The Universe Inside: Intelligence from Cells to Galaxies – Dr. Michael Levin, DSPod #213 – YouTube Bioelectricity Podcast Notes

PRINT ENGLISH BIOELECTRICITY GUIDE

PRINT CHINESE BIOELECTRICITY GUIDE


Minimum Unit of Intelligence, Evolution, and Behavior

  • Levin discusses embodied cognition and intelligence, focusing on minimal systems exhibiting cognitive behavior, beyond consciousness. He uses sorting algorithms as simple models.
  • These algorithms, though deterministic and transparent, demonstrate unexpected behaviors like “delayed gratification” (going temporarily backward to achieve a larger goal) and emergent goals.
  • There are surpises to be had on intelligence and competency, they can exist outside the known algorithm.
  • The algorithms cluster according to type (“algo-type”) during the sorting process, an emergent property not explicitly coded. This suggests “surprise minimization” as a driving force.
  • This work challenges assumptions about intelligence requiring late evolution, complex neural systems, or human-level goals. Even simple systems have “basal intelligence” (problem-solving to reach a goal by different means).
  • The focus is not *inventing* the intelligences. But *discovering* them.

Biobots (Anthrobots and Xenobots) and Emergent Goals

  • Levin’s lab works with Anthrobots (human tracheal cells) and Xenobots (frog skin cells), showing unexpected behaviors and capabilities.
  • Anthrobots, created by altering growth conditions (3D matrix then low-viscosity medium), spontaneously form motile structures that *heal neural wounds*.
  • Naming them “bots,” not just “organoids,” encourages exploring their potential for programmability and diverse applications beyond simply modeling organs.
  • It also reveals an issue that cells already know how to work together towards some competency before being programatically commanded, which is like and unlike AI safety issues.
  • These bots reveal “emergent competencies”—capabilities not directly programmed or selected for. The question becomes: Where do these novel goals originate?
  • Levin believes understanding and controlling these emergent goals is crucial, not just for medicine but also for understanding collective intelligence in general (e.g., swarms, AI).
  • This suggests new field of figuring out new goals of novel systems when and where evolution has no selectional history, this will become important.

Regenerative Medicine and Philosophical Implications

  • Levin’s research aims to “crack the morphogenetic code”—understanding how cell collectives make decisions about form—with the long-term goal of *in vivo* regeneration (regenerating tissues within the body).
  • There are also some interesting and emergent ideas: 1. cells could perhaps inherit other minds and also it opens up what it would actually mean for “you” or your identity. 2. Is it actually limited. 3. Do we actually only know what a living organism wants when they leave the constraints of what we normally observe it in, what constraints can we observe. 4. A science of mind at a distance.
  • This challenges traditional bioengineering approaches focused on micromanaging at the molecular level. The goal is to “persuade” cell collectives to achieve a desired form.
  • His view, intelligence is likely common, often appearing in “unfamiliar guises,” and the active inference framework of “free energy/surpise minimization” is perhaps key, cells strive for predictability.
  • This suggests there could be a type of chemistry of of the platonic/idea space, even in a non-spiritual way, if one knows where/how to observe.
  • One can make discoveries into what to do when you observe, instead of having an invention/creativity or something with no limits, we all have limited discoveries to make based on observation.

Broader Context of Diverse Intelligence

  • Traditional frameworks of “mind” (based on our usual forms and space-scales) “break” when one confronts the different capabilities across the scales/environments of lifeforms/organizations.
  • This isn’t just about the human scale, for example the potential for large scale behaviors, from the cellular all the way up to, perhaps, a planetary scale.
  • Intelligence and consciousness should *not* be conflated. Intelligence (problem-solving) is easier to study objectively; consciousness remains difficult.
  • Embodiment can occur in various “spaces” (physiological, transcriptional, etc.), not just 3D physical space. Perception-action loops are key, not necessarily 3D movement.
  • We need humility in recognizing intelligence in unconventional forms. Perturbational experiments, not just observation, are crucial for testing whether systems exhibit cognitive capacities.

