Intelligence and Emergence

• Intelligence is deeply ingrained in the multi-scale architecture of living organisms. It exists from molecular networks up. AI can help understand biology, and biology can inspire new AI.
• Emergence is important, but biology’s key is closed-loop, goal-directed (cybernetic) agents at *all* scales, not just emergent outcomes. These agents actively reduce errors from set points, using energy to achieve goals.
• Higher levels of organization can do causal work that’s not reducible to lower levels (referencing Julio Tononi and Eric Hoel’s work on Information Theory). This can be mathematically determined, not just philosophically debated.
• Membrane voltage, a high-level aggregate, is more useful for regenerative medicine and other interventions than tracking individual ion positions. This highlights higher-level causal power.

Agency and Observer Relativity

• Agency is best described along a continuum (a “persuadability continuum”), not as a binary property. Sharp categories are imposed by observers on continuous phenomena.
• Levels of system, is based of what is being observed. It can shift with what the observer.
• The claim that a system occupies a particular place on this agency continuum is observer-relative. It’s not an objective fact, but depends on the observer’s model and how well that model lets them interact with the system.
• A crucial point is that the system *itself* can be a valid observer, creating internal models and controlling itself. This “self-observation” is a key aspect of agency.
• Making the claims about systems are done so in respect of a hypothesis on an individual system by a researcher/obvserver.

Physicalism, Reductionism, and Matter

• Challenging reductionist physicalism doesn’t negate physicalism itself. It argues that matter is capable of far more than often assumed, including higher-level organization with non-reducible causal power.
• Saying an entity isn’t made of a kind of matter isn’t helpful in describing intelligence, given that all things have the same underlying “stuff”.
• “Pseudo-problems” arise from making objective claims in this area. Specifying the vantage point (observer) clarifies many issues.
• Any computation or intelligence isn’t necessarily reliant upon some sort of external observers but instead it itself can observe, as this defines this form of being (a legitimate entity in-and-of-itself).

Phase Transitions and Continuity

• While sharp, non-linear changes in behavior (akin to phase transitions) exist in AI and other fields, biological cognition appears more continuous. Sharpness in phase transitions often increases with system size.
• It’s difficult to pinpoint a specific “phase transition” where cognition *suddenly* emerges in development or evolution. The substrate is continuous; it’s a transformation of the *same* material.
• Attributing cognitive capabilities to single cells is controversial, but critics need to provide a concrete explanation of *how* and *when* full cognition emerges during development or evolution. Simply stating this point in time isn’t concrete.
• This doesn’t equate the lack of understanding transitions in understanding and quantifying them for future predictions (on future research, technologies, etc etc).

Defining and Measuring Intelligence

• Even gene regulatory networks (seemingly mechanical) exhibit various forms of learning (e.g., associative conditioning), showcasing non-zero intelligence at very low levels.
• The absolute minimum of intelligence likely involves: (1) some level of goal-directedness (least action principles, even in particles) and (2) some indeterminacy (local conditions not fully determining behavior).
• This doesn’t need to equate to 0 levels. Even at atomic scales and even in a “cognitive vacuum” these principles can exist as its “basement”, so long there are actions, reactions and movements of its components.
• Measuring intelligence is taking an IQ test ourselves. We must define a problem space, identify the system’s policy, and test the policy’s “cleverness.” We might not always recognize intelligence, especially outside familiar domains.
• Intelligence doesn’t rely on conventional intelligence such as with brains.
• Researchers must actively observe, analyze and quantify intelligence in non-human systems by defining their: behavior and environmental space (that it is doing it’s intelligent “stuff”).
• Intelligence is related to its capabilities in such a “problem space” and may depend, for example, on navigating this space and/or physiology.
• A continuum of competencies exists, from simple systems following energy gradients (like magnets) to complex systems overcoming obstacles to reach goals (like Romeo and Juliet). The is quantifiable in this way.
• Research shows that there is intelligence in simpler systems.
• The challenge is perturbation. It changes goals of individual parts that normally seem fixed on fixed behaviours.
• A system’s intelligence is revealed by challenging its “normal” behavior with perturbations. Often, systems are more capable than initially assumed. This makes assumptions hardwired (fixed on set behaviours, that would have difficulty adjusting/doing so within the limits/timescale for measuring and/or perturbation.

