Introduction and Levels of Explanation

• A major myth in biology is that the best explanations are always at the molecular level. Different levels of explanation (biology, chemistry, physics) offer different insights, and higher levels have autonomy.
• Emergence is a measure of surprise for the observer – how much a system does that wasn’t anticipated from the properties of its parts. It is relative, not absolute.
• Cognitive functions (learning, memory, conditioning) can be found even in very simple systems like gene regulatory networks. It is important to go past the levels.

Unifying Themes: Embodied Mind and Intelligence

• Levin’s work across various fields (cancer, development, regeneration, AI) is unified by an effort to understand embodied mind and intelligence in diverse, unconventional forms.
• Goals of organisms are typically attributed to evolutionary history. Synthetic constructs (xenobots, anthropods) allow studying the origins of goals in systems *without* such history.
• The “cognitive light cone” represents the size of the largest goal a system can pursue, in space and time. This concept helps understand cancer as a *shrinking* of this light cone, where cells revert to individualistic goals.
• Cancer cells aren’t necessarily more *selfish*; they have *smaller selves*. This leads to research on reconnecting cancer cells to bioelectrical networks, normalizing them *without* killing them.

Defining Intelligence and the Platonic View

• Intelligence (following William James) is the ability to reach the same goal by different means. This definition emphasizes problem-solving in a specific problem space, regardless of the physical substrate (brain, synthetic system, etc.).
• Levin leans toward a Platonic view of intelligence, akin to the mathematical Platonism, there exists, in fact, a space where mathematical properties of computation that think and compute “live” in such that we merely discover. He suggests that not only do rules of maths “live” there, but other cognitive states live as well, not limited by material states, we don’t just invent minds; physical systems can “harness” pre-existing intelligence.
• Estimates of intelligence are *not* objective properties of a system. They are guesses about its problem-solving capabilities, reflecting *our* knowledge (or lack thereof).

Self, Memory, and Dynamic Interpretation

• We are not static entities, but rather collections of interacting perspectives. “Selflets” are thin slices of experience.
• The continuity of self is perceived by *other observers* based on the consistency of behaviors and properties. We are interested on how the system behaves to ourselves, how to better get “messages” (engrams).
• Our access to the past is through memory engrams – *interpreted* traces, not direct access. These traces must be dynamically reinterpreted in new contexts.
• Our current actions are like messages to our future selves, constraining or enabling their possibilities. This creates a symmetry between our future self and *others’* future selves, with ethical implications.
• Under resource constraints, agents *must* coarse-grain; they cannot track everything. They create compressed representations (like memory engrams), focusing on salience, not fidelity.

Compression, Perspectives, and Synchronicity

• Highly compressed data can appear random because correlations are removed. This has implications for interpreting potential signals from advanced civilizations.
• What we experience depends on an interpretive agent. What we compress for memories, have to be de-compressed by us in future times, thus leading us to needing a reinterpretation of the rule to the given observation.
• Perspectives are fundamental: commitments to what to measure, what to pay attention to, and how to weave that into a model. Every perspective necessarily shuts out more than it lets in.
• If we are part of a larger cognitive system, recognizing that might look like *synchronicity* – meaningful events without apparent causal connection at our level.

Bioelectricity as Cognitive Glue

• Bioelectricity is *not* magic, but a crucial mechanism for enabling collective intelligence, by being used to create a policy for cooperation.
• Bioelectricity allows the cognitive light cone to scale up. Cells connected in electrical networks form larger emergent individuals with higher-level goals and capabilities.
• Cells create their agency with signals that enable other parts of their larger organism to move.