智能、进化和行为的最小单位

  • 莱文讨论具身认知和智能,重点关注表现出认知行为的最小系统,超越了意识。他使用排序算法作为简单的模型。
  • 这些算法虽然是确定性的和透明的,但表现出意想不到的行为,如“延迟满足”(为了实现更大的目标而暂时倒退)和涌现的目标。
  • 在智能和能力方面存在惊喜,它们可以存在于已知算法之外。
  • 这些算法在排序过程中根据类型(“算法类型”)进行聚类,这是一个没有明确编码的涌现属性。这表明“惊喜最小化”是一种驱动力。
  • 这项工作挑战了关于智能需要晚期进化、复杂神经系统或人类水平目标的假设。即使是简单的系统也具有“基础智能”(通过不同方式达到目标的解决问题的能力)。
  • 重点不是*发明*智能,而是*发现*它们。

生物机器人(人源机器人和异种机器人)与涌现目标

  • 莱文的实验室使用人源机器人(人类气管细胞)和异种机器人(青蛙皮肤细胞),展示出意想不到的行为和能力。
  • 人源机器人,通过改变生长条件(3D 基质然后是低粘度介质)创建,自发地形成可移动的结构,这些结构可以*治愈神经伤口*。
  • 将它们命名为“机器人”,而不仅仅是“类器官”,鼓励探索它们的可编程性和多样化应用潜力,而不仅仅是简单地模拟器官。
  • 它还揭示了一个问题,即细胞在被编程控制之前已经知道如何协同工作以实现某些能力,这与人工智能安全问题既相似又不同。
  • 这些机器人揭示了“涌现能力”——不是直接编程或选择的能力。问题就变成了:这些新颖的目标从何而来?
  • 莱文认为,理解和控制这些涌现目标至关重要,不仅对医学,而且对理解一般的集体智能(例如,群体、人工智能)也是如此。
  • 这表明了一个新的领域,即在进化没有选择历史的情况下,弄清楚新系统的新目标,这将变得很重要。

再生医学和哲学意义

  • 莱文的研究旨在“破解形态发生密码”——理解细胞群体如何对形式做出决定——长期目标是在体内*再生*(在体内再生组织)。
  • 还有一些有趣和涌现的想法:1. 细胞也许可以继承其他思维,这也开启了“你”或你的身份的真正含义。 2. 它实际上是有限的吗? 3. 我们是否真的只有在我们通常观察它的约束之外,才能知道一个活的生物体想要什么,我们可以观察到什么约束? 4. 一门远距离的心智科学。
  • 这挑战了传统的生物工程方法,这些方法侧重于在分子水平上进行微观管理。目标是“说服”细胞群体达到所需的形态。
  • 他的观点是,智能可能很普遍,经常以“不熟悉的形式”出现,而“自由能/惊喜最小化”的主动推理框架可能是关键,细胞力求可预测性。
  • 这表明,即使以非精神的方式,也可能存在柏拉图/理念空间类型的化学,如果一个人知道在哪里/如何观察。
  • 人们可以发现观察时该做什么,而不是拥有发明/创造力或没有限制的东西,我们都有基于观察的有限发现。

多样化智能的更广阔背景

  • 当面对跨越生命形式/组织的不同尺度/环境的不同能力时,传统的“心智”框架(基于我们通常的形式和空间尺度)会“崩溃”。
  • 这不仅仅是关于人类的尺度,例如大规模行为的潜力,从细胞一直到,也许是行星尺度。
  • 不应将智能和意识混为一谈。智能(解决问题)更容易客观地研究;意识仍然很困难。
  • 具身化可以发生在各种“空间”(生理的、转录的等)中,而不仅仅是 3D 物理空间。感知-行动循环是关键,不一定是 3D 运动。
  • 我们需要谦逊地认识到非常规形式的智能。扰动实验,而不仅仅是观察,对于测试系统是否表现出认知能力至关重要。