Introduction and Background

• Michael Levin’s early interest in how things are built (sparked by a TV) led him to explore both engineering/computer science and biology. He was interested in the way that both physical machines could create images on a screen and insects could perform behaviors.
• He became fascinated by how minds emerge from physics and chemistry, driving his focus on developmental biology.
• Levin’s perspective: standard developmental biology courses may offer a different, potentially more gene-centric view than his.

Genes, Hardware, and Software

• Genomes primarily describe the *hardware* of cells (protein sequences), not the overall organismal form (symmetry, number of limbs, etc.).
• Biological systems have both a hardware layer (genetics) and a crucial *software* layer (developmental physiology and decision-making processes). The software is dynamic and critical for navigating complex development.
• The mapping from genotype (genetic information) to phenotype (observable traits) is complex, not a simple one-to-one relationship, except in cases such as specific enzyme production.
• “Froggle” example: Mixing frog and axolotl cells creates a creature whose leg development can’t be predicted solely from the genomes, demonstrating the importance of cellular decision-making (software).
• Evolution selects for phenotypes (final outcomes, like anatomy and behavior) yet it’s genome (raw information) that get’s past on between generations, illustrating how interconnected and codependent each system.

Explaining Emergent Properties

• Explaining biological features requires considering both genetic factors and “free lunch” properties arising from physics, geometry, and computation.
• Galton board example: The bell curve shape arises from the system’s setup, not inherent in the wood, nails, or marbles individually. Emergent property from simple components and organization.
• Transistor example: Connecting transistors creates logic gates with truth tables that are inherent properties of the configuration, not something separately evolved.
• Biological systems can harness “generic laws” (as discussed by Stuart Kauffman and Stuart Newman) that are not explicitly encoded in the genes.

Cellular Decision-Making and Bioelectricity

• Cells and cell groups have agency (preferences, behaviors) and make decisions. Evolution shapes these behaviors through signals between cells.
• Individual cells have small-scale goals (pH, metabolic state). Multicellular collectives pursue larger goals (limb formation).
• Electrical networks, formed by connecting cells via gap junctions, are crucial for scaling up goals and collective intelligence.
• Gap Junctions allow a collective’s identity to fuzz out as molecules such as calcium merge together making multiple cells share a connected physiological signal. From a cells point of view it is a ‘false’ memory, from the collectie, it is real, causing a Mind-Meld, where memory is not longer ‘owned’.
• Homeostasis comparison: Single cells measure, remember, and act on their *local* environment. Connected cells in electrical networks measure, remember, and act on a *larger, non-local* scale, facilitating collective goals and response.
• Scaling of stress: Cells communicate stress (deviation from desired state) through shared signals, promoting plasticity and coordinated action to achieve collective goals. Individual stress becomes a shared problem.

Cancer as a Breakdown of Bioelectric Communication

• Bioelectrical signals create a larger sense of “self” across cell groups.
• Cancer can arise from cells becoming electrically disconnected from the larger network. They revert to their ancient, unicellular goal: to divide and go where resources are good (metastasis).
• Cells becoming disconnected (becoming cancerous) isn’t from increasing ‘selfishness’, rather it is a consequence of it shrinking, going back to their more selfish individual states, acting individually.
• Oncogenes often shut down gap junctions, the very first step towards bioelectrical disconnection, isolating cancer cells from the larger collective control.
• The study of bacterial biofilms shows that brain-like behaviors (using ion channels) evolved long before nervous systems, indicating the ancient roots of bioelectric communication.
• Morphespace: where good regions/bad regions, like barriers/obstacles are navigated similar to normal 3 dimensional space except now we are in a world of configuration.

Memory and Bioelectric Circuits

• Bioelectrical networks literally store a kind of memory, representing the “set point” or target morphology for regeneration.
• Planarian example: Cutting a flatworm into pieces results in each piece regenerating a complete, proportional worm.
• This implies that is must be a form of homeostatis going on; a complex non-neural form of a collective intelligence to keep such anatomical patterns consistent.
• Bioelectric pattern can be visualized (with voltage-sensitive dyes) and *rewritten* (using ion channel drugs) to change the body plan (e.g., creating two-headed worms). No genetic changes are required.
• That two-headed worms consistently produce two-headed worms is evidence of a true memory.
• The altered body plan is heritable through *fission* (splitting), demonstrating non-genetic inheritance. The bioelectric circuit acts as an additional hereditary medium. Not all inherited traits are DNA-based.
• Bioelectric circuit in planaria example; if the system’s physiology first ‘boots’ the bioelectric default circuit of number of head. This number-of-heads can also be edited non-genetically by altering the signals for short-term using inputs (tapping buttons on a calculator), creating a two-headed work without ‘rewriting’ the program.

