符号主义与联结主义人工智能的融合路径分析

Analysis of the Integration Path of Symbolism and Connectionism of Artificial Intelligence

人工智能的演化形成了符号主义与联结主义人工智能两条代表性进路。符号主义人工智能的优点在于推理过程透明、可解释,但存在不完备、框架问题和知识接收瓶颈等问题。联结主义人工智能的泛化能力和可移植能力强,但缺陷是算法不可解释性、过拟合等问题。符号主义与联结主义人工智能有着显著的融合可能性,在新一代人工智能背景下,两条进路的融合又产生了新的路径,在基于神经网络的逻辑推理、符号表示的机器学习、认知图谱及其推理、机器学习与论辩挖掘等交叉领域表现出强劲的融合潜力。

The evolution of AI has formed two representative approaches of symbolism and connectionism of AI. The advantage of symbolic AI is that the reasoning process is transparent and interpretable, but there are problems such as incompleteness, framework problems, and knowledge reception bottlenecks. Connectionist AI has strong abilities of generalization and transportability, but its drawbacks are the inexplicability and overfitting of the algorithm. Symbolic and connectionist AI have significant integration possibilities. Under the background of the new generation of AI, the integration of the two approaches has produced new paths, which have shown strong integrating potential in neural networks with logical reasoning, machine learning based on the symbolic representation, cognitive graph and its reasoning, machine learning and argumentation mining, and other cross-fields.