Goal-Directedness, Anthropomorphism, and Interaction

• A system’s “goal” is a useful lens or perspective for prediction and control, not necessarily an objective, discoverable fact. Positing a goal and competencies helps us interact effectively.
• Such can also involve more complicated relationships than what would be considered standard such as “friendships” if they allow for an optimized control and behaviour, for both sides involved in a system.
• “Anthropomorphism” isn’t a useful concept. We should make *specific*, testable claims about a system’s capabilities (human, robot, cell, etc.) and empirically determine their validity. We often *underestimate* intelligence, especially in biology.
• The assumption on an entities “naturality” also influences people on intelligence due to biases in a pre-garden assumption that doesn’t account that nature (life) doesn’t set you up, as Evolution simply aims for replicating with variations.

On Natural, Biomass and Goals

• Nature’s is set to survival by the “means of replication”, which is done from stochastic-gradient descent.
• Different solutions can apply/be found on how goals and intelligences can manifest that differs from humans.
• Humans desire/instinctually don’t enjoy nature because their current goals, even before major technological advances, often had goals (in many examples given, even primitive ones such as an umbrella, showed examples and reasoning on why and how to differ goals).
• Humans don’t have innate, fixed and/or objective/optimal goals (even pre-Garden as there isn’t necessarily something objective to begin with (on what would constitute these parameters).
• Humans also can influence these biases as with tools or modifications of any level.

Transhumanism and Categories

• Biology is highly interoperable. We can create chimeras/hybrids between biological tissues and machines, blurring boundaries between “human” and “machine.” There is it continuous with many variances in percentage. This challenges binary categories.
• A goal of intelligence research can mean to “better what is defined as a natural body” (of people for example).
• There exist human, biological extensions to even radical extremes that makes distinctions for transhumanism that blurs these parameters of distinction.
• Categorizing organisms/systems as natural kinds (with sharp, inherent boundaries) is limiting. Biology and technology push these boundaries, demanding a more nuanced perspective.
• “Natural” is often undesirable. Evolution optimizes for biomass, not human happiness or fulfillment. We can (and should) strive to do better than “natural” through science and technology.
• What is viewed as unnatural could well be natural too due to emergent, external factors that affect changes in human’s systems in question and the research or development being studied or discussed.

智能与涌现

• 智能深深植根于生命体的多尺度结构中。它从分子网络向上延伸。人工智能可以帮助理解生物学，生物学也可以激发新的人工智能。
• 涌现很重要，但生物学的关键在于在*所有*尺度上都有闭环、目标导向（控制论）的智能体，而不仅仅是涌现的结果。 这些智能体积极地减少与设定点的误差，利用能量来实现目标。
• 更高层次的组织可以完成无法简化为较低层次的因果工作（参考朱利奥·托诺尼和埃里克·霍尔的信息论研究）。这可以通过数学方法确定，而不仅仅是哲学上的争论。
• 膜电压，一个高层次的集合体，对于再生医学和其他干预措施比跟踪单个离子位置更有用。这突出了更高层次的因果力量。

自主性与观察者相对性

• 自主性最好沿着一个连续体（“可说服性连续体”）来描述，而不是作为一个二元属性。 尖锐的类别是观察者强加给连续现象的。
• 系统的层次，是基于正在观察的内容。 它可以随着观察者的不同而改变。
• 一个系统在这个自主性连续体上占据特定位置的说法是相对于观察者的。 这不是一个客观事实，而是取决于观察者的模型以及该模型允许他们与系统互动的程度。
• 一个关键点是，系统*自身*可以成为一个有效的观察者，创建内部模型并控制自身。这种“自我观察”是自主性的一个关键方面。
• 关于系统的声明是相对于研究人员/观察者对单个系统的假设而做出的。

物理主义、还原论和物质

• 挑战还原论的物理主义并不否定物理主义本身。 它认为物质的能力远不止通常假设的，包括具有不可还原的因果力量的更高层次的组织。
• 鉴于所有事物都具有相同的基础“物质”，说一个实体不是由某种物质构成的，无助于描述智能。
• “伪问题”源于在这个领域做出客观声明。指定观察点（观察者）可以澄清许多问题。
• 任何计算或智能都不一定依赖于某种外部观察者，而是它自身可以观察，因为这定义了这种存在形式（其本身就是一个合法的实体）。

相变与连续性

• 虽然行为中存在类似于相变的尖锐、非线性变化，在人工智能和其他领域中也存在，但生物认知似乎更具连续性。相变中的锐度通常随着系统尺寸的增大而增加。
• 很难确定一个特定的“相变”，在这个“相变”中，认知在发育或进化过程中*突然*出现。基质是连续的； 这是*相同*材料的转化。
• 将认知能力归因于单个细胞是有争议的，但批评者需要提供一个具体的解释，说明完全认知在发育或进化过程中是*如何*和*何时*出现的。仅仅陈述这个时间点是不具体的。
• 这并不等同于缺乏对过渡的理解，在理解和量化它们以用于未来的预测（关于未来的研究、技术等）。