The Eye Experiments

• Early frog embryos show a bioelectric pre-pattern that predicts the location of facial organs, including the eyes.
• Injecting ion channel RNA into other areas of the embryo (e.g., tail, gut) induces the formation of ectopic (out-of-place) eyes, demonstrating that bioelectricity is *instructive* for organ formation, and it is not merely for house-keeping.
• This revealed the modularity of development: The researchers didn’t need to specify *how* to build an eye, only *where*. The cells organized the complex process themselves.
• Cells recruited neighboring cells (even those not directly affected by the injected RNA) to participate in eye formation, showcasing multiple levels of instruction.
• Cells outside the traditionally defined “competent” regions (anterior neurectoderm) can, in fact, form eyes, highlighting the limitations of gene-centric views. It is not the top of the hierarchy, voltage is!
• Pac6: normally makes eyes, is found at anterior norectoderm, which will define competency for creating an eye, yet, other parts can.
• Ectopic eyes connected to the spinal cord (not the brain) could still mediate vision, demonstrating remarkable plasticity of the nervous system and its ability to interpret novel inputs.
• These examples were made early in Levin’s carreer, predating later research.
• Neutral Mutations; deleterious mutatuions, once lethal or severe can instead turn to be less sever or become neutural, broadening the landscape where evolution can progress towards.

Multiple Levels of Control and Goal-Directedness

• Biological systems exhibit multiple levels of emergence and control.
• Choosing the *right* level for intervention (e.g., bioelectric pattern vs. gene expression) is crucial for effective manipulation. Bioelectrical manipulation is often more effective and efficient than trying to micromanage genes.
• Instead of dealing with 10s of 1000s of individual parts/mechanisms, one could tap in further up the decision chain and deal with an intelligent system that navigates the complex decisions, taking away the stress of needing to handle the details.
• Biologists often exhibit “teleophobia” (fear of attributing goals or agency to biological systems), but cybernetics provides a framework for understanding machines with goals as a *continuum*, not a binary (dumb vs. smart). This is no longer considered “magical.”
• Telephobia came from needing to study all other entities as clocks since early days did not know how to interpret a human’s inner thoughts.
• Agency claims are *engineering* claims, testable by experiment (e.g., identifying, reading, and rewriting set points in a homeostatic system).
• Thermostat Example: how to test? look at what its level: is, is a setpoint?, 2 can we read/decode setpoint, rewrite? to rewrite is a new rewiring needed (like mechanical clock).
• Thermostat continued: after test; if thermostat work as expect, a trust/enginner dependency. No micromanaging is needed, temperature managed; good!.
• Genetic pathways can be *trained* (similar to neural networks), exhibiting various types of learning (including Pavlovian conditioning). This challenges the idea of purely deterministic gene regulatory networks.
• Implications for associative-learning, in a petri-dish there exists too powerful of a drug yet we cannot apply to humans, give both drugs, then give only nuetural one later; may or may not work.
• The molecular-placebo will activate only once paired enough times (Pavlov-style), where if the pairing stops, the original reaction (dog will drool from the bell ringing sound).

Placebo, Intention, and the Mind-Body Connection

• Intention can influence the body at multiple levels, including the bioelectric state of cells (e.g., deciding to stand up changes muscle cell voltage). The mind-body connection is strong and demonstrable.
• This suggests using levels of the hierarchy (chain of command) and speaking at the relevant part of the system to get tasks done.
• There is evidence of non-verbal “selves” within our bodies and we might be parts of larger selves.
• Learning is change withing one agent (human/machine learning, yet training (neural networks) implies multiple parties involved, at least one to create pressure (or lack-of pressure) and on receiver to change. The distinction could be helpful for exploring further discoveries.
• Minds emerge gradually during development. There’s no sharp dividing line between “just chemistry” and “having a mind.” This implies that minds exist in various forms across different scales of biological organization.
• We may be bad at recognizing unconventional intelligences because our perception is biased by our experience of the three-dimensional world of medium-sized objects moving at medium speeds. There are intelligences operating in other spaces (e.g., physiological space).
• Ethical questions arise from recognizing diverse forms of intelligence.

