共同演化：計算型交感系統之研究

Abstract

本篇論文旨在嘗試以複雜科學的觀念，研討各種人工智能的模型與學派， 更用適應性複雜系統的統一觀點，藉以析離出各類模型的共通觀念及基本 因子。進而形成新穎人工智能模型的構築要素，遂導致計算型交感系統概 念之開創。我們以一簡化的模擬網路為環境，嘗試植入一種初步的簡化型 交感系統－漫遊者系統，以觀察其網路控制效能，並為爾後的交感研究奠 下初基。於評析比較各類學派時，在演化問題方面，我們從經濟學中自由 市場流通的共通貨幣，並與熱力學中之消散性結構趨向複雜演化、及能謫 消長的廣泛自然現象，導出能量流遷的演化要素；另外我們亦由生態系□ ，萬物聚散生滅、共生合作於變異無常的環境中求存，思及共同演化的運 作機制。在效用問題方面，我們在需兼顧可控制系統以行應用、和可放任 系統以演變自發智能的兩相矛盾條件下，由生物學上各類物種及個體，競 相爭奪資源而產生的動態平衡現象加以思考，引出資源誘導的控制要素。 在以複雜度為分析利器下，我們由各類人工智能模型中歸納得四項共同性 質，複雜、演化、交感、適應；並萃取出三個共通構成要點，個體、互動 、競爭，並以此三點嘗試作成比較各模型複雜度的定性譜表。最後在結論 □，我們闡述了未來人工智能以此突破性發展的夢想遠景。 This thesis try to make a study of various paradigms of Artificial Intelligence from the viewpoint of the complexity science. Conceptually unified in the idea of Adaptive Complex Systems, these paradigms are analyzed in some common concepts and fundamental factors which we summarize. Moreover, the emergence of factors of neo-paradigms results in the concept of computational sympathizer systems. We attempt to develope an elementary simplified model, the stroller system, in the network environment for the sake of the premise of the study of sympathy mechanism. In the evolution aspect, we extract the evolution factor, Energy Transfer, from the ideas of the economical currency, the complexity of dissipative structures, and the thermal-dynamical transition between energy and entropy. In the mechanism aspect, we bring up the idea of coevolution associated with the ecology. In the utilization aspect, considering the contradiction between the utilization constraints and the spontaneous emergences, we derive the control factor, Resource Inducement, from the dynamical equilibrium in the competition of individuals for resources. In the overview of these paradigms, we summarize four common characteristics, which are Complexity, Evolution, Sympathy, and Adaptation. Futhermore, we abstract from the view of complexity three constitution factors; Individual, Interaction, and Competition - by means of which the complexity spectrums are described. Finally in the conclusions, we declare the prospects extended by these interesting ideas in the future.