Introduction and Core Concepts

• Levin argues that all living cells, not just neurons, possess a basic form of cognition. This means they have goals, make decisions, and solve problems within their specific domains (e.g., anatomical space, chemical space).
• The type of thinking performed by a cell depends on its environment. The type of thinking of the cells, collective is focused on creating/maintaing anatomical structures.
• Morphogenesis: The process by which organisms develop their shape. Levin emphasizes that this is a problem-solving process driven by cells collectively navigating “morphospace” (the space of possible anatomical forms).
• Bioelectricity: A crucial layer of communication and computation beyond biochemistry. All cells maintain a voltage difference across their membranes, and this voltage state is used for signaling. Ion Channels are protein “gates” in cell membranes that control the flow of ions (charged particles), creating voltage differences. They are like biological transistors. Gap Junctions are direct connections between cells that allow ions (and thus electrical signals) to flow freely. They erase ownership information on signals, promoting collective decision-making (“mind melt”). This makes is the start of a collection. Electrical Networks, collections of cells connected by gap junctions form networks that can store and process information, similar to (but slower than) neural networks.
• Memory in Biological Systems. Memory is not stored only in DNA; there exist multiple ways: Chemical networks (gene regulatory networks, dynamical attractors), Cytoskeletal structures (physical arrangements) and Bioelectrical states (like flip-flops, volatile RAM) where The configuration persists, acting like memory, even without changes to physical hardware.
• Teleophobia is being worried that the assignment of goals/agency is being misattributed, when really, any useful model which makes helpful decisions could and in his opinion, should be referred to as agentic and described with cognitive descriptors.

Planaria: A Model System

• Planaria do not age. They continuously regenerate, demonstrating that aging isn’t an inevitable thermodynamic process.
• Planaria can regrow any body part. Each fragment “knows” what’s missing and how to rebuild.
• A bioelectric network stores the “target morphology” (the ideal body plan). This network can be reprogrammed (e.g., to create two-headed worms) without changing the DNA. The altered body plan is heritable through fission (splitting), demonstrating non-genetic inheritance.
• Planaria accumulate mutations but maintain perfect anatomical control. This challenges the idea that DNA fully determines body plan, highlighting the role of bioelectric “software.”

Xenobots: Synthetic Organisms and “Engineered by Subtraction”

• Xenobots are self-assembling, bio-robotic organisms created from frog skin cells (Xenopus laevis).
• When isolated from the rest of the embryo, skin cells spontaneously form xenobots with novel behaviors. These include: Movement, navigation, collective behavior, and Kinematic self-replication, in which They build copies of themselves from loose cells, a behavior not found in frogs.
• Removing constraints (other cells) reveals the inherent plasticity and problem-solving capacity of cells. The default behavior of the frog cells is to be a xenobot, not skin.
• Collaboration with AI (Josh Bongard): Evolutionary algorithms are used to predict and design xenobot behaviors by manipulating cell interactions, not by changing DNA.

Multi-Scale Competency Architecture

• Biological systems have goals at multiple levels (molecules, cells, tissues, organs, organism). Each level has some degree of autonomy and problem-solving ability.
• Higher levels influence lower levels by altering the “landscape” of possibilities. Lower levels simply follow local gradients, contributing to the higher-level goal without needing to “know” the big picture. Like guiding water down a hill.
• This architecture allows for robust development and regeneration even in the face of noise, mutations, or environmental changes.
• The goals of an organism (e.g. a human climbing) can and will easily differ, and often run in direct conflict with, the lower-level organizational structures it comprises (e.g. skin cells).

Implications for Regenerative Medicine and Beyond

• Anatomical Compiler (Long-Term Goal): A system that translates a desired anatomical form into a set of stimuli that will guide cells to build it. This would revolutionize medicine by enabling the regeneration of limbs, organs, and potentially reversing aging.
• Somatic Psychiatry: Treating diseases by targeting the goal-directed behavior of cell collectives, rather than micromanaging at the molecular level.
• Understanding and Controlling Collective Intelligence: Developing a science to predict and manipulate the goals of complex systems (cells, swarms, AI). This is crucial for both biology and artificial systems.
• Ethical Considerations: Challenging binary distinctions (natural vs. artificial, living vs. non-living, human vs. non-human). Expanding our understanding of cognition to include diverse forms of intelligence.