The Nature of Self and Identity

• A self is a collection of parts working together towards *system-level* goals (goals of the collection, not individual parts).
• Selves can be compared by the size and scope of their goals (“cognitive light cone”). A bacterium’s goals are small and local; a human’s can encompass larger spatial and temporal scales. Selves are nested, and humans may not be at the top of the hierarchy.
• We may be unable to fully understand the goals of a larger system of which we are a part (analogous to ants being unaware of the context of human actions). Mathematical formalisms might provide evidence for or against being part of a larger system.
• “A self is a *temporary* bundle of activities that work toward specific goals”: highlighting the agency nature that systems perform in different and potentially multiple cognitive landscapes.
• There’s a distinction between *learning* (changing your mind with the assumption of no external agency) and *being trained* (being changed by an external agency). It is an open, empirical question whether the external world has agency.
• Consciousness itself: is a hierarchy where “the more indeterminism (the space between what you could/cannot) is an indicator of the more level of agency that this form has”.

Scientific Inquiry and Open Questions

• Many phenomena should be treated as empirical questions rather than philosophical beliefs.
• Scientific frameworks both enable and constrain the kinds of questions we ask and experiments we perform. Our pre-conceptions, including being human-centric affects or decisions of intelligence (a dog being called intelligence as it is closer to our level).
• All science begins with an act of faith: the assumption that the world is understandable and that there are patterns to be discovered. It is important to be aware of this foundational belief.
• Current focus of Levin’s group: understanding different kinds of minds in various embodiments, with implications for regenerative medicine, birth defects, cancer, synthetic biology, and artificial intelligence.

引言与背景

• 迈克尔·莱文（Michael Levin）早期对事物如何构建的兴趣（由电视机引发）引导他探索了工程学/计算机科学和生物学。他对物理机器如何在屏幕上创建图像以及昆虫如何执行行为的方式感兴趣。
• 他着迷于思想是如何从物理学和化学中产生的，这推动了他对发育生物学的关注。
• 莱文的观点：标准的发育生物学课程可能会提供一个与他的观点不同的，可能更以基因为中心的观点。

基因、硬件和软件

• 基因组主要描述细胞的*硬件*（蛋白质序列），而不是整个生物体的形态（对称性、肢体数量等）。
• 生物系统既有硬件层（遗传学），也有关键的*软件*层（发育生理学和决策过程）。软件是动态的，对于导航复杂的发展至关重要。
• 从基因型（遗传信息）到表型（可观察特征）的映射是复杂的，不是简单的一对一关系，除非在特定酶产生等情况下。
• “蛙蝾螈”例子：混合青蛙和蝾螈细胞会产生一种生物，其腿部发育不能仅从基因组预测，这表明了细胞决策（软件）的重要性。
• 进化选择表型（最终结果，如解剖结构和行为），但它的基因组（原始信息）在世代之间传递，说明每个系统如何相互关联和相互依赖。

解释涌现特性

• 解释生物学特征需要同时考虑遗传因素和源自物理学、几何学和计算的“免费午餐”特性。
• 高尔顿板示例：钟形曲线形状是由系统的设置产生的，而不是木材、钉子或弹珠本身固有的。简单组件和组织产生的涌现特性。
• 晶体管示例：连接晶体管会创建具有真值表的逻辑门，这些逻辑门是配置的固有属性，而不是单独进化的。
• 生物系统可以利用“通用定律”（正如斯图尔特·考夫曼和斯图尔特·纽曼所讨论的），这些定律并没有在基因中明确编码。

