(5)
|
| 人工智慧的本質與倫理邊界：身體現象學與群體智慧的觀點 |
| The Nature of Artificial Intelligence and Its Ethical Boundaries: Perspectives from Phenomenology of the Body and Collective Intelligence |
 |
| 作者 |
陳慶瀚 |
| Author |
Ching-Han Chen |
| 關鍵詞 |
人工智慧、
身體現象學、
非對稱交織、
群體智慧、
非對稱共生倫理
|
| Keywords |
Artificial Intelligence,
Phenomenology of the Body,
Asymmetric Interweaving (chiasme),
Collective Intelligence; Asymmetric Symbiotic Ethics
|
| 摘要 |
自大型語言模型（Large Language Models, LLMs）進入公共應用以來，人工智慧在若干任務與使用情境中，被部分研究描述為「突現能力」（emergent abilities）的非線性能力變化；該現象的性質與成因目前仍有爭議，卻已引發關於工作取代、系統失控與人類主體性消解的廣泛倫理焦慮。本文主張，這些焦慮並非僅源於技術能力的快速增長，而是根植於我們仍以「智慧作為內在、可控制屬性」的舊有理解框架，來面對一種已呈現關係性特徵的智慧形態。本文透過梅洛－龐蒂的身體現象學，重新檢視智慧的存在方式，主張智慧不宜被理解為封裝於單一系統中的內在屬性，而是在主體、行動與環境之間持續生成的關係性過程。本文以螞蟻群體智慧作為非集中式智慧生成的結構性類比，說明環境中介、局部回饋與分散協作如何形成群體層次的秩序，而非主張其可直接作為人工智慧的倫理規範模型。本文進一步指出，人類與大型語言模型之間的交織具有結構上的不對稱性：模型確實進入智慧生成的關係網路並發揮結構性參與作用，卻不佔據肉身可逆性的另一側，它改變環境而不被環境所觸及，其輸出造成傷害而自身不承受傷害。此一不對稱並非人類中心立場的殘留，而是同一套現象學分析的內在後果。本文主張，大型語言模型構成第一種在語意層次上閉合的間接溝通結構：模型輸出沉積於人類共享的符號環境，引導其後的人類表述，而該表述又回流為模型的訓練分佈；此一遞迴迴路在既有的自然範式中無所對應，亦為本文必須針對人工智慧提出倫理框架的理由。在此基礎上，本文提出「非對稱共生倫理」作為回應的框架：所謂共生，指人類活動與人工智慧系統在實踐上形成持續的相互依存與回饋；所謂非對稱，指兩者在責任能力上呈現根本差異。人工智慧的倫理邊界因此不應以全面控制或消除不確定性為唯一目標，而應同時著眼於智慧生成關係本身是否保有回饋、調適與價值介入的可能性；惟在回饋不可及且錯誤不可逆的情境中，該框架自身便要求退回嚴格的事前控制與人類否決權。最後，人類主體性並未因此消解，而是被重新定位為智慧生成網路中的關係調節者，此一位置來自其可受傷害性所衍生的責任能力，而非存在論上的特權。 |
| Abstract |
Since large language models (LLMs) entered widespread public use, artificial intelligence has exhibited non-linear shifts in capability that some studies describe as “emergent abilities” — a characterization whose nature and causes remain contested, yet which has already given rise to broad ethical anxieties concerning job displacement, loss of system control, and the erosion of human subjectivity. This paper argues that these concerns do not arise solely from the rapid growth of technical capabilities, but are rooted in a persistent conceptual framework that treats intelligence as an internal and controllable property, applied to a form of intelligence that has already taken on relational characteristics. Drawing on Maurice Merleau-Ponty’s phenomenology of the body, the paper re-examines the mode of being of intelligence, arguing that it should be understood not as an internal capacity encapsulated within a single system, but as a relational process continuously generated through the relations among subjects, actions, and environments. Ant colony intelligence is introduced as a structural analogy for the generation of decentralized intelligence, explaining how environmental mediation, local feedback, and distributed coordination give rise to collective-level order — not as a direct normative model for AI ethics. The paper further argues that the interweaving (chiasme) between humans and LLMs is structurally asymmetric: the model does enter the relational network of intelligence generation and exercises structural participation, yet it does not occupy the other side of carnal reversibility — it alters the environment without being touched by it, and its outputs cause harm that it does not itself undergo. This asymmetry is not a residue of anthropocentrism but an internal consequence of the same phenomenological analysis. The paper also argues that LLMs constitute the first form of stigmergy closed at the semantic level: model outputs are deposited in the shared symbolic environment, guide subsequent human expression, and that expression in turn flows back into the model's training distribution — a recursive loop with no counterpart in natural paradigms, and the reason an ethical framework must be formulated for AI in particular. On this basis, the paper proposes asymmetric symbiotic ethics as a responsive framework. Ethical boundaries should attend to whether the relations through which intelligence is generated retain possibilities for feedback, adaptation, and value intervention; yet where feedback is unavailable and errors are irreversible, the framework itself requires a return to strict ex ante control and human veto. Human subjectivity is thereby reconfigured as that of a relational mediator — this standing derives from the responsibility that vulnerability engenders, rather than an ontological privilege. |
|
|
|
