Introduction: Turing, Intelligence, and Morphogenesis

• Alan Turing was interested in both AI and morphogenesis, seeing a deep parallel. Levin believes these are fundamentally the same question: problem-solving in different spaces.
• It’s important to view intelligence not by scale, as everything (even human brains) comprises a larger collection (colony/collective intelligence) of smaller ones.
• There exist Smooth Cartesian transitions, starting from physics & chemistry which result in cognitions like: awareness & metacognitions.

Core Concepts: Multi-Scale Competency and Navigation

• Biology uses a multi-scale competency architecture: nested problem-solvers at different levels (cells, tissues, organs, organisms) each with autonomy and goals.
• Navigation, particularly of “spaces,” is central. These spaces aren’t just 3D physical space, but also physiological space (chemical parameters), transcriptional space (gene expression), and morphospace (anatomical configurations).
• Agents pursue *goals* relevant to them, despite not having the biggest scope (i.e., skin cell regeneration, even if that conficts with the higher level).
• Goal directed-ness should allow relations to *any* system.
• “Cognitive boundary” describes goal-scale.
• Goal-directedness is key for understanding and interacting with unconventional agents.

Bioelectricity and Morphogenesis: A Detailed Example

• Biological pattern formation is the behavior of a collective intelligence of cells navigating morphospace.
• Bioelectrical networks (precursors to brains) are the medium for this proto-cognitive activity. Cells communicate electrically via ion channels and gap junctions. This isn’t just philosophy, it has practical implications for biomedicine.
• Examples, cells handle local tasks: metabolic, morphogenesis, and behavior tasks.
• The morphogensis system does: vibrations and vibrations and soner sensing, creates a “map,” make decisions with respect to its environment.

Beyond Natural Kinds: Plasticity of Agents

• Agents are not fixed entities. Examples:
  • Caterpillar to butterfly: Radical body and brain reorganization, yet memory persists.
  • Planaria regeneration: Regrow any body part, including the brain, with memory retention even after head amputation. Information transfer between tissues.
  • Tadpole eye plasticity: Eyes grafted to the tail can still provide vision, even with novel neural connections. The brain adapts to new sensory input locations.
• Biological systems are nested structurally (cells, tissues, etc.) and functionally: each level has its own competency and solves problems in its space.

Beyond Traditional Spaces, Expanding Our Notion of Spaces

• Intelligence isnt only physical (3D). We can generalize intelligence by observing actions in non-3D, example, sensing and reacting to physiology states of the liver.
• Planaria can navigate a Barium solution despite the extreme dangers, and it takes a handful of the ~20k possible genes to allow Planaria to live through this process, showing non-random processes are taking place.

Intelligence as Problem-Solving, and Creating Systems

• Intelligence: the ability to creatively use new/existing informatoin creatively in new senarios, not just using pre-exisiting answers.
• Developing new cogntivie systems will require finding all the answers as nature can evolve solutions.
• TAME system (technological approach) needs a wide array: human to animal, but even new ones created in labs or by systems found outerspace, allowing us to recongize and compare them.

Navigating Morphospace: Goals and Homeostasis

• Cells *know* how and when to make the correct structure; regenerative system is similar but is about a signal *for* building.
• How cells work is a large scale question and *cant* be directly coded, example: you cannot tell Frogolotts grow leg by using genes; an unknown is whether Frogolott leg contains either axolotl or Frogolott.
• A long term is the Anatomical Compiler to make limbs/structures/shapes. It could revolutionize medicicine by being able to repair cancer, tramua, age-related illnesses, etc., etc.
• Morphospace is the space of all possible configurations for a structure. Embryogenesis is remarkably reliable at navigating morphospace, but it’s not hardwired. It’s homeostatic.
• Examples of morphogenetic homeostasis:
  • Monozygotic twins: Splitting an embryo results in two normal organisms, not two halves.
  • Axolotl limb regeneration: Regrows exactly what’s missing and then stops, demonstrating anatomical homeostasis.
  • Human liver regeneration, deer antler regeneration, child fingertip regeneration.
  • Newt Kidney Tubule: Cells adjust in size and number (even molecular mechanism) to form the correct lumen.
  • Frog-Leg Regeneration. Normal Frog limb Morphogensis will occur.
  • Picasso frogs: Messed up facial features still migrate to form a relatively normal frog face, showing error minimization, not hardwired movements.