细胞决策和生物电

• 细胞和细胞群具有自主性（偏好、行为）并做出决策。进化通过细胞之间的信号来塑造这些行为。
• 单个细胞具有小规模目标（pH值、代谢状态）。多细胞集体追求更大的目标（肢体形成）。
• 通过间隙连接将细胞连接起来形成的电网络，对于扩大目标和集体智能至关重要。
• 间隙连接允许集体的身份模糊化，因为钙等分子融合在一起，使多个细胞共享一个连接的生理信号。从细胞的角度来看，这是一个“虚假”记忆，从集体的角度来看，它是真实的，导致“心智融合”，其中记忆不再被“拥有”。
• 稳态比较：单个细胞测量、记忆并对其*局部*环境采取行动。电网络中连接的细胞测量、记忆并在*更大、非局部*的范围内采取行动，促进集体目标和反应。
• 应激的扩大：细胞通过共享信号传递应激（偏离期望状态），促进可塑性和协调行动以实现集体目标。个体应激成为一个共享问题。

癌症是生物电通信的崩溃

• 生物电信号在细胞群之间创造更大的“自我”感。
• 癌症可能源于细胞与更大网络的电连接断开。它们恢复到古老的、单细胞的目标：分裂并去资源丰富的地方（转移）。
• 细胞断开连接（变成癌细胞）不是因为“自私”增加，而是它收缩的结果，回到更自私的个体状态，单独行动。
• 癌基因通常会关闭间隙连接，这是生物电断开连接的第一步，将癌细胞与更大的集体控制隔离开来。
• 对细菌生物膜的研究表明，类似大脑的行为（使用离子通道）早在神经系统之前就已进化，表明生物电通信的古老根源。
• 形态空间：其中良好区域/不良区域，如障碍物/障碍物，类似于正常的三维空间导航，只是现在我们处于配置的世界中。

记忆和生物电回路

• 生物电网络确实存储了一种记忆，代表再生的“设定点”或目标形态。
• 涡虫示例：将扁虫切成碎片会导致每个碎片再生出一个完整、成比例的蠕虫。
• 这意味着这一定是某种形式的体内平衡在起作用；一种复杂的非神经形式的集体智能，以保持这种解剖模式的一致性。
• 生物电模式可以可视化（使用电压敏感染料）和*重写*（使用离子通道药物）以改变身体计划（例如，创建双头蠕虫）。不需要基因改变。
• 双头蠕虫持续产生双头蠕虫，证明了这是真正的记忆。
• 改变的身体计划通过*分裂*（分裂）是可遗传的，证明了非遗传继承。生物电回路充当了额外的遗传媒介。并非所有遗传特征都是基于DNA的。
• 涡虫中的生物电回路示例；如果系统的生理学首先“启动”头部数量的生物电默认回路。这个头数也可以通过使用输入（轻敲计算器上的按钮）短期改变信号来非遗传地编辑，在不“重写”程序的情况下创建一个双头蠕虫。

眼睛实验

• 早期青蛙胚胎显示出一种生物电预模式，可以预测面部器官（包括眼睛）的位置。
• 将离子通道RNA注入胚胎的其他区域（例如，尾部、肠道）会诱导异位（异位）眼睛的形成，表明生物电对于器官形成是*指导性的*，而不仅仅是为了维持。
• 这揭示了发育的模块化：研究人员不需要指定*如何*构建眼睛，只需要指定*在哪里*。细胞自己组织了复杂的过程。
• 细胞招募邻近的细胞（甚至那些没有直接受到注入RNA影响的细胞）参与眼睛的形成，展示了多层次的指导。
• 传统定义的“能力”区域（前神经外胚层）之外的细胞，实际上可以形成眼睛，突出了以基因为中心的观点的局限性。这不是层次结构的顶部，电压才是！
• Pac6：通常形成眼睛，位于前神经外胚层，这将定义形成眼睛的能力，然而，其他部分也可以。
• 连接到脊髓（而不是大脑）的异位眼睛仍然可以介导视觉，证明了神经系统的非凡可塑性及其解释新输入的能力。
• 这些例子是在莱文职业生涯的早期提出的，早于后来的研究。
• 中性突变；有害突变，一旦致命或严重，可以转变为不太严重或变得中性，拓宽了进化可以前进的范围。