Evolution and the Nature of Intelligence

• General phenomenon: evolution is probably quite ubiquitous because it stems from: heredity, heredity-error, competition.
• Evolution doesn’t create solutions to specific problems; it creates machines that can solve problems in various spaces (anatomical, chemical, behavioral).
• Even simple organisms may have a basic sense of agency, driven by the need to model themselves and their environment under energy constraints. The belief in free will may be a consequence of self-constructing systems.
• Unconventional Cognition: Recognizing and studying intelligence in systems that don’t fit traditional categories (plants, slime molds, synthetic organisms).

Key Metaphors and Analogies

• Dogs vs. Legos: Building with “agential materials” (dogs) is different from building with passive materials (Legos). Agential materials have their own goals and require training/persuasion, but they are also more resilient.
• The collective behavior of cells is like an orchestra, where the “music” (the emergent behavior) is the “dictator,” not any individual instrument (cell).
• Higher levels in the competency architecture “bend” the option space for lower levels, guiding their behavior without direct control. Analogy to relativity.
• Gap junctions create a shared cognitive space, blurring the boundaries between individual cells.

Concise Definitions (Some from Levin, some inferred)

• Agential Material: A material with its own goals, preferences, and some level of autonomy (e.g., cells).
• Target Morphology: The “ideal” body plan that a regenerating system strives to achieve.
• Cognitive Light Cone: The boundary of the largest goal a system can work towards, in space and time.
• Anatomical Compiler: A future system to design and build organisms by specifying their desired form.
• Ioniceutical: and intervention or agent which interacts directly with the bioelectrical state, perhaps through an ION channel, so that the anatomy may be guided in this manner.
• Software 2.0: A programming paradigm where, instead of writing explicit code, you train a system (like a neural network) to achieve a desired outcome. Analogous to training cells.
• Teleophobia: being wary of falsely attributing traits such as intelligence and intention onto something.

导言与核心概念

• 莱文认为，所有活细胞，不仅仅是神经元，都具有基本形式的认知能力。这意味着它们有目标，做决策，并在其特定领域（如解剖空间、化学空间）内解决问题。
• 细胞执行的思维类型取决于其环境。细胞群体的思维类型，集中于创建/维护解剖结构。
• 形态发生：生物体形成其形状的过程。莱文强调，这是一个由细胞集体在“形态空间”（可能的解剖形式空间）中导航的解决问题的过程。
• 生物电：超越生物化学的一个关键的沟通和计算层。所有细胞都在其细胞膜上保持电压差，并且该电压状态用于信号传导。离子通道是细胞膜中的蛋白质“门”，控制离子（带电粒子）的流动，产生电压差。它们就像生物晶体管。间隙连接是细胞之间的直接连接，允许离子（以及电信号）自由流动。它们消除了信号的所有权信息，促进了集体决策（“思维融合”）。这使其成为一个集合的开始。电网络，由间隙连接连接的细胞集合形成网络，可以存储和处理信息，类似于（但比）神经网络慢。
• 生物系统中的记忆：记忆不仅仅存储在DNA中。它以多种方式存在：化学网络（基因调控网络、动态吸引子）、细胞骨架结构（物理排列）和生物电状态（如触发器、易失性RAM），其中配置持续存在，充当记忆，即使没有物理硬件的改变。
• 目的恐惧症：担心目标/自主性的分配被错误归因，而实际上，任何做出有用决策的有用模型都可以，并且在他看来，应该被称为自主的，并用认知描述符来描述。

涡虫：一个模型系统

• 涡虫不会衰老。它们不断再生，表明衰老不是一个不可避免的热力学过程。
• 涡虫可以再生任何身体部位。每个片段都“知道”缺少什么以及如何重建。
• 生物电网络存储“目标形态”（理想的身体计划）。该网络可以在不改变DNA的情况下进行重新编程（例如，创建双头蠕虫）。改变后的身体计划通过分裂（分裂）是可遗传的，证明了非遗传继承。
• 涡虫积累突变，但保持完美的解剖控制。这挑战了DNA完全决定身体计划的想法，突出了生物电“软件”的作用。