Bioelectricity as the “Software” of Morphogenesis

• The standard developmental biology model needs feedback loops. This creates a homeostatic system. The set point is a complex data structure: a large-scale geometry (anatomical descriptor). This introduces goal-directedness, which is often avoided in biology.
• Brains are a good example of navigating space to a goal. Bioelectricity offers a similar mechanism:
  • Cells communicate electrically in networks (like neurons).
  • Ion channels set cell voltage.
  • Gap junctions allow electrical communication between cells.
• The commitment of neuroscience is decoding: electricial patterns representing the content: goals, memories.
• Bioelectrical signaling *before* genes for development.
• Neural electricity is an evolutionary pivot from morphospace to 3D space.

Tools to Read and Write Bioelectrical Information

• Voltage-sensitive fluorescent dyes reveal electrical conversations between cells.
• Computational modeling predicts voltage patterns from ion flows.
• The “electric face” in frog embryos: a bioelectrical pre-pattern that precedes gene expression and determines facial structure. Disrupting this pattern alters development.
• There exist Pathological electrical patterns.
• Tools to write: Modifying ion channels and gap junctions (like neuroscientists) to alter electrical states. No external fields/waves used, only endogenous mechanisms.
• Practical use of Bioelectricity, example: induce and suppress malformations/illnesses such as cancer, eye generation.

Examples of Bioelectric Control

• Inducing ectopic eye formation: A specific voltage pattern triggers eye development, even in inappropriate locations. This instruction is modular (doesn’t specify *how* to build an eye).
  • Cells recruit their neighbors.
  • Induction of ectopic organs: otocysts, hearts, forebrain, limbs, even fins (which tadpoles don’t normally have).
  • We have drug coctail through ion chanels to produce leg.
• Regenerating frog legs (which frogs don’t normally do) using a drug cocktail that targets ion channels.
• Altering planarian head number: An electrical circuit stores the “memory” of how many heads to regenerate. This can be reprogrammed, creating two-headed worms that continue to regenerate as two-headed even without further intervention. Non-genetic inheritance.

Exploring Morphospace and Attractors

• Tweaking the electrical circuit can lead to head shapes of other planarian species, or even novel shapes not found in nature. All with the same wild-type genome. This reveals different attractors in morphospace.
• “Full stack” models: Integrating molecular information (channels), physiology (voltage states), organ identity, and algorithms to understand the system and intervene. Computational platforms simulate tissue-level dynamics.
• Repairing birth defects: Using bioelectrical models to identify interventions (drugs targeting ion channels) to restore correct brain development in tadpoles with teratogen-induced or genetic defects. “Software” fixes for hardware problems.
  • Electroceuticals can help use bioelectrical *interface.*

Scaling of Cognition and Goals

• Borders change, between, and there is plasticity: individual organisms and the colonies.
• The boundary between self and world is flexible. Goal pursuit unifies diverse intelligences across scales.
  • Single cells have small goals (microns).
  • Cell collectives have larger goals (building a limb).
• Scaling of self is present in Cancer cells, whose electrical communications separate with a colony that exists within a bigger system.
• Cancer can sometimes be reversed by reconnecting cells electrically, emphasizing the importance of collective state.
• Any cognitive agent has some ability to care, where the abilities increase linearly but are cut off due to our capacity.

A Cognitive Light Cone Model

• A way to compare diverse intelligences: the size of their “cognitive light cone” – how big (in space and time) are the goals they can pursue.
  • Tick: Small goals (local butyrate concentration).
  • Dog: Larger goals (memory, some anticipation).
  • Human: Goals potentially larger than lifespan.
• Multi-scaling includes influence higher > lower organizational levels, e.g. genes expression is contorlled to “steer” systems.
• Multi-scale system: Higher levels “bend” the option space for subunits, guiding them to the overall goal.
• There is an example of this cycle for PH where each cell can adjust and do cycle processes; this continues up in higher structures and system, with even *evolution* pivoting it through different states/structures: behavior spaces to other, undiscovered areas.

Xenobots: An Extreme Example of Novelty

• Xenobot can arise spontaneously, no additional interference necessary: e.g. no cirucit/chemicals/instructions necessary, subtracting constraints *does* matter.
• Xenobots are made from frog skin cells, allowed to “reboot their multicellularity” without the normal embryonic context. They demonstrate:
  • Spontaneous self-organization.
  • Novel behaviors: Movement, navigation, calcium signaling (brain-like activity, without neurons).
  • Zenobot will regenreate, example: split it down half, see hinge clamp together, see how Zenobot retains this original shape.
  • Kinematic self-replication: They build copies of themselves from loose cells – a behavior not found in frogs.
  • Collect together into piles
  • Engineered to shape environment, which then the Zenobots would use it for another function.
• One Xenobot genone (frog’s Xenopus lavis) is also two other stages/behaviors
• These behaviors are not pre-programmed or selected for; they arise from the cells’ inherent plasticity. Skin cells *want* to be xenobots. Evolution is behavior shaping: finding signals to coax agential materials.