定义和测量智能

• 即使是基因调控网络（看似机械的）也表现出各种形式的学习（例如，联想条件反射），表明在非常低的水平上存在非零智能。
• 智能的绝对最小值可能包括：（1）某种程度的目标导向性（最小作用量原理，即使在粒子中也是如此）和（2）某种不确定性（局部条件不能完全决定行为）。
• 这不需要等于0级。即使在原子尺度上，即使在“认知真空”中，只要其组成部分存在动作、反应和运动，这些原则也可以作为其“基础”存在。
• 衡量智力就是我们自己在参加智商测试。 我们必须定义一个问题空间，确定系统的策略，并测试策略的“聪明程度”。我们可能并不总是能识别出智能，尤其是在熟悉的领域之外。
• 智能不依赖于传统的大脑等智能。
• 研究人员必须通过定义它们的行为和环境空间（它正在做它的智能“事情”）来积极观察、分析和量化非人类系统中的智能。
• 智能与它在这样一个“问题空间”中的能力有关，并且可能取决于例如，导航这个空间和/或生理机能。
• 存在一个连续的能力谱，从遵循能量梯度（如磁铁）的简单系统到克服障碍以实现目标（如罗密欧与朱丽叶）的复杂系统。这是可以用这种方式量化的。
• 研究表明，更简单的系统也存在智能。
• 挑战在于扰动。它改变了通常似乎固定在固定行为上的单个部分的目标。
• 一个系统的智能是通过用扰动挑战其“正常”行为来揭示的。 通常，系统比最初假设的更有能力。这使得假设是硬连接的（固定在设定的行为上，这些行为在用于测量和/或扰动的限制/时间尺度内难以调整/做到这一点）。

目标导向性、拟人化和互动

• 一个系统的“目标”是一个有用的视角或观点，用于预测和控制，而不一定是一个客观的、可发现的事实。 假定一个目标和能力可以帮助我们有效地互动。
• 这种也可以涉及比被认为是标准的更复杂的关系，例如“友谊”，如果它们允许对所涉及的系统中的双方进行优化的控制和行为。
• “拟人化”不是一个有用的概念。我们应该对系统（人类、机器人、细胞等）的能力做出*具体的*、可测试的声明，并通过经验确定其有效性。 我们经常*低估*智能，尤其是在生物学中。
• 对一个实体“自然性”的假设也会因为前花园假设中的偏见而影响人们对智能的看法，该假设没有考虑到自然（生命）不会为你做准备，因为进化只是为了用变异来复制。

关于自然，生物量和目标

• 自然的目标是通过“复制的手段”来实现生存，这是通过随机梯度下降来实现的。
• 对于目标和智能如何以不同于人类的方式表现，可以应用/找到不同的解决方案。
• 人类渴望/本能地不喜欢自然，因为他们当前的目标，即使在重大的技术进步之前，也经常有目标（在给出的许多例子中，甚至是原始的目标，例如雨伞，显示了关于为什么以及如何区分目标的例子和推理）。人类没有先天、固定和/或客观/最佳目标（即使是“伊甸园”之前的，因为一开始就不一定有客观的东西（关于什么构成这些参数））。
• 人类也可以通过工具或任何级别的修改来影响这些偏见。

超人类主义和范畴

• 生物学具有高度的互操作性。我们可以在生物组织和机器之间创造嵌合体/杂交体，模糊“人类”和“机器”之间的界限。它是连续的，有许多百分比差异。这挑战了二元范畴。
• 智能研究的目标可以意味着“改善被定义为自然身体的东西”（例如人）。
• 存在人类、生物扩展，甚至是极端的，这使得超人类主义的区分模糊了这些区分的参数。
• 将生物体/系统归类为自然种类（具有清晰、固有的边界）是具有限制性的。生物学和技术推动了这些界限，需要更细致的视角。
• “自然”通常是不受欢迎的。进化优化的是生物量，而不是人类的幸福或成就感。 我们可以（也应该）努力通过科学和技术做得比“自然”更好。
• 由于影响所讨论的人类系统变化的涌现的外部因素，以及正在研究或讨论的研究或开发，被视为不自然的也很可能是自然的。