Examples from the Levin Lab

• Early work showed bioelectricity’s role in left-right asymmetry in chicken embryos, manipulating this with ion channel constructs.
• “Electric face”: Bioelectric patterns in nascent ectoderm *prefigure* the formation of facial features. Birth defects disrupting these patterns can be corrected bioelectrically.
• The first demonstration of gaining the regeneration functions: Bioelectric manipulation can *induce* tail regeneration in tadpoles, demonstrating the control over large-scale anatomical outcomes. A 24-hour bioelectrical stimulation to frogs triggers *a year and a half* of leg growth, showing a high-level command (“build a leg”) without micromanaging the process.
• Bioelectric signaling is linked to cancer. Cells can be induced to become metastatic with *inappropriate* bioelectric cues. Conversely, bioelectrical connections can *normalize* cells expressing strong human oncogenes.
• Planaria: Two-headed worms show a *permanent, non-genetic* change in target morphology. This demonstrates physiological memory. Planaria’s highly chaotic genome and remarkable regenerative abilities suggest a prioritization of *algorithmic competency* over genetic fidelity.
• Anthropods (human-derived organoids) demonstrate that *tracheal cells* can exhibit novel behaviors, including neural repair, *without* any genetic changes. This showcases the plasticity and emergent capabilities of cell collectives.

Implications and Future Directions

• Evolution creates *problem-solving agents*, not just solutions to specific problems. These agents have tools (cytoskeleton, gene regulatory networks, bioelectricity) to handle novel circumstances.
• Work is shifting towards clinically relevant models: human cancer cells and spheroids for cancer research, and mice for regeneration.
• Understanding how information flows across levels (embryos communicating with each other, forming “hyper-embryos”) is a key focus.

Advice for Newcomers to Biology

• Paths are in science hard to predict, be vary of following peoples research agendas instead of making your own based on your goals and aspirations, follow your own “guts”.
• Develop your own intuition about which paths to take in science, and test those intuitions through experiment.
• Prioritize specific, technical critiques to improve your craft (how to make an experiment better), but be wary of general advice about career direction.
• Unifying seemingly disparate phenomena (like the different forces responsible for apple falls and planetary motions) is a major scientific achievement. Be open to this possibility.

导言与解释层面

• 生物学中的一个主要迷思是，最好的解释总是在分子水平上。不同的解释层面（生物学、化学、物理学）提供了不同的见解，而更高的层面具有自主性。
• 涌现是观察者惊奇程度的度量——一个系统所做的超出其组成部分属性预期的程度。它是相对的，而不是绝对的。
• 认知功能（学习、记忆、条件反射）甚至可以在非常简单的系统中找到，比如基因调控网络。重要的是要超越各个层面。

统一主题：具身心智与智能

• 莱文在各个领域（癌症、发育、再生、人工智能）的工作都围绕着一个统一的目标：理解多样化、非常规形式的具身心智和智能。
• 生物体的目标通常归因于进化历史。合成构建体（异种机器人、人源类器官）允许研究 *没有* 这种历史的系统中目标的起源。
• “认知光锥”代表一个系统可以追求的最大目标的规模，在空间和时间上。这个概念有助于将癌症理解为这个光锥的 *缩小*，其中细胞恢复到个体主义目标。
• 癌细胞不一定更 *自私*；它们有 *更小的自我*。这导致了对将癌细胞重新连接到生物电网络的研究，使它们 *不* 被杀死而正常化。

定义智能与柏拉图式观点

• 智能（遵循威廉·詹姆斯的定义）是通过不同手段达到同一目标的能力。这个定义强调在特定问题空间中的问题解决，无论物理基质如何（大脑、合成系统等）。
• 莱文倾向于一种柏拉图式的智能观，类似于数学的柏拉图主义，实际上存在一个空间，其中计算的数学属性“存在”于其中，我们仅仅是发现而已。 他认为，不仅数学规则“存在”于那里，而且其他认知状态也存在，不受物质状态的限制，我们不仅仅是发明思想；物理系统可以“驾驭”预先存在的智能。
• 对智能的估计 *不是* 系统的客观属性。它们是关于其问题解决能力的猜测，反映了 *我们* 的知识（或缺乏知识）。

