由生物学启发创建新的人工智能形式。
• 未来的生物学将涉及在不同尺度上演化/设计的材料与软件，因此伦理考量尤为重要。达尔文所说的“无尽的美丽形态”仅是冰山一角。
• 未来系统很可能依赖 AI，在自上而下的软件流程与底层执行机制之间建立联系。
• 更广泛的影响：我们需要新的伦理来处理那些并非传统分类的“混合”生物体，并拓展“智能”的概念以涵盖多样的生物系统。