异种机器人：合成生物体和“通过减法进行工程设计”

• 异种机器人是由青蛙皮肤细胞（非洲爪蟾）制成的自组装、生物机器人生物体。
• 当与胚胎的其余部分隔离时，皮肤细胞自发地形成具有新行为的异种机器人。这些包括：运动、导航、集体行为和运动学自我复制，其中它们从松散的细胞中构建自己的副本，这是一种在青蛙中找不到的行为。
• 消除约束（其他细胞）揭示了细胞固有的可塑性和解决问题的能力。青蛙细胞的默认行为是成为异种机器人，而不是皮肤。
• 与人工智能合作（Josh Bongard）：进化算法用于通过操纵细胞相互作用来预测和设计异种机器人行为，而不是通过改变DNA。

多尺度能力架构

• 生物系统在多个层次（分子、细胞、组织、器官、生物体）上都有目标。每个级别都具有一定程度的自主性和解决问题的能力。
• 较高层次通过改变可能性的“景观”来影响较低层次。较低层次只是遵循局部梯度，为较高层次的目标做出贡献，而不需要“知道”大局。就像引导水下山一样。
• 即使在噪声、突变或环境变化的情况下，这种架构也能实现强大的开发和再生。
• 生物体（例如，攀爬的人）的目标可以并且很容易不同，并且经常与构成它的较低级别的组织结构（例如，皮肤细胞）的目标发生直接冲突。

对再生医学及其他领域的影响

• 解剖编译器（长期目标）：一种将所需的解剖形式转换为一组刺激的系统，这些刺激将指导细胞构建它。这将通过实现肢体、器官的再生，并可能逆转衰老来彻底改变医学。
• 躯体精神病学：通过靶向细胞群体的目标导向行为来治疗疾病，而不是在分子水平上进行微观管理。
• 理解和控制集体智能：发展一门科学来预测和操纵复杂系统（细胞、群体、人工智能）的目标。这对生物学和人工系统都至关重要。
• 伦理考虑：挑战二元区别（自然与人工、生物与非生物、人类与非人类）。将我们对认知的理解扩展到包括各种形式的智能。

进化与智能的本质

• 普遍现象：进化可能是相当普遍的，因为它源于：遗传、遗传误差、竞争。
• 进化不会为特定问题创造解决方案；它创造了可以在各种空间（解剖、化学、行为）中解决问题的机器。
• 即使是简单的生物体也可能具有基本的自主性意识，这是由在能量约束下对自身和环境进行建模的需要所驱动的。对自由意志的信念可能是自我构建系统的结果。
• 非常规认知：识别和研究不符合传统类别的系统（植物、黏菌、合成生物体）中的智能。

关键隐喻和类比

• 狗与乐高：使用“自主材料”（狗）进行构建与使用被动材料（乐高）进行构建不同。自主材料有自己的目标，需要训练/说服，但它们也更有弹性。
• 细胞的集体行为就像一个管弦乐队，其中“音乐”（涌现行为）是“独裁者”，而不是任何单个乐器（细胞）。
• 能力架构中的较高层次“弯曲”较低层次的选项空间，在不直接控制的情况下指导它们的行为。类比相对论。
• 间隙连接创建了一个共享的认知空间，模糊了单个细胞之间的界限。

简洁定义（一些来自莱文，一些推断）

• 自主材料：具有自身目标、偏好和一定程度自主性的材料（例如，细胞）。
• 目标形态：再生系统努力实现的“理想”身体计划。
• 认知光锥：系统可以在空间和时间上努力实现的最大目标的边界。
• 解剖编译器：一种未来的系统，通过指定其所需的形式来设计和构建生物体。
• 离子药物：一种干预或药剂，可能通过离子通道直接与生物电状态相互作用，以便可以通过这种方式引导解剖结构。
• 软件 2.0：一种编程范式，其中，您不是编写显式代码，而是训练系统（如神经网络）以实现所需的结果。类似于训练细胞。
• 目的恐惧症：警惕错误地将智力和意图等特征归因于某物。