自我、记忆与动态解释

• 我们不是静态的实体，而是相互作用的视角的集合。“自我碎片”是经验的薄片。
• 自我的连续性是由 *其他观察者* 根据行为和属性的一致性来感知的。我们感兴趣的是系统如何对我们自己表现，如何更好地获取“信息”（记忆印迹）。
• 我们对过去的访问是通过记忆印迹—— *经过解释的* 痕迹，而不是直接访问。这些痕迹必须在新的情境中被动态地重新解释。
• 我们当前的行为就像给我们未来自我的信息，限制或开启他们的可能性。这在我们的未来自我和 *他人* 的未来自我之间创造了一种对称性，具有伦理含义。
• 在资源限制下，主体 *必须* 粗粒化；它们不能跟踪一切。它们创建压缩的表示（如记忆印迹），专注于显著性，而不是保真度。

压缩、视角与同步性

• 高度压缩的数据可能看起来是随机的，因为相关性被移除。这对解释来自先进文明的潜在信号有影响。
• 我们的经验取决于一个解释主体。我们为记忆压缩的内容，必须由我们在未来时间解压缩，从而导致我们需要对给定观察重新解释规则。
• 视角是基本的：承诺要测量什么，要注意什么，以及如何将其编织成一个模型。每个视角必然会关闭比它允许进入的更多。
• 如果我们是一个更大的认知系统的一部分，认识到这一点可能看起来像 *同步性*——在我们的层面上没有明显因果关系的意义深远的事件。

生物电作为认知粘合剂

• 生物电 *不是* 魔法，而是一种实现集体智能的关键机制，通过被用来创建合作的策略。
• 生物电允许认知光锥扩大规模。在电网络中连接的细胞形成更大的具有更高层次目标和能力的涌现个体。
• 细胞通过使它们的较大生物体的其他部分能够移动的信号来创建它们的自主性。

来自莱文实验室的例子

• 早期工作表明生物电在鸡胚胎左右不对称中的作用，用离子通道构建体操纵这一点。
• “电脸”：新生外胚层中的生物电模式 *预示* 面部特征的形成。破坏这些模式的先天缺陷可以通过生物电来纠正。
• 首次证明获得再生功能：生物电操纵可以 *诱导* 蝌蚪的尾巴再生，证明了对大规模解剖结果的控制。 对青蛙进行 24 小时的生物电刺激会触发 *一年半* 的腿部生长，显示了高级命令（“构建一条腿”），而无需微观管理该过程。
• 生物电信号与癌症有关。细胞可以通过 *不适当的* 生物电信号诱导转移。相反，生物电连接可以 *正常化* 表达强人类癌基因的细胞。
• 涡虫：双头蠕虫显示出目标形态的 *永久性、非遗传性* 改变。这证明了生理记忆。 涡虫高度混乱的基因组和显著的再生能力表明，优先考虑 *算法能力* 而不是遗传保真度。
• 人源类器官（人类衍生的类器官）表明 *气管细胞* 可以表现出新的行为，包括神经修复， *无需* 任何基因变化。这展示了细胞群体的可塑性和涌现能力。

影响与未来方向

• 进化创造 *解决问题的代理*，而不仅仅是特定问题的解决方案。这些代理拥有工具（细胞骨架、基因调控网络、生物电）来处理新的情况。
• 工作正在转向临床相关模型：用于癌症研究的人类癌细胞和球体，以及用于再生的小鼠。
• 了解信息如何在不同层次之间流动（胚胎相互交流，形成“超胚胎”）是一个关键焦点。

给生物学新手的建议

• 在科学中的道路很难预测，要小心追随别人的研究议程，而不是根据你自己的目标和愿望制定自己的研究议程，追随你自己的“直觉”。
• 发展你自己在科学中采取哪些路径的直觉，并通过实验来测试这些直觉。
• 优先考虑具体的技术批评来改进你的技艺（如何使实验更好），但要警惕关于职业方向的一般性建议。
• 统一看似不同的现象（如导致苹果坠落和行星运动的不同力）是一项重大的科学成就。对此可能性持开放态度。