Ethical Implications and a Future Taxonomy of Minds

• Viable Agents can comprise any materials, evolution/design: organic and inorganic materials/hybirds etc,
• Biology is incredibly interoperable. Any combination of evolved, designed, and software components can be a viable creature. We’re entering a world of unfamiliar agents.
• Ethics should be based on an agent/ability/functionality, no origin-story.
• Traditional ethical frameworks (based on origin and appearance) will become useless. We need new frameworks based on the capacity for sentience and goal-directedness.
• Bioelectricity (though the examples) shows great areas to explore with systems, and there could be many others we don’t yet.

Closing: Goal Directedness and Future Research

• Can there exist framework without *cultural* ideas like evolutionary natural history to *universially* apply like a mathematical axiom? Active-inference model from physics is close (cognitive functions emerge from minimize “surprises”/expectations).
• Intelligence could possibly organize through space by its scale, or other properties as well as their wavelengths.
• Relationships between systems (animals, plants, systems) might have hidden geometrical information.
• Exploring relationships with other non-animals. e.g. plants.
• High order-connectivy. Networks don’t show the whole picture (but this is hard to definitively define in this).
• A goal to see more novel solutions beyond evolution or other limited ideas, especially within *anotomical compliers.*
• There may be no such thing as *regenerations.*
  • For cells: new cell can be *envrionments* for one another.
  • Mind may arise: flat (cognitions rise at once).
• A general framework: test-compair-act and you get emergent *affordances,* where example, transitors do many amazing tasks/configurations like logic gates without this originally programmed in them, or the mathematical operations that will add all the triangle’s degrees = 180.
• Taxonomy of Mind could be: light cones: ability to see the highest end, biggest goalls.
• The work focuses on a continuum of agency, scaling of goals, bioelectricity as a cognitive medium, and implications for evolution, biomedicine, and understanding/creating diverse intelligences.
• All of these things we use today might have to be *increased* to adjust/account/plan. Example, maybe someone will literally care/understand all organisms instead of beign “linear,” there are other shapes too.

导言：图灵、智能与形态发生

• 艾伦·图灵对人工智能和形态发生都感兴趣，认为两者之间存在深刻的相似之处。莱文相信它们本质上是同一个问题：在不同空间中解决问题。
• 重要的是，观察智能不应以规模为标准，因为一切事物（甚至人类大脑）都包含着由更小个体组成的更大的集合（群落/集体智能）。
• 存在平滑的笛卡尔转换，从物理和化学开始，最终产生认知，如：意识和元认知。

核心概念：多尺度能力与导航

• 生物学使用多尺度能力架构：不同层次（细胞、组织、器官、生物体）嵌套的问题解决者，每个层次都有自主性和目标。
• 导航，特别是“空间”的导航，是核心。这些空间不仅仅是三维物理空间，还包括生理空间（化学参数）、转录空间（基因表达）和形态空间（解剖结构）。
• 行动者追求与它们相关的*目标*，尽管它们没有最大的范围（即，皮肤细胞再生，即使这与更高层次的目标冲突）。
• 目标导向性应该允许与*任何*系统建立关系。
• “认知边界”描述目标尺度。
• 目标导向性是理解和与非常规行动者互动的关键。

生物电与形态发生：一个详细的例子

• 生物模式形成是细胞集体智能在形态空间中导航的行为。
• 生物电网络（大脑的前身）是这种原始认知活动的媒介。细胞通过离子通道和间隙连接进行电交流。这不仅仅是哲学，它对生物医学具有实际意义。
• 例如，细胞处理局部任务：代谢、形态发生和行为任务。
• 形态发生系统执行：振动和声波感应，创建“地图”，根据其环境做出决策。

超越自然类别：行动者的可塑性

• 行动者不是固定的实体。例如：
  • 毛毛虫到蝴蝶：彻底的身体和大脑重组，但记忆仍然存在。
  • 涡虫再生：再生任何身体部位，包括大脑，即使在头部切除后也能保留记忆。组织之间的信息传递。
  • 蝌蚪眼睛可塑性：移植到尾巴的眼睛仍然可以提供视觉，即使有新的神经连接。大脑适应新的感官输入位置。
• 生物系统在结构上（细胞、组织等）和功能上都是嵌套的：每个层次都有自己的能力并在其空间中解决问题。