多层次的控制和目标导向

• 生物系统表现出多层次的涌现和控制。
• 为有效操作选择*正确的*干预水平（例如，生物电模式与基因表达）至关重要。生物电操作通常比试图微观管理基因更有效和高效。
• 与其处理数以万计的单个部件/机制，不如在决策链的更上游进行干预，并与一个智能系统打交道，该系统导航复杂的决策，消除了处理细节的压力。
• 生物学家经常表现出“目的恐惧症”（害怕将目标或自主性归因于生物系统），但控制论提供了一个框架，用于将具有目标的机器理解为*连续体*，而不是二元（哑巴与智能）。这不再被认为是“神奇的”。
• 目的恐惧症来自需要将所有其他实体研究为时钟，因为早期不知道如何解释人类的内在思想。
• 自主性主张是*工程*主张，可以通过实验进行测试（例如，识别、读取和重写稳态系统中的设定点）。
• 恒温器示例：如何测试？看看它的水平是什么：是设定点吗？2 我们可以读取/解码设定点，重写吗？重写需要新的布线（像机械钟）。
• 恒温器继续：测试后；如果恒温器按预期工作，则建立信任/工程师依赖。无需微观管理，温度得到管理；好！.
• 基因通路可以被*训练*（类似于神经网络），表现出各种类型的学习（包括巴甫洛夫条件反射）。这挑战了纯粹确定性基因调控网络的想法。
• 联想学习的含义，在培养皿中存在太强大的药物，但我们不能应用于人类，同时给予两种药物，然后只给予中性药物；可能会也可能不会奏效。
• 分子安慰剂只有在配对足够多次（巴甫洛夫式）后才会激活，如果配对停止，则会出现原始反应（狗会因铃声而流口水）。

安慰剂、意图和身心联系

• 意图可以在多个层面上影响身体，包括细胞的生物电状态（例如，决定站起来会改变肌肉细胞电压）。身心联系是强大且可证明的。
• 这建议使用层次结构（指挥链）的各个层次，并在系统的相关部分说话以完成任务。
• 有证据表明，我们的身体中存在非语言的“自我”，我们可能是更大自我的一部分。
• 学习是一个主体内部的变化（人类/机器学习），而训练（神经网络）则涉及多个参与方，至少一个参与方制造压力（或缺乏压力），而另一个参与方发生变化。这种区别可能有助于探索进一步的发现。
• 思想在发展过程中逐渐出现。“纯粹的化学”和“拥有思想”之间没有明确的界限。这意味着思想以各种形式存在于生物组织的不同尺度上。
• 我们可能不擅长识别非常规智能，因为我们的感知受到我们对中等大小物体以中等速度移动的三维世界经验的偏见。在其他空间（例如，生理空间）中存在着智能。
• 从承认各种形式的智能中产生了伦理问题。

自我和身份的本质

• 自我是朝着*系统级*目标（集体目标，而不是单个部分）共同努力的各个部分的集合。
• 可以通过其目标的大小和范围（“认知光锥”）来比较自我。细菌的目标是小而局部的；人类的目标可以包括更大的空间和时间尺度。自我是嵌套的，人类可能不在层次结构的顶部。
• 我们可能无法完全理解我们所属的更大系统的目标（类似于蚂蚁不知道人类行为的背景）。数学形式主义可能会提供支持或反对成为更大系统一部分的证据。
• “自我是朝着特定目标努力的活动的*临时*捆绑”：强调系统在不同且可能多个认知领域中执行的自主性本质。
• *学习*（在没有外部自主性的假设下改变你的想法）和*被训练*（被外部自主性改变）之间存在区别。外部世界是否有自主性是一个开放的、经验性的问题。
• 意识本身：是一个层次结构，其中“不确定性越多（你可以/不能做什么之间的空间）是这种形式具有更多自主性水平的指标”。

科学探究和开放性问题

• 许多现象应被视为经验问题，而不是哲学信念。
• 科学框架既促进也限制了我们提出的问题和我们进行的实验类型。我们的先入之见，包括以人为中心，会影响或决定智力（狗被称为智能，因为它更接近我们的水平）。
• 所有科学都始于一种信仰行为：假设世界是可以理解的，并且存在有待发现的模式。重要的是要意识到这种基本信念。
• 莱文小组目前的研究重点：了解各种实施中的不同类型的思想，对再生医学、出生缺陷、癌症、合成生物学和人工智能产生影响。