计算心智理论是一场关于人类心智本质的重要哲学与科学论争[1]。该理论的核心问题在于：心智是否可被理解为计算系统[1]。
这场论争的理论基础源于1936年阿兰·图灵提出的抽象计算模型——图灵机，它成为了现代计算理论的基石[1]。经典计算心智理论（CCTM）建立在这一基础之上，主张心智的作用方式类似图灵机，核心心理过程本质上就是图灵式计算[1]。哲学家Fodor进一步发展了这一观点，提出表征心智理论（RTM），认为心理活动涉及对"思维语言"（Mentalese）中符号的图灵式计算[1]。
然而，计算主义理论遭遇了来自多个方向的挑战。连接主义作为主要的对立范式应运而生，它摒弃符号计算的传统，转而采用神经网络模型，并在1980年代重获学术关注[1]。2010年代深度神经网络的突破性成就——包括AlexNet、AlphaGo和ChatGPT的成功——进一步推动了对计算模型的重新评估[1]。
这场理论之争触及多个核心问题，包括计算与表征的关系、宽内容与窄内容的区分，以及计算是否必然需要表征内容等[1]。其中，Gallistel和King提出的生产性论证尤为关键——他们认为认知计算的生产性特征需要可寻址的读写记忆系统，这一论点被视为反驳连接主义的重要依据[1]。
This 2015 scholarly review examines the computational theory of mind (CTM), a foundational debate in philosophy and cognitive science regarding whether the mind functions as a computational system [1]. The article traces the intellectual development from Alan Turing's abstract computational model introduced in 1936—which established the theoretical basis for modern computation—through various computational paradigms including artificial intelligence, classical computation theory, and neural networks [1].
The classical computational theory of mind (CCTM) posits that the mind operates similarly to a Turing machine, with core psychological processes functioning as Turing-style computations [1]. Jerry Fodor's representational theory of mind (RTM) extends this framework by proposing that mental activity involves Turing-like computations on symbols in a mental language known as Mentalese [1]. However, connectionism emerged as a major competing paradigm, employing neural network models rather than symbolic computation and experienced renewed attention during the 1980s [1]. Recent developments in deep neural networks throughout the 2010s—including breakthroughs such as AlexNet, AlphaGo, and ChatGPT—have prompted renewed examination of computational models [1].
The survey addresses central controversies in the field, including the relationship between computation and representation, distinctions between broad and narrow content, whether computation necessarily requires representational content, and mechanical and temporal concerns [1]. Notably, Gallistel and King advanced the productivity argument, contending that cognitive computation's generative capacity requires addressable read-write memory—a position they identify as critical to refuting connectionism [1].