超越传统空间，扩展我们的空间概念

• 智能不仅仅是物理的（3D）。我们可以通过观察非3D中的行动来概括智能，例如，感知和响应肝脏的生理状态。
• 涡虫可以在极度危险的钡溶液中导航，并且只需要大约2万个可能基因中的少数几个就可以让涡虫在这个过程中存活下来，这表明正在发生非随机过程。

智能作为问题解决，以及创建系统

• 智能：创造性地在新场景中使用新的/现有的信息的能力，而不仅仅是使用预先存在的答案。
• 开发新的认知系统将需要找到所有答案，因为自然可以进化出解决方案。
• TAME系统（技术方法）需要广泛的范围：从人类到动物，甚至是在实验室中创造的或在外太空发现的系统，使我们能够识别和比较它们。

导航形态空间：目标和稳态

• 细胞*知道*如何以及何时构建正确的结构；再生系统类似，但它是关于构建的信号。
• 细胞如何工作是一个大规模的问题，不能直接编码，例如：你不能通过使用基因告诉蝾螈长出腿；一个未知数是蝾螈的腿包含蝾螈还是青蛙。
• 一个长期目标是解剖编译器，用于制造肢体/结构/形状。它可以通过修复癌症、创伤、与年龄相关的疾病等，彻底改变医学。
• 形态空间是结构所有可能配置的空间。胚胎发生在形态空间中的导航非常可靠，但它不是硬连接的。它是稳态的。
• 形态发生稳态的例子：
  • 同卵双胞胎：分裂一个胚胎会产生两个正常的生物体，而不是两半。
  • 蝾螈肢体再生：精确再生缺失的部分，然后停止，证明了解剖稳态。
  • 人类肝脏再生、鹿角再生、儿童指尖再生。
  • 蝾螈肾小管：细胞调整大小和数量（甚至分子机制）以形成正确的管腔。
  • 青蛙腿再生。将发生正常的青蛙肢体形态发生。
  • 毕加索青蛙：混乱的面部特征仍然会迁移，形成一个相对正常的青蛙脸，显示出误差最小化，而不是硬连接的运动。

生物电作为形态发生的“软件”

• 标准的发育生物学模型需要反馈回路。这创建了一个稳态系统。设定点是一个复杂的数据结构：一个大规模的几何结构（解剖描述符）。这引入了目标导向性，这在生物学中经常被避免。
• 大脑是导航空间到目标的一个很好的例子。生物电提供了一种类似的机制：
  • 细胞在网络中进行电交流（像神经元一样）。
  • 离子通道设定细胞电压。
  • 间隙连接允许细胞之间的电通信。
• 神经科学的承诺是解码：代表内容的电模式：目标、记忆。
• 生物电信号先于基因发育。
• 神经电是形态空间到三维空间的进化枢纽。

读取和写入生物电信息的工具

• 电压敏感荧光染料揭示了细胞之间的电对话。
• 计算模型从离子流预测电压模式。
• 青蛙胚胎中的“电面”：一种先于基因表达并决定面部结构的生物电预模式。破坏这种模式会改变发育。
• 存在病理性的电模式。
• 写入工具：修改离子通道和间隙连接（像神经科学家一样）以改变电状态。不使用外部场/波，仅使用内源性机制。
• 生物电的实际应用，例如：诱导和抑制畸形/疾病，如癌症、眼睛生成。

生物电控制的例子

• 诱导异位眼睛形成：特定的电压模式触发眼睛发育，即使在不适当的位置。这个指令是模块化的（不指定*如何*构建眼睛）。
  • 细胞招募它们的邻居。
  • 诱导异位器官：耳囊、心脏、前脑、四肢，甚至鳍（蝌蚪通常没有）。
  • 我们有通过离子通道的药物混合物来产生腿。
• 使用靶向离子通道的药物混合物再生青蛙腿（青蛙通常不会这样做）。
• 改变涡虫头部数量：电路存储要再生的头部数量的“记忆”。这可以重新编程，创建双头蠕虫，即使没有进一步的干预，它们也会继续再生为双头。非遗传继承。

探索形态空间和吸引子

• 调整电路可以导致其他涡虫物种的头部形状，甚至在自然界中找不到的新形状。所有这些都具有相同的野生型基因组。这揭示了形态空间中的不同吸引子。
• “全栈”模型：整合分子信息（通道）、生理学（电压状态）、器官标识和算法，以了解系统并进行干预。计算平台模拟组织级动力学。
• 修复出生缺陷：使用生物电模型识别干预措施（靶向离子通道的药物），以恢复具有致畸剂诱导或遗传缺陷的蝌蚪的正确大脑发育。“软件”修复硬件问题。
  • 电药可以通过生物电*界面*来提供帮助。

认知的规模和目标

• 边界改变，之间存在可塑性：个体生物体和群体。
• 自我和世界之间的界限是灵活的。目标追求将不同尺度的各种智能统一起来。
  • 单个细胞有小的目标（微米）。
  • 细胞集体有更大的目标（建造肢体）。
• 自我的缩放存在于癌细胞中，其电通信与存在于更大系统中的群体分离。
• 有时可以通过电连接细胞来逆转癌症，强调集体状态的重要性。
• 任何认知主体都有一定的关怀能力，其能力线性增加，但由于我们的能力而受到限制。

认知光锥模型

• 一种比较不同智能的方法：它们的“认知光锥”的大小 – 它们可以追求的目标有多大（在空间和时间上）。
  • 蜱虫：小目标（局部丁酸浓度）。
  • 狗：更大的目标（记忆，一些预期）。
  • 人类：目标可能比寿命更长。
• 多尺度包括影响较高 > 较低的组织层次，例如，基因表达被控制以“引导”系统。
• 多尺度系统：较高层次“弯曲”子单元的选项空间，引导它们实现总体目标。
• 有一个 PH 周期的例子，每个细胞都可以调整和执行循环过程；这在更高的结构和系统中继续存在，甚至*进化*也通过不同的状态/结构（行为空间到其他未发现的领域）使其旋转。

异种机器人：新颖性的一个极端例子

• 异种机器人可以自发产生，不需要额外的干预：例如，不需要电路/化学物质/指令，减去约束*确实*很重要。
• 异种机器人由青蛙皮肤细胞制成，允许在没有正常胚胎环境的情况下“重新启动其多细胞性”。它们证明：
  • 自发自组织。
  • 新行为：运动、导航、钙信号传导（类似大脑的活动，没有神经元）。
  • 异种机器人会再生，例如：将其分成两半，看到铰链夹在一起，看看异种机器人如何保留其原始形状。
  • 运动学自复制：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。
  • 聚集在一起成堆。
  • 被设计用来塑造环境，然后异种机器人会将其用于另一个功能。
• 一个异种机器人基因组（青蛙的非洲爪蟾）也是另外两个阶段/行为。
• 这些行为不是预先编程或选择的；它们来自细胞固有的可塑性。皮肤细胞*想要*成为异种机器人。进化是行为塑造：寻找信号来哄骗自主材料。

伦理影响和未来心智分类法

• 可行的行动者可以包含任何材料，进化/设计：有机和无机材料/混合物等。
• 生物学具有令人难以置信的互操作性。进化、设计和软件组件的任何组合都可以成为可行的生物。我们正在进入一个不熟悉的行动者的世界。
• 伦理应该基于行动者/能力/功能，而不是起源故事。
• 传统的伦理框架（基于起源和外观）将变得无用。我们需要基于感知能力和目标导向性的新框架。
• 生物电（通过示例）显示了探索系统的巨大领域，并且可能还有许多我们尚未探索的领域。

结束：目标导向性和未来研究

• 是否存在没有*文化*思想（如进化自然历史）的框架，可以像数学公理一样*普遍*应用？来自物理学的主动推理模型很接近（认知功能从最小化“惊喜”/期望中产生）。
• 智能可能会通过其规模或其他属性以及它们的波长来组织空间。
• 系统（动物、植物、系统）之间的关系可能隐藏着几何信息。
• 探索与其他非动物的关系。例如植物。
• 高阶连接。网络没有显示出完整的图景（但这很难在其中明确定义）。
• 一个目标是看到超越进化或其他有限想法的更多新颖解决方案，特别是在*解剖编译器*中。
• 可能没有*再生*这样的东西。
  • 对于细胞：新细胞可以彼此成为*环境*。
  • 心灵可能会出现：扁平（认知同时上升）。
• 一个通用框架：测试-比较-行动，你会得到涌现的*可供性*，例如，晶体管执行许多惊人的任务/配置，如逻辑门，而这最初并没有在其中编程，或者数学运算将所有三角形的度数相加 = 180。
• 心智分类法可能是：光锥：看到最高端、最大目标的能力。
• 这项工作侧重于自主性的连续性、目标的规模、生物电作为认知媒介，以及对进化、生物医学和理解/创造多样化智能的影响。
• 我们今天使用的所有这些东西可能都必须*增加*以进行调整/考虑/计划。例如，也许有人会真正关心/理解所有生物，而不是“线性”，还有其他形状。